llvm-6502/lib/Target/R600/AMDILInstrInfo.td
Tom Stellard 6b7d99d473 R600: New control flow for SI v2
This patch replaces the control flow handling with a new
pass which structurize the graph before transforming it to
machine instruction. This has a couple of different advantages
and currently fixes 20 piglit tests without a single regression.

It is now a general purpose transformation that could be not
only be used for SI/R6xx, but also for other hardware
implementations that use a form of structurized control flow.

v2: further cleanup, fixes and documentation

Patch by: Christian König

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170591 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-19 22:10:31 +00:00

209 lines
9.4 KiB
TableGen

//===------------ AMDILInstrInfo.td - AMDIL Target ------*-tablegen-*------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//==-----------------------------------------------------------------------===//
//
// This file describes the AMDIL instructions in TableGen format.
//
//===----------------------------------------------------------------------===//
// AMDIL Instruction Predicate Definitions
// Predicate that is set to true if the hardware supports double precision
// divide
def HasHWDDiv : Predicate<"Subtarget.device()"
"->getGeneration() > AMDGPUDeviceInfo::HD4XXX && "
"Subtarget.device()->usesHardware(AMDGPUDeviceInfo::DoubleOps)">;
// Predicate that is set to true if the hardware supports double, but not double
// precision divide in hardware
def HasSWDDiv : Predicate<"Subtarget.device()"
"->getGeneration() == AMDGPUDeviceInfo::HD4XXX &&"
"Subtarget.device()->usesHardware(AMDGPUDeviceInfo::DoubleOps)">;
// Predicate that is set to true if the hardware support 24bit signed
// math ops. Otherwise a software expansion to 32bit math ops is used instead.
def HasHWSign24Bit : Predicate<"Subtarget.device()"
"->getGeneration() > AMDGPUDeviceInfo::HD5XXX">;
// Predicate that is set to true if 64bit operations are supported or not
def HasHW64Bit : Predicate<"Subtarget.device()"
"->usesHardware(AMDGPUDeviceInfo::LongOps)">;
def HasSW64Bit : Predicate<"Subtarget.device()"
"->usesSoftware(AMDGPUDeviceInfo::LongOps)">;
// Predicate that is set to true if the timer register is supported
def HasTmrRegister : Predicate<"Subtarget.device()"
"->isSupported(AMDGPUDeviceInfo::TmrReg)">;
// Predicate that is true if we are at least evergreen series
def HasDeviceIDInst : Predicate<"Subtarget.device()"
"->getGeneration() >= AMDGPUDeviceInfo::HD5XXX">;
// Predicate that is true if we have region address space.
def hasRegionAS : Predicate<"Subtarget.device()"
"->usesHardware(AMDGPUDeviceInfo::RegionMem)">;
// Predicate that is false if we don't have region address space.
def noRegionAS : Predicate<"!Subtarget.device()"
"->isSupported(AMDGPUDeviceInfo::RegionMem)">;
// Predicate that is set to true if 64bit Mul is supported in the IL or not
def HasHW64Mul : Predicate<"Subtarget.calVersion()"
">= CAL_VERSION_SC_139"
"&& Subtarget.device()"
"->getGeneration() >="
"AMDGPUDeviceInfo::HD5XXX">;
def HasSW64Mul : Predicate<"Subtarget.calVersion()"
"< CAL_VERSION_SC_139">;
// Predicate that is set to true if 64bit Div/Mod is supported in the IL or not
def HasHW64DivMod : Predicate<"Subtarget.device()"
"->usesHardware(AMDGPUDeviceInfo::HW64BitDivMod)">;
def HasSW64DivMod : Predicate<"Subtarget.device()"
"->usesSoftware(AMDGPUDeviceInfo::HW64BitDivMod)">;
// Predicate that is set to true if 64bit pointer are used.
def Has64BitPtr : Predicate<"Subtarget.is64bit()">;
def Has32BitPtr : Predicate<"!Subtarget.is64bit()">;
//===--------------------------------------------------------------------===//
// Custom Operands
//===--------------------------------------------------------------------===//
def brtarget : Operand<OtherVT>;
//===--------------------------------------------------------------------===//
// Custom Selection DAG Type Profiles
//===--------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Generic Profile Types
//===----------------------------------------------------------------------===//
def SDTIL_GenBinaryOp : SDTypeProfile<1, 2, [
SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
]>;
def SDTIL_GenTernaryOp : SDTypeProfile<1, 3, [
SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisSameAs<2, 3>
]>;
def SDTIL_GenVecBuild : SDTypeProfile<1, 1, [
SDTCisEltOfVec<1, 0>
]>;
//===----------------------------------------------------------------------===//
// Flow Control Profile Types
//===----------------------------------------------------------------------===//
