6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-29 10:32:47 +00:00
Code Issues Projects Releases Wiki Activity
3d47855394
llvm-6502/test/CodeGen/ARM/build-attributes.ll

801 lines
38 KiB
LLVM
Raw Normal View History

[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
; This tests that MC/asm header conversion is smooth and that the
; build attributes are correct
Move the xscale build attribute test to the proper place and remove the old one. The encoding of build attributes is already tested in CodeGen/ARM/build-attributes-encoding.s git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199393 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-16 15:11:54 +00:00
; RUN: llc < %s -mtriple=thumbv5-linux-gnueabi -mcpu=xscale | FileCheck %s --check-prefix=XSCALE
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi | FileCheck %s --check-prefix=V6
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi -enable-unsafe-fp-math | FileCheck %s --check-prefix=V6-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=thumbv6m-linux-gnueabi | FileCheck %s --check-prefix=V6M
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv6m-linux-gnueabi -enable-unsafe-fp-math | FileCheck %s --check-prefix=V6M-FAST
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi -mcpu=arm1156t2f-s | FileCheck %s --check-prefix=ARM1156T2F-S
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi -mcpu=arm1156t2f-s -enable-unsafe-fp-math | FileCheck %s --check-prefix=ARM1156T2F-S-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv6-linux-gnueabi -mcpu=arm1156t2f-s -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi | FileCheck %s --check-prefix=V7M
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -enable-unsafe-fp-math | FileCheck %s --check-prefix=V7M-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi | FileCheck %s --check-prefix=V7
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -enable-unsafe-fp-math | FileCheck %s --check-prefix=V7-FAST
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi | FileCheck %s --check-prefix=V8
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -enable-unsafe-fp-math | FileCheck %s --check-prefix=V8-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; RUN: llc < %s -mtriple=thumbv8-linux-gnueabi | FileCheck %s --check-prefix=Vt8
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv8-linux-gnueabi -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Add subtarget feature support for Cortex-A53 Some previous implicit defaults have changed, for example FP and NEON are now on by default. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192590 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:16:57 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mattr=-neon,-crypto | FileCheck %s --check-prefix=V8-FPARMv8
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mattr=-fp-armv8,-crypto | FileCheck %s --check-prefix=V8-NEON
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mattr=-crypto | FileCheck %s --check-prefix=V8-FPARMv8-NEON
; RUN: llc < %s -mtriple=armv8-linux-gnueabi | FileCheck %s --check-prefix=V8-FPARMv8-NEON-CRYPTO
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 | FileCheck %s --check-prefix=CORTEX-A5-DEFAULT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A5-DEFAULT-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 -mattr=-neon,+d16 | FileCheck %s --check-prefix=CORTEX-A5-NONEON
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 -mattr=-vfp2 | FileCheck %s --check-prefix=CORTEX-A5-NOFPU
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a5 -mattr=-vfp2 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A5-NOFPU-FAST
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -float-abi=soft | FileCheck %s --check-prefix=CORTEX-A9-SOFT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -float-abi=soft -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A9-SOFT-FAST
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -float-abi=hard | FileCheck %s --check-prefix=CORTEX-A9-HARD
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -float-abi=hard -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A9-HARD-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a12 | FileCheck %s --check-prefix=CORTEX-A12-DEFAULT
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a9 -float-abi=soft | FileCheck %s --check-prefix=CORTEX-A9-SOFT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a12 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A12-DEFAULT-FAST
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a12 -mattr=-vfp2 | FileCheck %s --check-prefix=CORTEX-A12-NOFPU
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a12 -mattr=-vfp2 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A12-NOFPU-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a12 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a15 | FileCheck %s --check-prefix=CORTEX-A15
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a15 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A15-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a15 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a17 | FileCheck %s --check-prefix=CORTEX-A17-DEFAULT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a17 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A17-FAST
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a17 -mattr=-vfp2 | FileCheck %s --check-prefix=CORTEX-A17-NOFPU
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a17 -mattr=-vfp2 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A17-NOFPU-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a17 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=thumbv6m-linux-gnueabi -mcpu=cortex-m0 | FileCheck %s --check-prefix=CORTEX-M0
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv6m-linux-gnueabi -mcpu=cortex-m0 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M0-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv6m-linux-gnueabi -mcpu=cortex-m0 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m3 | FileCheck %s --check-prefix=CORTEX-M3
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m3 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M3-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m3 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m4 -float-abi=soft | FileCheck %s --check-prefix=CORTEX-M4-SOFT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m4 -float-abi=soft -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M4-SOFT-FAST
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m4 -float-abi=hard | FileCheck %s --check-prefix=CORTEX-M4-HARD
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m4 -float-abi=hard -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M4-HARD-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv7m-linux-gnueabi -mcpu=cortex-m4 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218747 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 09:02:17 +00:00
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 -mattr=-vfp2 | FileCheck %s --check-prefix=CORTEX-M7 --check-prefix=CORTEX-M7-SOFT
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 -mattr=-vfp2 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M7-NOFPU-FAST
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218747 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 09:02:17 +00:00
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 -mattr=+fp-only-sp | FileCheck %s --check-prefix=CORTEX-M7 --check-prefix=CORTEX-M7-SINGLE
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 -mattr=+fp-only-sp -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-M7-FAST
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 | FileCheck %s --check-prefix=CORTEX-M7-DOUBLE
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=thumbv7em-linux-gnueabi -mcpu=cortex-m7 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=armv7r-linux-gnueabi -mcpu=cortex-r5 | FileCheck %s --check-prefix=CORTEX-R5
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7r-linux-gnueabi -mcpu=cortex-r5 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-R5-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7r-linux-gnueabi -mcpu=cortex-r5 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a53 | FileCheck %s --check-prefix=CORTEX-A53
