[llvm] [RFC][GlobalISel] Use Builders in MatchTable (PR #65955)
Pierre van Houtryve via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 11 05:45:52 PDT 2023
https://github.com/Pierre-vh created https://github.com/llvm/llvm-project/pull/65955:
The MatchTableExecutor did not use the MachineIRBuilder nor Observers to build instructions.
This meant that it only had a very superficial view of what's going on when rules are being applied. Custom code could create insts that the executor didn't know about.
Using a builder & an observer allows it to collect any instruction that's created as long as the same builder is used by the supporting C++ code. For instance, flags are propagated more thoroughly now.
Another benefit that may come later is that I'm trying to improve how constants are created in apply MIR patterns.
`MachineIRBuilder::buildConstant` automatically handles splats for us, this means that we may change `addCImm` to use that and handle vector cases automatically.
Overall it improves the flexibility and consistency of the builder
>From 07f0d51cdc5776612fd6370dc3e9ef35fdea09d7 Mon Sep 17 00:00:00 2001
From: pvanhout <pierre.vanhoutryve at amd.com>
Date: Mon, 11 Sep 2023 14:27:59 +0200
Subject: [PATCH 1/2] [AMDGPU] Regen combine-fma-add-mul-pre-legalize.mir
---
.../combine-fma-add-mul-pre-legalize.mir | 84 +++++++++++++++++++
1 file changed, 84 insertions(+)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
index d8866561f181fef..ee775dc674cfecd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
@@ -24,6 +24,7 @@ body: |
; GFX9-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
; GFX9-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-CONTRACT-LABEL: name: test_f32_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -33,6 +34,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-DENORM-LABEL: name: test_f32_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-DENORM-NEXT: {{ $}}
@@ -43,6 +45,7 @@ body: |
; GFX9-DENORM-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-UNSAFE-LABEL: name: test_f32_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -52,6 +55,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-LABEL: name: test_f32_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
@@ -62,6 +66,7 @@ body: |
; GFX10-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
; GFX10-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-CONTRACT-LABEL: name: test_f32_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -71,6 +76,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-DENORM-LABEL: name: test_f32_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-DENORM-NEXT: {{ $}}
@@ -81,6 +87,7 @@ body: |
; GFX10-DENORM-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[FMUL]], [[COPY2]]
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-UNSAFE-LABEL: name: test_f32_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -115,6 +122,7 @@ body: |
; GFX9-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
; GFX9-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-CONTRACT-LABEL: name: test_f32_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -124,6 +132,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-DENORM-LABEL: name: test_f32_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-DENORM-NEXT: {{ $}}
@@ -134,6 +143,7 @@ body: |
; GFX9-DENORM-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-UNSAFE-LABEL: name: test_f32_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -143,6 +153,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-LABEL: name: test_f32_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
@@ -153,6 +164,7 @@ body: |
; GFX10-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
; GFX10-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-CONTRACT-LABEL: name: test_f32_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -162,6 +174,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-DENORM-LABEL: name: test_f32_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-DENORM-NEXT: {{ $}}
@@ -172,6 +185,7 @@ body: |
; GFX10-DENORM-NEXT: [[FADD:%[0-9]+]]:_(s32) = reassoc G_FADD [[COPY2]], [[FMUL]]
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[FADD]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-UNSAFE-LABEL: name: test_f32_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -210,6 +224,7 @@ body: |
; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-CONTRACT-LABEL: name: test_half_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -223,6 +238,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-DENORM-LABEL: name: test_half_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-DENORM-NEXT: {{ $}}
@@ -237,6 +253,7 @@ body: |
; GFX9-DENORM-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-UNSAFE-LABEL: name: test_half_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -250,6 +267,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-LABEL: name: test_half_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
@@ -264,6 +282,7 @@ body: |
; GFX10-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX10-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-CONTRACT-LABEL: name: test_half_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -277,6 +296,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-DENORM-LABEL: name: test_half_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-DENORM-NEXT: {{ $}}
