diff --git a/programming/build/llvm/files/amdgpu-avoid-an-illegal-operand-in-si-shrink-instr.patch b/programming/build/llvm/files/amdgpu-avoid-an-illegal-operand-in-si-shrink-instr.patch new file mode 100644 index 0000000000..f96f73262a --- /dev/null +++ b/programming/build/llvm/files/amdgpu-avoid-an-illegal-operand-in-si-shrink-instr.patch @@ -0,0 +1,85 @@ +commit b08a140a8fe8d0b0d16a93042b4952d6e34ab913 +Author: Piotr Sobczak +Date: Wed Jan 27 16:02:49 2021 +0100 + + [AMDGPU] Avoid an illegal operand in si-shrink-instructions + + Before the patch it was possible to trigger a constant bus + violation when folding immediates into a shrunk instruction. + + The patch adds a check to enforce the legality of the new operand. + + Differential Revision: https://reviews.llvm.org/D95527 + +diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp +index 9c6833a7dab6..6c1b16eddc84 100644 +--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp ++++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp +@@ -84,21 +84,23 @@ static bool foldImmediates(MachineInstr &MI, const SIInstrInfo *TII, + MachineOperand &MovSrc = Def->getOperand(1); + bool ConstantFolded = false; + +- if (MovSrc.isImm() && (isInt<32>(MovSrc.getImm()) || +- isUInt<32>(MovSrc.getImm()))) { +- // It's possible to have only one component of a super-reg defined by +- // a single mov, so we need to clear any subregister flag. +- Src0.setSubReg(0); +- Src0.ChangeToImmediate(MovSrc.getImm()); +- ConstantFolded = true; +- } else if (MovSrc.isFI()) { +- Src0.setSubReg(0); +- Src0.ChangeToFrameIndex(MovSrc.getIndex()); +- ConstantFolded = true; +- } else if (MovSrc.isGlobal()) { +- Src0.ChangeToGA(MovSrc.getGlobal(), MovSrc.getOffset(), +- MovSrc.getTargetFlags()); +- ConstantFolded = true; ++ if (TII->isOperandLegal(MI, Src0Idx, &MovSrc)) { ++ if (MovSrc.isImm() && ++ (isInt<32>(MovSrc.getImm()) || isUInt<32>(MovSrc.getImm()))) { ++ // It's possible to have only one component of a super-reg defined ++ // by a single mov, so we need to clear any subregister flag. ++ Src0.setSubReg(0); ++ Src0.ChangeToImmediate(MovSrc.getImm()); ++ ConstantFolded = true; ++ } else if (MovSrc.isFI()) { ++ Src0.setSubReg(0); ++ Src0.ChangeToFrameIndex(MovSrc.getIndex()); ++ ConstantFolded = true; ++ } else if (MovSrc.isGlobal()) { ++ Src0.ChangeToGA(MovSrc.getGlobal(), MovSrc.getOffset(), ++ MovSrc.getTargetFlags()); ++ ConstantFolded = true; ++ } + } + + if (ConstantFolded) { +diff --git a/llvm/test/CodeGen/AMDGPU/shrink-instructions-illegal-fold.mir b/llvm/test/CodeGen/AMDGPU/shrink-instructions-illegal-fold.mir +new file mode 100644 +index 000000000000..7889f437facf +--- /dev/null ++++ b/llvm/test/CodeGen/AMDGPU/shrink-instructions-illegal-fold.mir +@@ -0,0 +1,23 @@ ++# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=si-shrink-instructions --verify-machineinstrs %s -o - | FileCheck %s ++ ++# Make sure immediate folding into V_CNDMASK respects constant bus restrictions. ++--- ++ ++name: shrink_cndmask_illegal_imm_folding ++tracksRegLiveness: true ++body: | ++ bb.0: ++ liveins: $vgpr0, $vgpr1 ++ ; CHECK-LABEL: name: shrink_cndmask_illegal_imm_folding ++ ; CHECK: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0 ++ ; CHECK: [[MOV:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 32768, implicit $exec ++ ; CHECK: V_CMP_EQ_U32_e32 0, [[COPY]], implicit-def $vcc, implicit $exec ++ ; CHECK: V_CNDMASK_B32_e32 [[MOV]], killed [[COPY]], implicit $vcc, implicit $exec ++ ++ %0:vgpr_32 = COPY $vgpr0 ++ %1:vgpr_32 = V_MOV_B32_e32 32768, implicit $exec ++ V_CMP_EQ_U32_e32 0, %0:vgpr_32, implicit-def $vcc, implicit $exec ++ %2:vgpr_32 = V_CNDMASK_B32_e64 0, %1:vgpr_32, 0, killed %0:vgpr_32, $vcc, implicit $exec ++ S_NOP 0 ++ ++... diff --git a/programming/build/llvm/pspec.xml b/programming/build/llvm/pspec.xml index 76c44cced9..0cb7fb35c9 100755 --- a/programming/build/llvm/pspec.xml +++ b/programming/build/llvm/pspec.xml @@ -37,7 +37,7 @@ --> 0001-SystemZ-Use-LA-instead-of-AGR-in-eliminateFrameIndex.patch - + amdgpu-avoid-an-illegal-operand-in-si-shrink-instr.patch @@ -298,6 +298,7 @@ 32-bit shared libraries for llvm emul32 + zlib-32bit libffi-32bit libxml2-32bit