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a53 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A53-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a53 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Add Cortex-A57 support git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192591 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:17:07 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a57 | FileCheck %s --check-prefix=CORTEX-A57
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a57 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A57-FAST
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv8-linux-gnueabi -mcpu=cortex-a57 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 | FileCheck %s --check-prefix=CORTEX-A7-CHECK
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A7-CHECK-FAST
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -mattr=-vfp2,-vfp3,-vfp4,-neon | FileCheck %s --check-prefix=CORTEX-A7-NOFPU
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -mattr=-vfp2,-vfp3,-vfp4,-neon -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A7-NOFPU-FAST
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -mattr=+vfp4,-neon | FileCheck %s --check-prefix=CORTEX-A7-FPUV4
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -enable-sign-dependent-rounding-fp-math | FileCheck %s --check-prefix=DYN-ROUNDING
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -mattr=+vfp4,-neon -enable-unsafe-fp-math | FileCheck %s --check-prefix=CORTEX-A7-FPUV4-FAST
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -mattr=+vfp4,,+d16,-neon | FileCheck %s --check-prefix=CORTEX-A7-FPUV4
[ARM] Emit correct build attributes for the relocation models. Patch by Asiri Rathnayake. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209656 91177308-0d34-0410-b5e6-96231b3b80d8
2014-05-27 13:30:21 +00:00
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi -relocation-model=pic | FileCheck %s --check-prefix=RELOC-PIC
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi -relocation-model=static | FileCheck %s --check-prefix=RELOC-OTHER
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi -relocation-model=default | FileCheck %s --check-prefix=RELOC-OTHER
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi -relocation-model=dynamic-no-pic | FileCheck %s --check-prefix=RELOC-OTHER
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi | FileCheck %s --check-prefix=RELOC-OTHER
[ARM] Emit ABI_PCS_R9_use build attribute. Patch by Ben Foster! Differential Revision: http://reviews.llvm.org/D4657 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213944 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 14:03:14 +00:00
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi | FileCheck %s --check-prefix=PCS-R9-USE
; RUN: llc < %s -mtriple=arm-none-linux-gnueabi -arm-reserve-r9 | FileCheck %s --check-prefix=PCS-R9-RESERVE
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv8a (AArch32)
; RUN: llc < %s -mtriple=armv8-none-linux-gnueabi -mcpu=cortex-a57 -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv8-none-linux-gnueabi -mcpu=cortex-a57 -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=armv8-none-linux-gnueabi -mcpu=cortex-a57 | FileCheck %s --check-prefix=NO-STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv7a
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=armv7-none-linux-gnueabi -mcpu=cortex-a7 | FileCheck %s --check-prefix=NO-STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv7r
; RUN: llc < %s -mtriple=armv7r-none-linux-gnueabi -mcpu=cortex-r5 -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv7r-none-linux-gnueabi -mcpu=cortex-r5 -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=armv7r-none-linux-gnueabi -mcpu=cortex-r5 | FileCheck %s --check-prefix=NO-STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv7m
; RUN: llc < %s -mtriple=thumbv7m-none-linux-gnueabi -mcpu=cortex-m3 -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=thumbv7m-none-linux-gnueabi -mcpu=cortex-m3 -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=thumbv7m-none-linux-gnueabi -mcpu=cortex-m3 | FileCheck %s --check-prefix=NO-STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv6
; RUN: llc < %s -mtriple=armv6-none-netbsd-gnueabi -mcpu=arm1136j-s | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv6-none-linux-gnueabi -mcpu=arm1136j-s | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=armv6-none-linux-gnueabi -mcpu=arm1136j-s -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv6-none-linux-gnueabi -mcpu=arm1136j-s -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv6m
; RUN: llc < %s -mtriple=thumb-none-linux-gnueabi -arm-no-strict-align -mcpu=cortex-m0 | FileCheck %s --check-prefix=STRICT-ALIGN
; RUN: llc < %s -mtriple=thumb-none-linux-gnueabi -arm-strict-align -mcpu=cortex-m0 | FileCheck %s --check-prefix=STRICT-ALIGN
; RUN: llc < %s -mtriple=thumbv6m-none-linux-gnueabi -arm-no-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
; RUN: llc < %s -mtriple=thumbv6m-none-linux-gnueabi -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=thumb-none-linux-gnueabi -mcpu=cortex-m0 | FileCheck %s --check-prefix=STRICT-ALIGN
; RUN: llc < %s -mtriple=thumb-none-linux-gnueabi -mcpu=cortex-m0 | FileCheck %s --check-prefix=STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; ARMv5
; RUN: llc < %s -mtriple=armv5-none-linux-gnueabi -mcpu=arm1022e -arm-no-strict-align | FileCheck %s --check-prefix=NO-STRICT-ALIGN
; RUN: llc < %s -mtriple=armv5-none-linux-gnueabi -mcpu=arm1022e -arm-strict-align | FileCheck %s --check-prefix=STRICT-ALIGN
ARM: test default values for TAG_CPU_unaligned_access attribute. It should be on for every target that supports unaligned accesses (e.g. not v6m). Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220912 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-30 17:05:44 +00:00
; RUN: llc < %s -mtriple=armv5-none-linux-gnueabi -mcpu=arm1022e | FileCheck %s --check-prefix=STRICT-ALIGN
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
Move the xscale build attribute test to the proper place and remove the old one. The encoding of build attributes is already tested in CodeGen/ARM/build-attributes-encoding.s git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199393 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-16 15:11:54 +00:00
; XSCALE: .eabi_attribute 6, 5
; XSCALE: .eabi_attribute 8, 1
; XSCALE: .eabi_attribute 9, 1
Emit ABI_FP_rounding attribute. LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When the user has specified this option, the Tag_ABI_FP_rounding attribute should be emitted with value 1. This option currently does not appear to disable transformations and optimizations that assume default floating point rounding behavior, AFAICT, but the intention should be recorded in the build attributes, regardless of what the compiler actually does with the intention. Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223218 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 08:12:26 +00:00
; DYN-ROUNDING: .eabi_attribute 19, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V6: .eabi_attribute 6, 6
; V6: .eabi_attribute 8, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
;; We assume round-to-nearest by default (matches GCC)
; V6-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The default choice made by llc is for a V6 CPU without an FPU.
;; This is not an interesting detail, but for such CPUs, the default intention is to use
;; software floating-point support. The choice is not important for targets without
;; FPU support!
; V6: .eabi_attribute 20, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V6: .eabi_attribute 24, 1
; V6: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; V6-NOT: .eabi_attribute 27
; V6-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; V6-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; V6-NOT: .eabi_attribute 42
; V6-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V6-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite the V6 CPU having no FPU by default, we chose to flush to
;; positive zero here. There's no hardware support doing this, but the
;; fast maths software library might.
