[llvm] 5b18ef7 - [AMDGPU] Add verification for mandatory literals
Jay Foad via llvm-commits
llvm-commits at lists.llvm.org
Fri May 20 08:16:52 PDT 2022
Author: Jay Foad
Date: 2022-05-20T16:14:00+01:00
New Revision: 5b18ef7256a129273b6786545f8b66e928442e81
URL: https://github.com/llvm/llvm-project/commit/5b18ef7256a129273b6786545f8b66e928442e81
DIFF: https://github.com/llvm/llvm-project/commit/5b18ef7256a129273b6786545f8b66e928442e81.diff
LOG: [AMDGPU] Add verification for mandatory literals
Extend the literal operand checking in SIInstrInfo::verifyInstruction to
check VOP2 instructions like V_FMAAK_F32 which have a mandatory literal
operand. The rule is that src0 can also be a literal, but only if it is
the same literal value.
AMDGPUAsmParser::validateConstantBusLimitations already handles this
correctly.
Differential Revision: https://reviews.llvm.org/D126063
Added:
Modified:
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
llvm/test/CodeGen/AMDGPU/fold-fmaak-bug.ll
llvm/test/CodeGen/AMDGPU/verify-constant-bus-violations.mir
llvm/test/CodeGen/AMDGPU/verify-duplicate-literal.mir
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index cb1a4a269c946..0b0f2c4a64b19 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -4252,8 +4252,12 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr &MI,
bool UsesLiteral = false;
const MachineOperand *LiteralVal = nullptr;
- if (AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::imm) != -1)
+ int ImmIdx = AMDGPU::getNamedOperandIdx(Opcode, AMDGPU::OpName::imm);
+ if (ImmIdx != -1) {
++ConstantBusCount;
+ UsesLiteral = true;
+ LiteralVal = &MI.getOperand(ImmIdx);
+ }
SmallVector<Register, 2> SGPRsUsed;
Register SGPRUsed;
@@ -4280,8 +4284,8 @@ bool SIInstrInfo::verifyInstruction(const MachineInstr &MI,
UsesLiteral = true;
LiteralVal = &MO;
} else if (!MO.isIdenticalTo(*LiteralVal)) {
- assert(isVOP3(MI));
- ErrInfo = "VOP3 instruction uses more than one literal";
+ assert(isVOP2(MI) || isVOP3(MI));
+ ErrInfo = "VOP2/VOP3 instruction uses more than one literal";
return false;
}
}
diff --git a/llvm/test/CodeGen/AMDGPU/fold-fmaak-bug.ll b/llvm/test/CodeGen/AMDGPU/fold-fmaak-bug.ll
index bfa6f0c2598f6..8ac8b6b427989 100644
--- a/llvm/test/CodeGen/AMDGPU/fold-fmaak-bug.ll
+++ b/llvm/test/CodeGen/AMDGPU/fold-fmaak-bug.ll
@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck %s -check-prefix GFX10
+; RUN: not --crash llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s 2>&1 | FileCheck %s -check-prefix GFX10
declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float)
declare void @llvm.amdgcn.exp.compr.v2f16(i32 immarg, i32 immarg, <2 x half>, <2 x half>, i1 immarg, i1 immarg)
@@ -6,7 +6,8 @@ declare void @llvm.amdgcn.exp.compr.v2f16(i32 immarg, i32 immarg, <2 x half>, <2
; FIXME: This instruction uses two
diff erent literal constants which is not
; allowed.