@@ -291,6 +311,7 @@ body: |
; GFX10-DENORM-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-UNSAFE-LABEL: name: test_half_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -337,6 +358,7 @@ body: |
; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-CONTRACT-LABEL: name: test_half_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -350,6 +372,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-DENORM-LABEL: name: test_half_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-DENORM-NEXT: {{ $}}
@@ -364,6 +387,7 @@ body: |
; GFX9-DENORM-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX9-UNSAFE-LABEL: name: test_half_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -377,6 +401,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-LABEL: name: test_half_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-NEXT: {{ $}}
@@ -391,6 +416,7 @@ body: |
; GFX10-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX10-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-CONTRACT-LABEL: name: test_half_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -404,6 +430,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-DENORM-LABEL: name: test_half_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-DENORM-NEXT: {{ $}}
@@ -418,6 +445,7 @@ body: |
; GFX10-DENORM-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16)
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
+ ;
; GFX10-UNSAFE-LABEL: name: test_half_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -468,6 +496,7 @@ body: |
; GFX9-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-CONTRACT-LABEL: name: test_double_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -485,6 +514,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-DENORM-LABEL: name: test_double_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-DENORM-NEXT: {{ $}}
@@ -503,6 +533,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-UNSAFE-LABEL: name: test_double_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -520,6 +551,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-LABEL: name: test_double_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-NEXT: {{ $}}
@@ -538,6 +570,7 @@ body: |
; GFX10-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-CONTRACT-LABEL: name: test_double_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -555,6 +588,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-DENORM-LABEL: name: test_double_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-DENORM-NEXT: {{ $}}
@@ -573,6 +607,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-UNSAFE-LABEL: name: test_double_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -631,6 +666,7 @@ body: |
; GFX9-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-CONTRACT-LABEL: name: test_double_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -648,6 +684,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-DENORM-LABEL: name: test_double_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-DENORM-NEXT: {{ $}}
@@ -666,6 +703,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-UNSAFE-LABEL: name: test_double_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -683,6 +721,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-LABEL: name: test_double_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-NEXT: {{ $}}
@@ -701,6 +740,7 @@ body: |
; GFX10-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-CONTRACT-LABEL: name: test_double_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -718,6 +758,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-DENORM-LABEL: name: test_double_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-DENORM-NEXT: {{ $}}
@@ -736,6 +777,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-UNSAFE-LABEL: name: test_double_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -802,6 +844,7 @@ body: |
; GFX9-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX9-CONTRACT-LABEL: name: test_4xfloat_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -827,6 +870,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-CONTRACT-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX9-DENORM-LABEL: name: test_4xfloat_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX9-DENORM-NEXT: {{ $}}
@@ -853,6 +897,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-DENORM-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX9-UNSAFE-LABEL: name: test_4xfloat_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -878,6 +923,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-UNSAFE-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX10-LABEL: name: test_4xfloat_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX10-NEXT: {{ $}}
@@ -904,6 +950,7 @@ body: |
; GFX10-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX10-CONTRACT-LABEL: name: test_4xfloat_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -929,6 +976,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-CONTRACT-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX10-DENORM-LABEL: name: test_4xfloat_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX10-DENORM-NEXT: {{ $}}
@@ -955,6 +1003,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-DENORM-NEXT: $vgpr3 = COPY [[UV3]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+ ;