; V6-FAST-NOT: .eabi_attribute 20
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V6M: .eabi_attribute 6, 12
Must not produce Tag_CPU_arch_profile for pre-ARMv7 cores (e.g. cortex-m0) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@198945 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-10 16:42:55 +00:00
; V6M-NOT: .eabi_attribute 7
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V6M: .eabi_attribute 8, 0
; V6M: .eabi_attribute 9, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V6M-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The default choice made by llc is for a V6M CPU without an FPU.
;; This is not an interesting detail, but for such CPUs, the default intention is to use
;; software floating-point support. The choice is not important for targets without
;; FPU support!
; V6M: .eabi_attribute 20, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V6M: .eabi_attribute 24, 1
; V6M: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; V6M-NOT: .eabi_attribute 27
; V6M-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; V6M-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; V6M-NOT: .eabi_attribute 42
; V6M-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V6M-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite the V6M CPU having no FPU by default, we chose to flush to
;; positive zero here. There's no hardware support doing this, but the
;; fast maths software library might.
; V6M-FAST-NOT: .eabi_attribute 20
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; ARM1156T2F-S: .cpu arm1156t2f-s
; ARM1156T2F-S: .eabi_attribute 6, 8
; ARM1156T2F-S: .eabi_attribute 8, 1
; ARM1156T2F-S: .eabi_attribute 9, 2
; ARM1156T2F-S: .fpu vfpv2
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; ARM1156T2F-S-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; ARM1156T2F-S: .eabi_attribute 20, 1
; ARM1156T2F-S: .eabi_attribute 21, 1
; ARM1156T2F-S: .eabi_attribute 23, 3
; ARM1156T2F-S: .eabi_attribute 24, 1
; ARM1156T2F-S: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; ARM1156T2F-S-NOT: .eabi_attribute 27
; ARM1156T2F-S-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; ARM1156T2F-S-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; ARM1156T2F-S-NOT: .eabi_attribute 42
; ARM1156T2F-S-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; V6 cores default to flush to positive zero (value 0). Note that value 2 is also equally
;; valid for this core, it's an implementation defined question as to which of 0 and 2 you
;; select. LLVM historically picks 0.
; ARM1156T2F-S-FAST-NOT: .eabi_attribute 20
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V7M: .eabi_attribute 6, 10
; V7M: .eabi_attribute 7, 77
; V7M: .eabi_attribute 8, 0
; V7M: .eabi_attribute 9, 2
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V7M-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The default choice made by llc is for a V7M CPU without an FPU.
;; This is not an interesting detail, but for such CPUs, the default intention is to use
;; software floating-point support. The choice is not important for targets without
;; FPU support!
; V7M: .eabi_attribute 20, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V7M: .eabi_attribute 24, 1
; V7M: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; V7M-NOT: .eabi_attribute 27
; V7M-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; V7M-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; V7M-NOT: .eabi_attribute 42
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; V7M-NOT: .eabi_attribute 44
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; V7M-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite the V7M CPU having no FPU by default, we chose to flush
;; preserving sign. This matches what the hardware would do in the
;; architecture revision were to exist on the current target.
; V7M-FAST: .eabi_attribute 20, 2
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; V7: .syntax unified
; V7: .eabi_attribute 6, 10
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V7-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; In safe-maths mode we default to an IEEE 754 compliant choice.
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; V7: .eabi_attribute 20, 1
; V7: .eabi_attribute 21, 1
; V7: .eabi_attribute 23, 3
; V7: .eabi_attribute 24, 1
; V7: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; V7-NOT: .eabi_attribute 27
; V7-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; V7-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; V7-NOT: .eabi_attribute 42
; V7-NOT: .eabi_attribute 68
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V7-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The default CPU does have an FPU and it must be VFPv3 or better, so it flushes
;; denormals to zero preserving the sign.
; V7-FAST: .eabi_attribute 20, 2
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; V8: .syntax unified
; V8: .eabi_attribute 6, 14
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V8-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; V8: .eabi_attribute 20, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; V8-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The default does have an FPU, and for V8-A, it flushes preserving sign.
; V8-FAST: .eabi_attribute 20, 2
Add a subtarget feature 'v8' to the ARM backend. This allows for targeting the ARMv8 AArch32 variant. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184967 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-26 16:58:26 +00:00
; Vt8: .syntax unified
; Vt8: .eabi_attribute 6, 14
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; Vt8-NOT: .eabi_attribute 19
Tiny patch for proof-of-concept cleanup of ARMAsmPrinter::EmitStartOfAsmFile() Small test for sanity check of resulting ARM .s file. Tested against -r115129. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115133 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 02:45:56 +00:00
[ARMv8] Change hasV8Fp to hasFPARMv8, and other command line options to be more consistent. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190692 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-13 13:46:57 +00:00
; V8-FPARMv8: .syntax unified
; V8-FPARMv8: .eabi_attribute 6, 14
; V8-FPARMv8: .fpu fp-armv8
Add a Subtarget feature 'v8fp' to the ARM backend. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185073 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-27 11:49:26 +00:00
; V8-NEON: .syntax unified
; V8-NEON: .eabi_attribute 6, 14
[arm] Enhance the test case by checking .fpu directive. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@191891 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-03 12:18:56 +00:00
; V8-NEON: .fpu neon
Add a Subtarget feature 'v8fp' to the ARM backend. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185073 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-27 11:49:26 +00:00
; V8-NEON: .eabi_attribute 12, 3
[ARMv8] Change hasV8Fp to hasFPARMv8, and other command line options to be more consistent. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190692 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-13 13:46:57 +00:00
; V8-FPARMv8-NEON: .syntax unified
; V8-FPARMv8-NEON: .eabi_attribute 6, 14
; V8-FPARMv8-NEON: .fpu neon-fp-armv8
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; V8-FPARMv8-NEON: .eabi_attribute 12, 3
; V8-FPARMv8-NEON-CRYPTO: .syntax unified
; V8-FPARMv8-NEON-CRYPTO: .eabi_attribute 6, 14
; V8-FPARMv8-NEON-CRYPTO: .fpu crypto-neon-fp-armv8
; V8-FPARMv8-NEON-CRYPTO: .eabi_attribute 12, 3
Emit unaligned access build attribute for ARM Patch by Charlie Turner. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-08 12:26:22 +00:00
; Tag_CPU_unaligned_access
; NO-STRICT-ALIGN: .eabi_attribute 34, 1
; STRICT-ALIGN: .eabi_attribute 34, 0
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; Tag_CPU_arch 'ARMv7'
; CORTEX-A7-CHECK: .eabi_attribute 6, 10
; CORTEX-A7-NOFPU: .eabi_attribute 6, 10
; CORTEX-A7-FPUV4: .eabi_attribute 6, 10
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_CPU_arch_profile 'A'
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 7, 65
; CORTEX-A7-NOFPU: .eabi_attribute 7, 65
; CORTEX-A7-FPUV4: .eabi_attribute 7, 65
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_ARM_ISA_use
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 8, 1
; CORTEX-A7-NOFPU: .eabi_attribute 8, 1
; CORTEX-A7-FPUV4: .eabi_attribute 8, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_THUMB_ISA_use
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 9, 2
; CORTEX-A7-NOFPU: .eabi_attribute 9, 2
; CORTEX-A7-FPUV4: .eabi_attribute 9, 2
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .fpu neon-vfpv4
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; CORTEX-A7-NOFPU-NOT: .fpu
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-FPUV4: .fpu vfpv4
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A7-CHECK-NOT: .eabi_attribute 19
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_ABI_FP_denormal
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
; CORTEX-A7-CHECK: .eabi_attribute 20, 1
;; The A7 has VFPv3 support by default, so flush preserving sign.