; GFX10-LABEL: _amdgpu_ps_main:
-; GFX10: v_fmaak_f32 {{v[0-9]+}}, 0x40490fdb, {{v[0-9]+}}, 0xbfc90fdb
+; GFX10: Bad machine code: VOP2/VOP3 instruction uses more than one literal
+; GFX10: instruction: %4:vgpr_32 = nnan nsz arcp contract afn reassoc nofpexcept V_FMAAK_F32 1078530011, %0:vgpr_32, -1077342245, implicit $mode, implicit $exec
define amdgpu_ps void @_amdgpu_ps_main(float %arg) {
bb:
%i = fmul reassoc nnan nsz arcp contract afn float %arg, 0x400921FB60000000
diff --git a/llvm/test/CodeGen/AMDGPU/verify-constant-bus-violations.mir b/llvm/test/CodeGen/AMDGPU/verify-constant-bus-violations.mir
index a036bf5c4807f..a9e34c0279491 100644
--- a/llvm/test/CodeGen/AMDGPU/verify-constant-bus-violations.mir
+++ b/llvm/test/CodeGen/AMDGPU/verify-constant-bus-violations.mir
@@ -1,21 +1,26 @@
-# RUN: not --crash llc -march=amdgcn -mcpu=gfx900 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -check-prefix=GFX9-ERR %s
-# RUN: not --crash llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -check-prefix=GFX10PLUS-ERR %s
-# RUN: not --crash llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -check-prefix=GFX10PLUS-ERR %s
+# RUN: not --crash llc -march=amdgcn -mcpu=gfx900 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -implicit-check-not="Bad machine code" -check-prefix=GFX9-ERR %s
+# RUN: not --crash llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -implicit-check-not="Bad machine code" -check-prefix=GFX10PLUS-ERR %s
+# RUN: not --crash llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -run-pass machineverifier -o /dev/null %s 2>&1 | FileCheck -implicit-check-not="Bad machine code" -check-prefix=GFX10PLUS-ERR %s
# GFX9-ERR: *** Bad machine code: VOP* instruction violates constant bus restriction ***
# GFX9-ERR: $vgpr0 = V_CNDMASK_B32_e64 0, $sgpr0, 0, -1, killed $sgpr0_sgpr1, implicit $exec
+# GFX9-ERR: *** Bad machine code: VOP* instruction violates constant bus restriction ***
+# GFX9-ERR: $vgpr0 = V_FMAAK_F32 $sgpr2, $vgpr0, 1077936128, implicit $mode, implicit $exec
---
name: sgpr_reuse_2sgpr
liveins:
- { reg: '$sgpr0_sgpr1', virtual-reg: '' }
body: |
bb.0:
- liveins: $sgpr0_sgpr1
+ liveins: $sgpr0_sgpr1, $sgpr2
$vgpr0 = V_CNDMASK_B32_e64 0, $sgpr0, 0, -1, killed $sgpr0_sgpr1, implicit $exec
+ $vgpr0 = V_FMAAK_F32 $sgpr2, $vgpr0, 1077936128, implicit $mode, implicit $exec
...
# GFX10PLUS-ERR: *** Bad machine code: VOP* instruction violates constant bus restriction ***
# GFX10PLUS-ERR: $vgpr0 = V_CNDMASK_B32_e64 0, $sgpr0, 0, $sgpr2, killed $sgpr0_sgpr1, implicit $exec
+# GFX10PLUS-ERR: *** Bad machine code: VOP2/VOP3 instruction uses more than one literal ***
+# GFX10PLUS-ERR: $vgpr0 = V_FMAAK_F32 1077936128, $vgpr0, 0, implicit $mode, implicit $exec
---
name: sgpr_reuse_3sgpr
liveins:
@@ -25,4 +30,6 @@ body: |
bb.0:
liveins: $sgpr0_sgpr1, $sgpr2
$vgpr0 = V_CNDMASK_B32_e64 0, $sgpr0, 0, $sgpr2, killed $sgpr0_sgpr1, implicit $exec
+ $vgpr0 = V_FMAAK_F32 1077936128, $vgpr0, 0, implicit $mode, implicit $exec
+ $vgpr0 = V_FMAAK_F32 $sgpr2, $vgpr0, 1077936128, implicit $mode, implicit $exec
...
diff --git a/llvm/test/CodeGen/AMDGPU/verify-duplicate-literal.mir b/llvm/test/CodeGen/AMDGPU/verify-duplicate-literal.mir
index a5920d3cac07c..ba303fe8f8c85 100644
--- a/llvm/test/CodeGen/AMDGPU/verify-duplicate-literal.mir
+++ b/llvm/test/CodeGen/AMDGPU/verify-duplicate-literal.mir
@@ -27,6 +27,15 @@ body: |
$vgpr0 = V_FMA_F32_e64 0, $vgpr0, 0, 1077936128, 0, 1077936128, 0, 0, implicit $mode, implicit $exec
...
+---
+name: use_duplicate_literal_fmaak
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0
+ $vgpr0 = V_FMAAK_F32 1077936128, $vgpr0, 1077936128, implicit $mode, implicit $exec
+...
+
---
name: use_duplicate_literal_sop2
tracksRegLiveness: true
More information about the llvm-commits
mailing list