; GFX10-UNSAFE-LABEL: name: test_4xfloat_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -1033,6 +1082,7 @@ body: |
; GFX9-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX9-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -1054,6 +1104,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-CONTRACT-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX9-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX9-DENORM-NEXT: {{ $}}
@@ -1076,6 +1127,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-DENORM-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX9-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -1097,6 +1149,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX9-UNSAFE-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX10-LABEL: name: test_3xfloat_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX10-NEXT: {{ $}}
@@ -1119,6 +1172,7 @@ body: |
; GFX10-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX10-CONTRACT-LABEL: name: test_3xfloat_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -1140,6 +1194,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-CONTRACT-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX10-DENORM-LABEL: name: test_3xfloat_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX10-DENORM-NEXT: {{ $}}
@@ -1162,6 +1217,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr1 = COPY [[UV1]](s32)
; GFX10-DENORM-NEXT: $vgpr2 = COPY [[UV2]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+ ;
; GFX10-UNSAFE-LABEL: name: test_3xfloat_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -1228,6 +1284,7 @@ body: |
; GFX9-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-CONTRACT-LABEL: name: test_4xhalf_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -1245,6 +1302,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-DENORM-LABEL: name: test_4xhalf_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-DENORM-NEXT: {{ $}}
@@ -1263,6 +1321,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-DENORM-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-UNSAFE-LABEL: name: test_4xhalf_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -1280,6 +1339,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-LABEL: name: test_4xhalf_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-NEXT: {{ $}}
@@ -1298,6 +1358,7 @@ body: |
; GFX10-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-CONTRACT-LABEL: name: test_4xhalf_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -1315,6 +1376,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-DENORM-LABEL: name: test_4xhalf_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-DENORM-NEXT: {{ $}}
@@ -1333,6 +1395,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-DENORM-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-UNSAFE-LABEL: name: test_4xhalf_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -1403,6 +1466,7 @@ body: |
; GFX9-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -1432,6 +1496,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-DENORM-NEXT: {{ $}}
@@ -1462,6 +1527,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-DENORM-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX9-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -1491,6 +1557,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-LABEL: name: test_3xhalf_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-NEXT: {{ $}}
@@ -1521,6 +1588,7 @@ body: |
; GFX10-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-CONTRACT-LABEL: name: test_3xhalf_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -1550,6 +1618,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-DENORM-LABEL: name: test_3xhalf_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-DENORM-NEXT: {{ $}}
@@ -1580,6 +1649,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-DENORM-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1
+ ;
; GFX10-UNSAFE-LABEL: name: test_3xhalf_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -1692,6 +1762,7 @@ body: |
; GFX9-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX9-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX9-CONTRACT-LABEL: name: test_4xdouble_add_mul
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -1745,6 +1816,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX9-CONTRACT-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX9-DENORM-LABEL: name: test_4xdouble_add_mul
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX9-DENORM-NEXT: {{ $}}
@@ -1799,6 +1871,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX9-DENORM-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX9-UNSAFE-LABEL: name: test_4xdouble_add_mul
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -1852,6 +1925,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX9-UNSAFE-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX10-LABEL: name: test_4xdouble_add_mul
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX10-NEXT: {{ $}}
@@ -1906,6 +1980,7 @@ body: |
; GFX10-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX10-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX10-CONTRACT-LABEL: name: test_4xdouble_add_mul
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -1959,6 +2034,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX10-CONTRACT-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX10-DENORM-LABEL: name: test_4xdouble_add_mul