; CORTEX-A7-CHECK-FAST: .eabi_attribute 20, 2
; CORTEX-A7-NOFPU: .eabi_attribute 20, 1
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-A7-NOFPU-FAST: .eabi_attribute 20, 2
; CORTEX-A7-FPUV4: .eabi_attribute 20, 1
;; The VFPv4 FPU flushes preserving sign.
; CORTEX-A7-FPUV4-FAST: .eabi_attribute 20, 2
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_ABI_FP_exceptions
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 21, 1
; CORTEX-A7-NOFPU: .eabi_attribute 21, 1
; CORTEX-A7-FPUV4: .eabi_attribute 21, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_ABI_FP_number_model
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 23, 3
; CORTEX-A7-NOFPU: .eabi_attribute 23, 3
; CORTEX-A7-FPUV4: .eabi_attribute 23, 3
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_ABI_align_needed
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 24, 1
; CORTEX-A7-NOFPU: .eabi_attribute 24, 1
; CORTEX-A7-FPUV4: .eabi_attribute 24, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
ARM: update build attributes for ABI r2.09 Update names for the names as per the current ABI errata. Mark deprecated tags as such. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199576 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-19 08:25:35 +00:00
; Tag_ABI_align_preserved
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 25, 1
; CORTEX-A7-NOFPU: .eabi_attribute 25, 1
; CORTEX-A7-FPUV4: .eabi_attribute 25, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_FP_HP_extension
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 36, 1
; CORTEX-A7-NOFPU: .eabi_attribute 36, 1
; CORTEX-A7-FPUV4: .eabi_attribute 36, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_MPextension_use
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 42, 1
; CORTEX-A7-NOFPU: .eabi_attribute 42, 1
; CORTEX-A7-FPUV4: .eabi_attribute 42, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_DIV_use
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 44, 2
; CORTEX-A7-NOFPU: .eabi_attribute 44, 2
; CORTEX-A7-FPUV4: .eabi_attribute 44, 2
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
; Tag_Virtualization_use
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
; CORTEX-A7-CHECK: .eabi_attribute 68, 3
; CORTEX-A7-NOFPU: .eabi_attribute 68, 3
; CORTEX-A7-FPUV4: .eabi_attribute 68, 3
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-DEFAULT: .cpu cortex-a5
; CORTEX-A5-DEFAULT: .eabi_attribute 6, 10
; CORTEX-A5-DEFAULT: .eabi_attribute 7, 65
; CORTEX-A5-DEFAULT: .eabi_attribute 8, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 9, 2
; CORTEX-A5-DEFAULT: .fpu neon-vfpv4
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A5-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-DEFAULT: .eabi_attribute 20, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 21, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 23, 3
; CORTEX-A5-DEFAULT: .eabi_attribute 24, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 25, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 42, 1
; CORTEX-A5-DEFAULT: .eabi_attribute 68, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A5-DEFAULT-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A5 defaults to a VFPv4 FPU, so it flushed preserving sign when -ffast-math
;; is given.
; CORTEX-A5-DEFAULT-FAST: .eabi_attribute 20, 2
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-NONEON: .cpu cortex-a5
; CORTEX-A5-NONEON: .eabi_attribute 6, 10
; CORTEX-A5-NONEON: .eabi_attribute 7, 65
; CORTEX-A5-NONEON: .eabi_attribute 8, 1
; CORTEX-A5-NONEON: .eabi_attribute 9, 2
; CORTEX-A5-NONEON: .fpu vfpv4-d16
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-NONEON: .eabi_attribute 20, 1
; CORTEX-A5-NONEON: .eabi_attribute 21, 1
; CORTEX-A5-NONEON: .eabi_attribute 23, 3
; CORTEX-A5-NONEON: .eabi_attribute 24, 1
; CORTEX-A5-NONEON: .eabi_attribute 25, 1
; CORTEX-A5-NONEON: .eabi_attribute 42, 1
; CORTEX-A5-NONEON: .eabi_attribute 68, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A5-NONEON-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A5 defaults to a VFPv4 FPU, so it flushed preserving sign when -ffast-math
;; is given.
