6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-07-24 22:24:54 +00:00
Code Issues Projects Releases Wiki Activity

[AArch64] Disable some optimization cases for type conversion from sint to fp, because those optimization cases are micro-architecture dependent and only make sense for Cyclone. A new predicate Cyclone is introduced in .td file.

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213827 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Jiangning Liu
2014-07-24 01:29:59 +00:00
parent d28cfd150b
commit 1bc34d71b7
2 changed files with 30 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
lib/Target/AArch64/AArch64InstrInfo.td
View File

@@ -24,6 +24,7 @@ def HasCRC : Predicate<"Subtarget->hasCRC()">,
AssemblerPredicate<"FeatureCRC", "crc">; AssemblerPredicate<"FeatureCRC", "crc">;
def IsLE : Predicate<"Subtarget->isLittleEndian()">; def IsLE : Predicate<"Subtarget->isLittleEndian()">;
def IsBE : Predicate<"!Subtarget->isLittleEndian()">; def IsBE : Predicate<"!Subtarget->isLittleEndian()">;
def IsCyclone : Predicate<"Subtarget->isCyclone()">;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// AArch64-specific DAG Nodes. // AArch64-specific DAG Nodes.
@@ -4386,7 +4387,7 @@ class SExtLoadi8CVTf32Pat<dag addrmode, dag INST>
0), 0),
dsub)), dsub)),
0), 0),
ssub)))>, Requires<[NotForCodeSize]>; ssub)))>, Requires<[NotForCodeSize, IsCyclone]>;
def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext), def : SExtLoadi8CVTf32Pat<(ro8.Wpat GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext),
(LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>; (LDRBroW GPR64sp:$Rn, GPR32:$Rm, ro8.Wext:$ext)>;
@@ -4439,7 +4440,7 @@ class SExtLoadi16CVTf64Pat<dag addrmode, dag INST>
0), 0),
dsub)), dsub)),
0), 0),
dsub)))>, Requires<[NotForCodeSize]>; dsub)))>, Requires<[NotForCodeSize, IsCyclone]>;
def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext), def : SExtLoadi16CVTf64Pat<(ro16.Wpat GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext),
(LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>; (LDRHroW GPR64sp:$Rn, GPR32:$Rm, ro16.Wext:$ext)>;