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX10-DENORM-NEXT: {{ $}}
@@ -2013,6 +2089,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr6 = COPY [[UV6]](s32)
; GFX10-DENORM-NEXT: $vgpr7 = COPY [[UV7]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
+ ;
; GFX10-UNSAFE-LABEL: name: test_4xdouble_add_mul
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23
; GFX10-UNSAFE-NEXT: {{ $}}
@@ -2168,6 +2245,7 @@ body: |
; GFX9-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX9-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX9-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX9-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
; GFX9-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX9-CONTRACT-NEXT: {{ $}}
@@ -2210,6 +2288,7 @@ body: |
; GFX9-CONTRACT-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX9-CONTRACT-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX9-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
; GFX9-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX9-DENORM-NEXT: {{ $}}
@@ -2253,6 +2332,7 @@ body: |
; GFX9-DENORM-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX9-DENORM-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX9-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX9-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
; GFX9-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX9-UNSAFE-NEXT: {{ $}}
@@ -2295,6 +2375,7 @@ body: |
; GFX9-UNSAFE-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX9-UNSAFE-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX10-LABEL: name: test_3xdouble_add_mul_rhs
; GFX10: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX10-NEXT: {{ $}}
@@ -2338,6 +2419,7 @@ body: |
; GFX10-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX10-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX10-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX10-CONTRACT-LABEL: name: test_3xdouble_add_mul_rhs
; GFX10-CONTRACT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX10-CONTRACT-NEXT: {{ $}}
@@ -2380,6 +2462,7 @@ body: |
; GFX10-CONTRACT-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX10-CONTRACT-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX10-DENORM-LABEL: name: test_3xdouble_add_mul_rhs
; GFX10-DENORM: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX10-DENORM-NEXT: {{ $}}
@@ -2423,6 +2506,7 @@ body: |
; GFX10-DENORM-NEXT: $vgpr4 = COPY [[UV4]](s32)
; GFX10-DENORM-NEXT: $vgpr5 = COPY [[UV5]](s32)
; GFX10-DENORM-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5
+ ;
; GFX10-UNSAFE-LABEL: name: test_3xdouble_add_mul_rhs
; GFX10-UNSAFE: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17
; GFX10-UNSAFE-NEXT: {{ $}}
>From 4cb4b58faa5f64dfec71f7bc05d6274ee48ca65e Mon Sep 17 00:00:00 2001
From: pvanhout <pierre.vanhoutryve at amd.com>
Date: Mon, 11 Sep 2023 14:43:47 +0200
Subject: [PATCH 2/2] [RFC][GlobalISel] Use Builders in MatchTable
The MatchTableExecutor did not use the MachineIRBuilder nor Observers to build instructions.
This meant that it only had a very superficial view of what's going on when rules are being applied. Custom code could create insts that the executor didn't know about.
Using a builder & an observer allows it to collect any instruction that's created as long as the same builder is used by the supporting C++ code. For instance, flags are propagated more thoroughly now.
Another benefit that may come later is that I'm trying to improve how constants are created in apply MIR patterns.
`MachineIRBuilder::buildConstant` automatically handles splats for us,
this means that we may change `addCImm` to use that and handle vector cases automatically.
---
.../CodeGen/GlobalISel/GIMatchTableExecutor.h | 40 +++++---
.../GlobalISel/GIMatchTableExecutorImpl.h | 61 +++++++-----
.../CodeGen/GlobalISel/MachineIRBuilder.h | 10 ++
.../GlobalISel/GIMatchTableExecutor.cpp | 4 +
.../AArch64/GlobalISel/combine-sdiv.mir | 28 +++---
...galizercombiner-ashr-shl-to-sext-inreg.mir | 4 +-
.../combine-fma-add-mul-pre-legalize.mir | 96 +++++++++----------
.../GlobalISel/combine-fma-unmerge-values.mir | 20 ++--
.../combine-fold-binop-into-select.mir | 20 ++--
.../AMDGPU/GlobalISel/combine-fsub-fneg.mir | 28 +++---
.../GlobalISelCombinerEmitter/match-table.td | 6 +-
llvm/test/TableGen/GlobalISelEmitter.td | 6 +-
.../TableGen/GlobalISelCombinerEmitter.cpp | 8 +-
llvm/utils/TableGen/GlobalISelEmitter.cpp | 6 +-
.../GlobalISelMatchTableExecutorEmitter.cpp | 2 +-
15 files changed, 190 insertions(+), 149 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h b/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h
index 2b0733cf9353e6c..cc7ebd130ce23f7 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h
@@ -19,6 +19,7 @@
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/GlobalISel/Utils.h"
+#include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
#include "llvm/CodeGen/LowLevelType.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/IR/Function.h"
@@ -40,6 +41,7 @@ class APInt;
class APFloat;
class GISelKnownBits;
class MachineInstr;
+class MachineIRBuilder;
class MachineInstrBuilder;
class MachineFunction;
class MachineOperand;
@@ -501,10 +503,24 @@ class GIMatchTableExecutor {
}
protected:
+ /// Observer used by \ref executeMatchTable to record all instructions created
+ /// by the rule.