; CORTEX-A5-NONEON-FAST: .eabi_attribute 20, 2
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-NOFPU: .cpu cortex-a5
; CORTEX-A5-NOFPU: .eabi_attribute 6, 10
; CORTEX-A5-NOFPU: .eabi_attribute 7, 65
; CORTEX-A5-NOFPU: .eabi_attribute 8, 1
; CORTEX-A5-NOFPU: .eabi_attribute 9, 2
; CORTEX-A5-NOFPU-NOT: .fpu
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A5-NOFPU-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Enable FeatureMP for Cortex-A5 by default. Patch by Oliver Stannard. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195640 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-25 13:17:15 +00:00
; CORTEX-A5-NOFPU: .eabi_attribute 20, 1
; CORTEX-A5-NOFPU: .eabi_attribute 21, 1
; CORTEX-A5-NOFPU: .eabi_attribute 23, 3
; CORTEX-A5-NOFPU: .eabi_attribute 24, 1
; CORTEX-A5-NOFPU: .eabi_attribute 25, 1
; CORTEX-A5-NOFPU: .eabi_attribute 42, 1
; CORTEX-A5-NOFPU: .eabi_attribute 68, 1
[ARM] add the overlooked tests for Cortex-A7 build attributes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195365 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-21 16:22:39 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A5-NOFPU-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-A5-NOFPU-FAST: .eabi_attribute 20, 2
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A9-SOFT: .cpu cortex-a9
; CORTEX-A9-SOFT: .eabi_attribute 6, 10
; CORTEX-A9-SOFT: .eabi_attribute 7, 65
; CORTEX-A9-SOFT: .eabi_attribute 8, 1
; CORTEX-A9-SOFT: .eabi_attribute 9, 2
; CORTEX-A9-SOFT: .fpu neon
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A9-SOFT-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A9-SOFT: .eabi_attribute 20, 1
; CORTEX-A9-SOFT: .eabi_attribute 21, 1
; CORTEX-A9-SOFT: .eabi_attribute 23, 3
; CORTEX-A9-SOFT: .eabi_attribute 24, 1
; CORTEX-A9-SOFT: .eabi_attribute 25, 1
; CORTEX-A9-SOFT-NOT: .eabi_attribute 27
; CORTEX-A9-SOFT-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-A9-SOFT: .eabi_attribute 36, 1
Remove the cortex-a9-mp CPU. This CPU definition is redundant. The Cortex-A9 is defined as supporting multiprocessing extensions. Remove its definition and update appropriate tests. LLVM defines both a cortex-a9 CPU and a cortex-a9-mp CPU. The only difference between the two CPU definitions in ARM.td is that cortex-a9-mp contains the feature FeatureMP for multiprocessing extensions. This is redundant since the Cortex-A9 is defined as having multiprocessing extensions in the TRMs. armcc also defines the Cortex-A9 as having multiprocessing extensions by default. Change-Id: Ifcadaa6c322be0a33d9d2a39cfdd7da1d75981a7 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@221166 91177308-0d34-0410-b5e6-96231b3b80d8
2014-11-03 17:38:00 +00:00
; CORTEX-A9-SOFT: .eabi_attribute 42, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A9-SOFT: .eabi_attribute 68, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A9-SOFT-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A9 defaults to a VFPv3 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-A9-SOFT-FAST: .eabi_attribute 20, 2
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A9-HARD: .cpu cortex-a9
; CORTEX-A9-HARD: .eabi_attribute 6, 10
; CORTEX-A9-HARD: .eabi_attribute 7, 65
; CORTEX-A9-HARD: .eabi_attribute 8, 1
; CORTEX-A9-HARD: .eabi_attribute 9, 2
; CORTEX-A9-HARD: .fpu neon
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A9-HARD-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A9-HARD: .eabi_attribute 20, 1
; CORTEX-A9-HARD: .eabi_attribute 21, 1
; CORTEX-A9-HARD: .eabi_attribute 23, 3
; CORTEX-A9-HARD: .eabi_attribute 24, 1
; CORTEX-A9-HARD: .eabi_attribute 25, 1
; CORTEX-A9-HARD-NOT: .eabi_attribute 27
; CORTEX-A9-HARD: .eabi_attribute 28, 1
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-A9-HARD: .eabi_attribute 36, 1
Remove the cortex-a9-mp CPU. This CPU definition is redundant. The Cortex-A9 is defined as supporting multiprocessing extensions. Remove its definition and update appropriate tests. LLVM defines both a cortex-a9 CPU and a cortex-a9-mp CPU. The only difference between the two CPU definitions in ARM.td is that cortex-a9-mp contains the feature FeatureMP for multiprocessing extensions. This is redundant since the Cortex-A9 is defined as having multiprocessing extensions in the TRMs. armcc also defines the Cortex-A9 as having multiprocessing extensions by default. Change-Id: Ifcadaa6c322be0a33d9d2a39cfdd7da1d75981a7 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@221166 91177308-0d34-0410-b5e6-96231b3b80d8
2014-11-03 17:38:00 +00:00
; CORTEX-A9-HARD: .eabi_attribute 42, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A9-HARD: .eabi_attribute 68, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A9-HARD-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A9 defaults to a VFPv3 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-A9-HARD-FAST: .eabi_attribute 20, 2
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; CORTEX-A12-DEFAULT: .cpu cortex-a12
; CORTEX-A12-DEFAULT: .eabi_attribute 6, 10
; CORTEX-A12-DEFAULT: .eabi_attribute 7, 65
; CORTEX-A12-DEFAULT: .eabi_attribute 8, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 9, 2
; CORTEX-A12-DEFAULT: .fpu neon-vfpv4
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A12-DEFAULT-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; CORTEX-A12-DEFAULT: .eabi_attribute 20, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 21, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 23, 3
; CORTEX-A12-DEFAULT: .eabi_attribute 24, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 25, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 42, 1
; CORTEX-A12-DEFAULT: .eabi_attribute 44, 2