// Branch instruction where second and third are basic blocks
def SDTIL_BRCond : SDTypeProfile<0, 2, [
SDTCisVT<0, OtherVT>
]>;
//===--------------------------------------------------------------------===//
// Custom Selection DAG Nodes
//===--------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Flow Control DAG Nodes
//===----------------------------------------------------------------------===//
def IL_brcond : SDNode<"AMDGPUISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
//===----------------------------------------------------------------------===//
// Call/Return DAG Nodes
//===----------------------------------------------------------------------===//
def IL_retflag : SDNode<"AMDGPUISD::RET_FLAG", SDTNone,
[SDNPHasChain, SDNPOptInGlue]>;
//===--------------------------------------------------------------------===//
// Instructions
//===--------------------------------------------------------------------===//
// Floating point math functions
def IL_div_inf : SDNode<"AMDGPUISD::DIV_INF", SDTIL_GenBinaryOp>;
def IL_mad : SDNode<"AMDGPUISD::MAD", SDTIL_GenTernaryOp>;
//===----------------------------------------------------------------------===//
// Integer functions
//===----------------------------------------------------------------------===//
def IL_umul : SDNode<"AMDGPUISD::UMUL" , SDTIntBinOp,
[SDNPCommutative, SDNPAssociative]>;
//===--------------------------------------------------------------------===//
// Custom Pattern DAG Nodes
//===--------------------------------------------------------------------===//
def global_store : PatFrag<(ops node:$val, node:$ptr),
(store node:$val, node:$ptr), [{
return isGlobalStore(dyn_cast<StoreSDNode>(N));
}]>;
//===----------------------------------------------------------------------===//
// Load pattern fragments
//===----------------------------------------------------------------------===//
// Global address space loads
def global_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return isGlobalLoad(dyn_cast<LoadSDNode>(N));
}]>;
// Constant address space loads
def constant_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
return isConstantLoad(dyn_cast<LoadSDNode>(N), -1);
}]>;
//===----------------------------------------------------------------------===//
// Complex addressing mode patterns
//===----------------------------------------------------------------------===//
def ADDR : ComplexPattern<i32, 2, "SelectADDR", [], []>;
def ADDRF : ComplexPattern<i32, 2, "SelectADDR", [frameindex], []>;
def ADDR64 : ComplexPattern<i64, 2, "SelectADDR64", [], []>;
def ADDR64F : ComplexPattern<i64, 2, "SelectADDR64", [frameindex], []>;
//===----------------------------------------------------------------------===//
// Instruction format classes
//===----------------------------------------------------------------------===//
class ILFormat<dag outs, dag ins, string asmstr, list<dag> pattern>
: Instruction {
let Namespace = "AMDGPU";
dag OutOperandList = outs;
dag InOperandList = ins;
let Pattern = pattern;
let AsmString = !strconcat(asmstr, "\n");
let isPseudo = 1;
let Itinerary = NullALU;
bit hasIEEEFlag = 0;
bit hasZeroOpFlag = 0;
let mayLoad = 0;
let mayStore = 0;
let hasSideEffects = 0;
}
//===--------------------------------------------------------------------===//
// Multiclass Instruction formats
//===--------------------------------------------------------------------===//
// Multiclass that handles branch instructions
multiclass BranchConditional<SDNode Op> {
def _i32 : ILFormat<(outs),
(ins brtarget:$target, GPRI32:$src0),
"; i32 Pseudo branch instruction",
[(Op bb:$target, GPRI32:$src0)]>;
def _f32 : ILFormat<(outs),
(ins brtarget:$target, GPRF32:$src0),
"; f32 Pseudo branch instruction",
[(Op bb:$target, GPRF32:$src0)]>;
}
// Only scalar types should generate flow control
multiclass BranchInstr<string name> {
def _i32 : ILFormat<(outs), (ins GPRI32:$src),
!strconcat(name, " $src"), []>;
def _f32 : ILFormat<(outs), (ins GPRF32:$src),
!strconcat(name, " $src"), []>;
}
// Only scalar types should generate flow control
multiclass BranchInstr2<string name> {
def _i32 : ILFormat<(outs), (ins GPRI32:$src0, GPRI32:$src1),
!strconcat(name, " $src0, $src1"), []>;
def _f32 : ILFormat<(outs), (ins GPRF32:$src0, GPRF32:$src1),
!strconcat(name, " $src0, $src1"), []>;
}
//===--------------------------------------------------------------------===//
// Intrinsics support
//===--------------------------------------------------------------------===//
include "AMDILIntrinsics.td"