+ class GIMatchTableObserver : public GISelChangeObserver {
+ public:
+ virtual ~GIMatchTableObserver();
+
+ void erasingInstr(MachineInstr &MI) override { CreatedInsts.erase(&MI);}
+ void createdInstr(MachineInstr &MI) override { CreatedInsts.insert(&MI); }
+ void changingInstr(MachineInstr &MI) override {}
+ void changedInstr(MachineInstr &MI) override {}
+
+ // Keeps track of all instructions that have been created when applying a rule.
+ SmallDenseSet<MachineInstr *, 4> CreatedInsts;
+ };
+
using ComplexRendererFns =
std::optional<SmallVector<std::function<void(MachineInstrBuilder &)>, 4>>;
using RecordedMIVector = SmallVector<MachineInstr *, 4>;
- using NewMIVector = SmallVector<MachineInstrBuilder, 4>;
struct MatcherState {
std::vector<ComplexRendererFns::value_type> Renderers;
@@ -555,15 +571,17 @@ class GIMatchTableExecutor {
/// and false otherwise.
template <class TgtExecutor, class PredicateBitset, class ComplexMatcherMemFn,
class CustomRendererFn>
- bool executeMatchTable(
- TgtExecutor &Exec, NewMIVector &OutMIs, MatcherState &State,
- const ExecInfoTy<PredicateBitset, ComplexMatcherMemFn, CustomRendererFn>
- &ISelInfo,
- const int64_t *MatchTable, const TargetInstrInfo &TII,
- MachineRegisterInfo &MRI, const TargetRegisterInfo &TRI,
- const RegisterBankInfo &RBI, const PredicateBitset &AvailableFeatures,
- CodeGenCoverage *CoverageInfo,
- GISelChangeObserver *Observer = nullptr) const;
+ bool executeMatchTable(TgtExecutor &Exec, MatcherState &State,
+ const ExecInfoTy<PredicateBitset, ComplexMatcherMemFn,
+ CustomRendererFn> &ExecInfo,
+ MachineIRBuilder &Builder,
+ const int64_t *MatchTable, const TargetInstrInfo &TII,
+ MachineRegisterInfo &MRI,
+ const TargetRegisterInfo &TRI,
+ const RegisterBankInfo &RBI,
+ const PredicateBitset &AvailableFeatures,
+ CodeGenCoverage *CoverageInfo
+ ) const;
virtual const int64_t *getMatchTable() const {
llvm_unreachable("Should have been overridden by tablegen if used");
@@ -592,7 +610,7 @@ class GIMatchTableExecutor {
}
virtual void runCustomAction(unsigned, const MatcherState &State,
- NewMIVector &OutMIs) const {
+ ArrayRef<MachineInstrBuilder> OutMIs) const {
llvm_unreachable("Subclass does not implement runCustomAction!");
}
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h b/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h
index 883c1ca0fe350b0..5e2371cddffd79a 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/GIMatchTableExecutorImpl.h
@@ -18,6 +18,7 @@
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/GlobalISel/GIMatchTableExecutor.h"
#include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
+#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
#include "llvm/CodeGen/GlobalISel/Utils.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineOperand.h"
@@ -42,17 +43,32 @@ namespace llvm {
template <class TgtExecutor, class PredicateBitset, class ComplexMatcherMemFn,
class CustomRendererFn>
bool GIMatchTableExecutor::executeMatchTable(
- TgtExecutor &Exec, NewMIVector &OutMIs, MatcherState &State,
+ TgtExecutor &Exec, MatcherState &State,
const ExecInfoTy<PredicateBitset, ComplexMatcherMemFn, CustomRendererFn>
&ExecInfo,
- const int64_t *MatchTable, const TargetInstrInfo &TII,
- MachineRegisterInfo &MRI, const TargetRegisterInfo &TRI,
- const RegisterBankInfo &RBI, const PredicateBitset &AvailableFeatures,
- CodeGenCoverage *CoverageInfo, GISelChangeObserver *Observer) const {
+ MachineIRBuilder &Builder, const int64_t *MatchTable,
+ const TargetInstrInfo &TII, MachineRegisterInfo &MRI,
+ const TargetRegisterInfo &TRI, const RegisterBankInfo &RBI,
+ const PredicateBitset &AvailableFeatures,
+ CodeGenCoverage *CoverageInfo) const {
+
+ // Setup observer
+ GIMatchTableObserver MTObserver;
+ GISelObserverWrapper Observer(&MTObserver);
+ if (auto *CurObs = Builder.getChangeObserver())
+ Observer.addObserver(CurObs);
+
+ // TODO: Set MF delegate?