; CORTEX-A12-DEFAULT: .eabi_attribute 68, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A12-DEFAULT-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A12 defaults to a VFPv3 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-A12-DEFAULT-FAST: .eabi_attribute 20, 2
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; CORTEX-A12-NOFPU: .cpu cortex-a12
; CORTEX-A12-NOFPU: .eabi_attribute 6, 10
; CORTEX-A12-NOFPU: .eabi_attribute 7, 65
; CORTEX-A12-NOFPU: .eabi_attribute 8, 1
; CORTEX-A12-NOFPU: .eabi_attribute 9, 2
; CORTEX-A12-NOFPU-NOT: .fpu
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A12-NOFPU-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
Add support for Cortex-A12. Patch by Oliver Stannard! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195448 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-22 11:53:16 +00:00
; CORTEX-A12-NOFPU: .eabi_attribute 20, 1
; CORTEX-A12-NOFPU: .eabi_attribute 21, 1
; CORTEX-A12-NOFPU: .eabi_attribute 23, 3
; CORTEX-A12-NOFPU: .eabi_attribute 24, 1
; CORTEX-A12-NOFPU: .eabi_attribute 25, 1
; CORTEX-A12-NOFPU: .eabi_attribute 42, 1
; CORTEX-A12-NOFPU: .eabi_attribute 44, 2
; CORTEX-A12-NOFPU: .eabi_attribute 68, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A12-NOFPU-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-A12-NOFPU-FAST: .eabi_attribute 20, 2
[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
; CORTEX-A15: .cpu cortex-a15
; CORTEX-A15: .eabi_attribute 6, 10
; CORTEX-A15: .eabi_attribute 7, 65
; CORTEX-A15: .eabi_attribute 8, 1
; CORTEX-A15: .eabi_attribute 9, 2
; CORTEX-A15: .fpu neon-vfpv4
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A15-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
; CORTEX-A15: .eabi_attribute 20, 1
; CORTEX-A15: .eabi_attribute 21, 1
; CORTEX-A15: .eabi_attribute 23, 3
; CORTEX-A15: .eabi_attribute 24, 1
; CORTEX-A15: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A15-NOT: .eabi_attribute 27
; CORTEX-A15-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-A15: .eabi_attribute 36, 1
; CORTEX-A15: .eabi_attribute 42, 1
; CORTEX-A15: .eabi_attribute 44, 2
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A15: .eabi_attribute 68, 3
[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A15-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A15 defaults to a VFPv3 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-A15-FAST: .eabi_attribute 20, 2
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; CORTEX-A17-DEFAULT: .cpu cortex-a17
; CORTEX-A17-DEFAULT: .eabi_attribute 6, 10
; CORTEX-A17-DEFAULT: .eabi_attribute 7, 65
; CORTEX-A17-DEFAULT: .eabi_attribute 8, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 9, 2
; CORTEX-A17-DEFAULT: .fpu neon-vfpv4
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A17-DEFAULT-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; CORTEX-A17-DEFAULT: .eabi_attribute 20, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 21, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 23, 3
; CORTEX-A17-DEFAULT: .eabi_attribute 24, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 25, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 42, 1
; CORTEX-A17-DEFAULT: .eabi_attribute 44, 2
; CORTEX-A17-DEFAULT: .eabi_attribute 68, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A17-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A17 defaults to a VFPv3 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-A17-FAST: .eabi_attribute 20, 2
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; CORTEX-A17-NOFPU: .cpu cortex-a17
; CORTEX-A17-NOFPU: .eabi_attribute 6, 10
; CORTEX-A17-NOFPU: .eabi_attribute 7, 65
; CORTEX-A17-NOFPU: .eabi_attribute 8, 1
; CORTEX-A17-NOFPU: .eabi_attribute 9, 2
; CORTEX-A17-NOFPU-NOT: .fpu
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A17-NOFPU-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
Adds support for the Cortex-A17 to the ARM backend Patch by Matthew Wahab. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219606 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-13 10:22:19 +00:00
; CORTEX-A17-NOFPU: .eabi_attribute 20, 1
; CORTEX-A17-NOFPU: .eabi_attribute 21, 1
; CORTEX-A17-NOFPU: .eabi_attribute 23, 3
; CORTEX-A17-NOFPU: .eabi_attribute 24, 1
; CORTEX-A17-NOFPU: .eabi_attribute 25, 1
; CORTEX-A17-NOFPU: .eabi_attribute 42, 1
; CORTEX-A17-NOFPU: .eabi_attribute 44, 2
; CORTEX-A17-NOFPU: .eabi_attribute 68, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A17-NOFPU-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-A17-NOFPU-FAST: .eabi_attribute 20, 2
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-M0: .cpu cortex-m0
; CORTEX-M0: .eabi_attribute 6, 12
Must not produce Tag_CPU_arch_profile for pre-ARMv7 cores (e.g. cortex-m0) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@198945 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-10 16:42:55 +00:00
; CORTEX-M0-NOT: .eabi_attribute 7
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-M0: .eabi_attribute 8, 0
; CORTEX-M0: .eabi_attribute 9, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M0-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
; CORTEX-M0: .eabi_attribute 20, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-M0: .eabi_attribute 24, 1
; CORTEX-M0: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-M0-NOT: .eabi_attribute 27
; CORTEX-M0-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-M0-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-M0-NOT: .eabi_attribute 42
; CORTEX-M0-NOT: .eabi_attribute 68
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M0-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite the M0 CPU having no FPU in this scenario, we chose to
;; flush to positive zero here. There's no hardware support doing
;; this, but the fast maths software library might and such behaviour
;; would match hardware support on this architecture revision if it
;; existed.