27
test/CodeGen/AArch64/arm64-scvt.ll
View File

@@ -1,4 +1,5 @@
; RUN: llc < %s -march=arm64 -aarch64-neon-syntax=apple | FileCheck %s ; RUN: llc < %s -march=arm64 -mcpu=cyclone -aarch64-neon-syntax=apple | FileCheck %s
; RUN: llc < %s -march=arm64 -mcpu=cortex-a57 | FileCheck --check-prefix=CHECK-A57 %s
; rdar://13082402 ; rdar://13082402
define float @t1(i32* nocapture %src) nounwind ssp { define float @t1(i32* nocapture %src) nounwind ssp {
@@ -409,6 +410,10 @@ define float @sfct1(i8* nocapture %sp0) {
; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]] ; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]]
; CHECK-NEXT: fmul s0, [[REG]], [[REG]] ; CHECK-NEXT: fmul s0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct1:
; CHECK-A57: ldrsb w[[REGNUM:[0-9]+]], [x0, #1]
; CHECK-A57-NEXT: scvtf [[REG:s[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul s0, [[REG]], [[REG]]
entry: entry:
%addr = getelementptr i8* %sp0, i64 1 %addr = getelementptr i8* %sp0, i64 1
%pix_sp0.0.copyload = load i8* %addr, align 1 %pix_sp0.0.copyload = load i8* %addr, align 1
@@ -466,6 +471,10 @@ define float @sfct5(i8* nocapture %sp0, i64 %offset) {
; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]] ; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]]
; CHECK-NEXT: fmul s0, [[REG]], [[REG]] ; CHECK-NEXT: fmul s0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct5:
; CHECK-A57: ldrsb w[[REGNUM:[0-9]+]], [x0, x1]
; CHECK-A57-NEXT: scvtf [[REG:s[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul s0, [[REG]], [[REG]]
entry: entry:
%addr = getelementptr i8* %sp0, i64 %offset %addr = getelementptr i8* %sp0, i64 %offset
%pix_sp0.0.copyload = load i8* %addr, align 1 %pix_sp0.0.copyload = load i8* %addr, align 1
@@ -536,6 +545,10 @@ define double @sfct10(i16* nocapture %sp0) {
; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]] ; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]]
; CHECK-NEXT: fmul d0, [[REG]], [[REG]] ; CHECK-NEXT: fmul d0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct10:
; CHECK-A57: ldrsh w[[REGNUM:[0-9]+]], [x0, #2]
; CHECK-A57-NEXT: scvtf [[REG:d[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul d0, [[REG]], [[REG]]
entry: entry:
%addr = getelementptr i16* %sp0, i64 1 %addr = getelementptr i16* %sp0, i64 1
%pix_sp0.0.copyload = load i16* %addr, align 1 %pix_sp0.0.copyload = load i16* %addr, align 1
@@ -592,6 +605,10 @@ define double @sfct14(i16* nocapture %sp0, i64 %offset) {
; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]] ; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]]
; CHECK-NEXT: fmul d0, [[REG]], [[REG]] ; CHECK-NEXT: fmul d0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct14:
; CHECK-A57: ldrsh w[[REGNUM:[0-9]+]], [x0, x1, lsl #1]
; CHECK-A57-NEXT: scvtf [[REG:d[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul d0, [[REG]], [[REG]]
entry: entry:
%addr = getelementptr i16* %sp0, i64 %offset %addr = getelementptr i16* %sp0, i64 %offset
%pix_sp0.0.copyload = load i16* %addr, align 1 %pix_sp0.0.copyload = load i16* %addr, align 1
@@ -636,6 +653,10 @@ entry:
; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.4s v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]] ; CHECK: scvtf [[REG:s[0-9]+]], s[[SEXTREG]]
; CHECK-NEXT: fmul s0, [[REG]], [[REG]] ; CHECK-NEXT: fmul s0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct17:
; CHECK-A57: ldursb w[[REGNUM:[0-9]+]], [x0, #-1]
; CHECK-A57-NEXT: scvtf [[REG:s[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul s0, [[REG]], [[REG]]
%bitcast = ptrtoint i8* %sp0 to i64 %bitcast = ptrtoint i8* %sp0 to i64
%add = add i64 %bitcast, -1 %add = add i64 %bitcast, -1
%addr = inttoptr i64 %add to i8* %addr = inttoptr i64 %add to i8*
@@ -713,6 +734,10 @@ define double @sfct22(i16* nocapture %sp0) {
; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0 ; CHECK-NEXT: sshll.2d v[[SEXTREG:[0-9]+]], [[SEXTREG1]], #0
; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]] ; CHECK: scvtf [[REG:d[0-9]+]], d[[SEXTREG]]
; CHECK-NEXT: fmul d0, [[REG]], [[REG]] ; CHECK-NEXT: fmul d0, [[REG]], [[REG]]
; CHECK-A57-LABEL: sfct22:
; CHECK-A57: ldursh w[[REGNUM:[0-9]+]], [x0, #1]
; CHECK-A57-NEXT: scvtf [[REG:d[0-9]+]], w[[REGNUM]]
; CHECK-A57-NEXT: fmul d0, [[REG]], [[REG]]
%bitcast = ptrtoint i16* %sp0 to i64 %bitcast = ptrtoint i16* %sp0 to i64
%add = add i64 %bitcast, 1 %add = add i64 %bitcast, 1
%addr = inttoptr i64 %add to i16* %addr = inttoptr i64 %add to i16*