+
+ // Setup builder.
+ auto RestoreOldObserver = Builder.setTemporaryChangeObserver(Observer);
uint64_t CurrentIdx = 0;
SmallVector<uint64_t, 4> OnFailResumeAt;
+ // We also record MachineInstrs manually in this vector so opcodes can address them.
+ SmallVector<MachineInstrBuilder, 4> OutMIs;
+
// Bypass the flag check on the instruction, and only look at the MCInstrDesc.
bool NoFPException = !State.MIs[0]->getDesc().mayRaiseFPException();
@@ -71,14 +87,16 @@ bool GIMatchTableExecutor::executeMatchTable(
return RejectAndResume;
};
- auto propagateFlags = [=](NewMIVector &OutMIs) {
- for (auto MIB : OutMIs) {
+ auto propagateFlags = [&]() {
+ for (auto *MI : MTObserver.CreatedInsts) {
// Set the NoFPExcept flag when no original matched instruction could
// raise an FP exception, but the new instruction potentially might.
uint16_t MIBFlags = Flags;
- if (NoFPException && MIB->mayRaiseFPException())
+ if (NoFPException && MI->mayRaiseFPException())
MIBFlags |= MachineInstr::NoFPExcept;
- MIB.setMIFlags(MIBFlags);
+ Observer.changingInstr(*MI);
+ MI->setFlags(MIBFlags);
+ Observer.changedInstr(*MI);
}
return true;
@@ -901,6 +919,7 @@ bool GIMatchTableExecutor::executeMatchTable(
OutMIs[NewInsnID] = MachineInstrBuilder(*State.MIs[OldInsnID]->getMF(),
State.MIs[OldInsnID]);
OutMIs[NewInsnID]->setDesc(TII.get(NewOpcode));
+ MTObserver.CreatedInsts.insert(OutMIs[NewInsnID]);
DEBUG_WITH_TYPE(TgtExecutor::getName(),
dbgs() << CurrentIdx << ": GIR_MutateOpcode(OutMIs["
<< NewInsnID << "], MIs[" << OldInsnID << "], "
@@ -914,8 +933,7 @@ bool GIMatchTableExecutor::executeMatchTable(
if (NewInsnID >= OutMIs.size())
OutMIs.resize(NewInsnID + 1);
- OutMIs[NewInsnID] = BuildMI(*State.MIs[0]->getParent(), State.MIs[0],
- MIMetadata(*State.MIs[0]), TII.get(Opcode));
+ OutMIs[NewInsnID] = Builder.buildInstr(Opcode);
DEBUG_WITH_TYPE(TgtExecutor::getName(),
dbgs() << CurrentIdx << ": GIR_BuildMI(OutMIs["
<< NewInsnID << "], " << Opcode << ")\n");
@@ -1239,8 +1257,7 @@ bool GIMatchTableExecutor::executeMatchTable(
DEBUG_WITH_TYPE(TgtExecutor::getName(),
dbgs() << CurrentIdx << ": GIR_EraseFromParent(MIs["
<< InsnID << "])\n");
- if (Observer)
- Observer->erasingInstr(*MI);
+ Observer.erasingInstr(*MI);
MI->eraseFromParent();
break;
}
@@ -1269,11 +1286,9 @@ bool GIMatchTableExecutor::executeMatchTable(
Register Old = State.MIs[OldInsnID]->getOperand(OldOpIdx).getReg();
Register New = State.MIs[NewInsnID]->getOperand(NewOpIdx).getReg();
- if (Observer)
- Observer->changingAllUsesOfReg(MRI, Old);
+ Observer.changingAllUsesOfReg(MRI, Old);
MRI.replaceRegWith(Old, New);
- if (Observer)
- Observer->finishedChangingAllUsesOfReg();
+ Observer.finishedChangingAllUsesOfReg();
break;
}
case GIR_ReplaceRegWithTempReg: {
@@ -1288,11 +1303,9 @@ bool GIMatchTableExecutor::executeMatchTable(