; CORTEX-M0-FAST-NOT: .eabi_attribute 20
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-M3: .cpu cortex-m3
; CORTEX-M3: .eabi_attribute 6, 10
; CORTEX-M3: .eabi_attribute 7, 77
; CORTEX-M3: .eabi_attribute 8, 0
; CORTEX-M3: .eabi_attribute 9, 2
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M3-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-M3: .eabi_attribute 20, 1
; CORTEX-M3: .eabi_attribute 21, 1
; CORTEX-M3: .eabi_attribute 23, 3
; CORTEX-M3: .eabi_attribute 24, 1
; CORTEX-M3: .eabi_attribute 25, 1
; CORTEX-M3-NOT: .eabi_attribute 27
; CORTEX-M3-NOT: .eabi_attribute 28
; CORTEX-M3-NOT: .eabi_attribute 36
; CORTEX-M3-NOT: .eabi_attribute 42
; CORTEX-M3-NOT: .eabi_attribute 44
; CORTEX-M3-NOT: .eabi_attribute 68
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M3-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-M3-FAST: .eabi_attribute 20, 2
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-M4-SOFT: .cpu cortex-m4
; CORTEX-M4-SOFT: .eabi_attribute 6, 13
; CORTEX-M4-SOFT: .eabi_attribute 7, 77
; CORTEX-M4-SOFT: .eabi_attribute 8, 0
; CORTEX-M4-SOFT: .eabi_attribute 9, 2
; CORTEX-M4-SOFT: .fpu vfpv4-d16
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M4-SOFT-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-M4-SOFT: .eabi_attribute 20, 1
; CORTEX-M4-SOFT: .eabi_attribute 21, 1
; CORTEX-M4-SOFT: .eabi_attribute 23, 3
; CORTEX-M4-SOFT: .eabi_attribute 24, 1
; CORTEX-M4-SOFT: .eabi_attribute 25, 1
; CORTEX-M4-SOFT: .eabi_attribute 27, 1
; CORTEX-M4-SOFT-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-M4-SOFT: .eabi_attribute 36, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-M4-SOFT-NOT: .eabi_attribute 42
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-M4-SOFT-NOT: .eabi_attribute 44
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-M4-SOFT-NOT: .eabi_attribute 68
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M4-SOFT-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The M4 defaults to a VFPv4 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-M4-SOFT-FAST: .eabi_attribute 20, 2
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-M4-HARD: .cpu cortex-m4
; CORTEX-M4-HARD: .eabi_attribute 6, 13
; CORTEX-M4-HARD: .eabi_attribute 7, 77
; CORTEX-M4-HARD: .eabi_attribute 8, 0
; CORTEX-M4-HARD: .eabi_attribute 9, 2
; CORTEX-M4-HARD: .fpu vfpv4-d16
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M4-HARD-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-M4-HARD: .eabi_attribute 20, 1
; CORTEX-M4-HARD: .eabi_attribute 21, 1
; CORTEX-M4-HARD: .eabi_attribute 23, 3
; CORTEX-M4-HARD: .eabi_attribute 24, 1
; CORTEX-M4-HARD: .eabi_attribute 25, 1
; CORTEX-M4-HARD: .eabi_attribute 27, 1
; CORTEX-M4-HARD: .eabi_attribute 28, 1
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-M4-HARD: .eabi_attribute 36, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-M4-HARD-NOT: .eabi_attribute 42
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-M4-HARD-NOT: .eabi_attribute 44
Fix broken CHECK lines git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201479 91177308-0d34-0410-b5e6-96231b3b80d8
2014-02-16 07:31:05 +00:00
; CORTEX-M4-HARD-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M4-HARD-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The M4 defaults to a VFPv4 FPU, so it flushes preseving sign when
;; -ffast-math is specified.
; CORTEX-M4-HARD-FAST: .eabi_attribute 20, 2
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218747 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 09:02:17 +00:00
; CORTEX-M7: .cpu cortex-m7
; CORTEX-M7: .eabi_attribute 6, 13
; CORTEX-M7: .eabi_attribute 7, 77
; CORTEX-M7: .eabi_attribute 8, 0
; CORTEX-M7: .eabi_attribute 9, 2
; CORTEX-M7-SOFT-NOT: .fpu
; CORTEX-M7-SINGLE: .fpu fpv5-d16
; CORTEX-M7-DOUBLE: .fpu fpv5-d16
; CORTEX-M7: .eabi_attribute 17, 1
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M7-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218747 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 09:02:17 +00:00
; CORTEX-M7: .eabi_attribute 20, 1
; CORTEX-M7: .eabi_attribute 21, 1
; CORTEX-M7: .eabi_attribute 23, 3
; CORTEX-M7: .eabi_attribute 24, 1
; CORTEX-M7: .eabi_attribute 25, 1
; CORTEX-M7-SOFT-NOT: .eabi_attribute 27
; CORTEX-M7-SINGLE: .eabi_attribute 27, 1
; CORTEX-M7-DOUBLE-NOT: .eabi_attribute 27
; CORTEX-M7: .eabi_attribute 36, 1
; CORTEX-M7: .eabi_attribute 14, 0
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-M7-NOFPU-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The M7 has the ARMv8 FP unit, which always flushes preserving sign.
; CORTEX-M7-FAST: .eabi_attribute 20, 2
;; Despite there being no FPU, we chose to flush to zero preserving
;; sign. This matches what the hardware would do for this architecture
;; revision.
; CORTEX-M7-NOFPU-FAST: .eabi_attribute 20, 2
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-R5: .cpu cortex-r5
; CORTEX-R5: .eabi_attribute 6, 10
; CORTEX-R5: .eabi_attribute 7, 82
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-R5: .eabi_attribute 8, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-R5: .eabi_attribute 9, 2
[arm] Implement eabi_attribute, cpu, and fpu directives. This commit allows the ARM integrated assembler to parse and assemble the code with .eabi_attribute, .cpu, and .fpu directives. To implement the feature, this commit moves the code from AttrEmitter to ARMTargetStreamers, and several new test cases related to cortex-m4, cortex-r5, and cortex-a15 are added. Besides, this commit also change the Subtarget->isFPOnlySP() to Subtarget->hasD16() to match the usage of .fpu directive. This commit changes the test cases: * Several .eabi_attribute directives in 2010-09-29-mc-asm-header-test.ll are removed because the .fpu directive already cover the functionality. * In the Cortex-A15 test case, the value for Tag_Advanced_SIMD_arch has be changed from 1 to 2, which is more precise. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-28 17:51:12 +00:00
; CORTEX-R5: .fpu vfpv3-d16
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-R5-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-R5: .eabi_attribute 20, 1
; CORTEX-R5: .eabi_attribute 21, 1
; CORTEX-R5: .eabi_attribute 23, 3
; CORTEX-R5: .eabi_attribute 24, 1
; CORTEX-R5: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-R5: .eabi_attribute 27, 1
; CORTEX-R5-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-R5-NOT: .eabi_attribute 36
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-R5-NOT: .eabi_attribute 42
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-R5: .eabi_attribute 44, 2
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-R5-NOT: .eabi_attribute 68
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-R5-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The R5 has the VFPv3 FP unit, which always flushes preserving sign.
; CORTEX-R5-FAST: .eabi_attribute 20, 2
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-A53: .cpu cortex-a53
; CORTEX-A53: .eabi_attribute 6, 14
; CORTEX-A53: .eabi_attribute 7, 65
; CORTEX-A53: .eabi_attribute 8, 1
; CORTEX-A53: .eabi_attribute 9, 2
Add subtarget feature support for Cortex-A53 Some previous implicit defaults have changed, for example FP and NEON are now on by default. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192590 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:16:57 +00:00
; CORTEX-A53: .fpu crypto-neon-fp-armv8
; CORTEX-A53: .eabi_attribute 12, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A53-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
; CORTEX-A53: .eabi_attribute 20, 1
[ARM] Improve build attributes emission. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192111 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-07 16:55:23 +00:00
; CORTEX-A53: .eabi_attribute 24, 1
; CORTEX-A53: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A53-NOT: .eabi_attribute 27
; CORTEX-A53-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-A53: .eabi_attribute 36, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A53: .eabi_attribute 42, 1
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-A53-NOT: .eabi_attribute 44
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A53: .eabi_attribute 68, 3
[ARMv8] Emit the proper .fpu directive. Patch by Bradley Smith! git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190683 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-13 11:51:52 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A53-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A53 has the ARMv8 FP unit, which always flushes preserving sign.
; CORTEX-A53-FAST: .eabi_attribute 20, 2
Add Cortex-A57 support git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192591 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:17:07 +00:00
; CORTEX-A57: .cpu cortex-a57
; CORTEX-A57: .eabi_attribute 6, 14
; CORTEX-A57: .eabi_attribute 7, 65
; CORTEX-A57: .eabi_attribute 8, 1
; CORTEX-A57: .eabi_attribute 9, 2
; CORTEX-A57: .fpu crypto-neon-fp-armv8
; CORTEX-A57: .eabi_attribute 12, 3
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A57-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; We default to IEEE 754 compliance
; CORTEX-A57: .eabi_attribute 20, 1
Add Cortex-A57 support git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192591 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:17:07 +00:00
; CORTEX-A57: .eabi_attribute 24, 1
; CORTEX-A57: .eabi_attribute 25, 1
[ARM] Fix Tag_ABI_HardFP_use build attribute Fix Tag_ABI_HardFP_use build attribute to handle single precision FP, replace deprecated Tag_ABI_HardFP_use value of 3 with 0 and also add some tests for Tag_ABI_VFP_args. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193856 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 11:21:16 +00:00
; CORTEX-A57-NOT: .eabi_attribute 27
; CORTEX-A57-NOT: .eabi_attribute 28
[ARM] Add support for FP_HP_extension build attribute git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194470 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-12 10:38:05 +00:00
; CORTEX-A57: .eabi_attribute 36, 1
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A57: .eabi_attribute 42, 1
[ARM] Do not generate Tag_DIV_use=AllowDIVExt when hardware div is non-optional: it should have the default value of AllowDIVIfExists git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199638 91177308-0d34-0410-b5e6-96231b3b80d8
2014-01-20 10:18:42 +00:00
; CORTEX-A57-NOT: .eabi_attribute 44
[ARM] Add Virtualization subtarget feature and more build attributes in this area Add a Virtualization ARM subtarget feature along with adding proper build attribute emission for Tag_Virtualization_use (encodes Virtualization and TrustZone) and Tag_MPextension_use. Also rework test/CodeGen/ARM/2010-10-19-mc-elf-objheader.ll testcase to something that is more maintainable. This changes the focus of this testcase away from testing CPU defaults (which is tested elsewhere), onto specifically testing that attributes are encoded correctly. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193859 91177308-0d34-0410-b5e6-96231b3b80d8
2013-11-01 13:27:35 +00:00
; CORTEX-A57: .eabi_attribute 68, 3
Add Cortex-A57 support git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192591 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 13:17:07 +00:00
Add tests for default value of Tag_ABI_FP_rounding. Change-Id: I051866d073fc6ce87ce3e693a3762da6d81f4393 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223217 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-03 07:59:50 +00:00
; CORTEX-A57-FAST-NOT: .eabi_attribute 19
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
;; The A57 has the ARMv8 FP unit, which always flushes preserving sign.
; CORTEX-A57-FAST: .eabi_attribute 20, 2
[ARM] Emit correct build attributes for the relocation models. Patch by Asiri Rathnayake. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209656 91177308-0d34-0410-b5e6-96231b3b80d8
2014-05-27 13:30:21 +00:00
; RELOC-PIC: .eabi_attribute 15, 1
; RELOC-PIC: .eabi_attribute 16, 1
; RELOC-PIC: .eabi_attribute 17, 2
; RELOC-OTHER: .eabi_attribute 17, 1
[ARM] Emit ABI_PCS_R9_use build attribute. Patch by Ben Foster! Differential Revision: http://reviews.llvm.org/D4657 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213944 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-25 14:03:14 +00:00
; PCS-R9-USE: .eabi_attribute 14, 0
; PCS-R9-RESERVE: .eabi_attribute 14, 3
Tiny patch for proof-of-concept cleanup of ARMAsmPrinter::EmitStartOfAsmFile() Small test for sanity check of resulting ARM .s file. Tested against -r115129. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115133 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 02:45:56 +00:00
define i32 @f(i64 %z) {
Emit Tag_ABI_FP_denormal correctly in fast-math mode. The default ARM floating-point mode does not support IEEE 754 mode exactly. Of relevance to this patch is that input denormals are flushed to zero. The way in which they're flushed to zero depends on the architecture, * For VFPv2, it is implementation defined as to whether the sign of zero is preserved. * For VFPv3 and above, the sign of zero is always preserved when a denormal is flushed to zero. When FP support has been disabled, the strategy taken by this patch is to assume the software support will mirror the behaviour of the hardware support for the target *if it existed*. That is, for architectures which can only have VFPv2, it is assumed the software will flush to positive zero. For later architectures it is assumed the software will flush to zero preserving sign. Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223110 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-02 08:22:29 +00:00
ret i32 0
Tiny patch for proof-of-concept cleanup of ARMAsmPrinter::EmitStartOfAsmFile() Small test for sanity check of resulting ARM .s file. Tested against -r115129. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115133 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 02:45:56 +00:00
}
Reference in New Issue Copy Permalink