Register Old = State.MIs[OldInsnID]->getOperand(OldOpIdx).getReg();
Register New = State.TempRegisters[TempRegID];
- if (Observer)
- Observer->changingAllUsesOfReg(MRI, Old);
+ Observer.changingAllUsesOfReg(MRI, Old);
MRI.replaceRegWith(Old, New);
- if (Observer)
- Observer->finishedChangingAllUsesOfReg();
+ Observer.finishedChangingAllUsesOfReg();
break;
}
case GIR_Coverage: {
@@ -1309,11 +1322,7 @@ bool GIMatchTableExecutor::executeMatchTable(
case GIR_Done:
DEBUG_WITH_TYPE(TgtExecutor::getName(),
dbgs() << CurrentIdx << ": GIR_Done\n");
- if (Observer) {
- for (MachineInstr *MI : OutMIs)
- Observer->createdInstr(*MI);
- }
- propagateFlags(OutMIs);
+ propagateFlags();
return true;
default:
llvm_unreachable("Unexpected command");
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h b/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
index e7db9547f03b694..3217c0ad01f857b 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/MachineIRBuilder.h
@@ -19,6 +19,7 @@
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/IR/DebugLoc.h"
+#include "llvm/Support/SaveAndRestore.h"
#include "llvm/IR/Module.h"
namespace llvm {
@@ -364,6 +365,15 @@ class MachineIRBuilder {
State.Observer = &Observer;
}
+ GISelChangeObserver* getChangeObserver() const {
+ return State.Observer;
+ }
+
+ // Replaces the change observer with \p Observer and returns an object that restores the old Observer on destruction.
+ SaveAndRestore<GISelChangeObserver*> setTemporaryChangeObserver(GISelChangeObserver &Observer) {
+ return SaveAndRestore<GISelChangeObserver*>(State.Observer, &Observer);
+ }
+
void stopObservingChanges() { State.Observer = nullptr; }
bool isObservingChanges() const { return State.Observer != nullptr; }
diff --git a/llvm/lib/CodeGen/GlobalISel/GIMatchTableExecutor.cpp b/llvm/lib/CodeGen/GlobalISel/GIMatchTableExecutor.cpp
index 26752369a7711a1..c1ababb8ed852d0 100644
--- a/llvm/lib/CodeGen/GlobalISel/GIMatchTableExecutor.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GIMatchTableExecutor.cpp
@@ -21,6 +21,10 @@
using namespace llvm;
+GIMatchTableExecutor::GIMatchTableObserver::~GIMatchTableObserver() {
+ // anchor
+}
+
GIMatchTableExecutor::MatcherState::MatcherState(unsigned MaxRenderers)
: Renderers(MaxRenderers) {}
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/combine-sdiv.mir b/llvm/test/CodeGen/AArch64/GlobalISel/combine-sdiv.mir
index e99ee84100a3991..13f9998952e6ed3 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/combine-sdiv.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/combine-sdiv.mir
@@ -22,10 +22,10 @@ body: |
; CHECK: liveins: $w0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 -991146299
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 3
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 -991146299
; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = exact G_ASHR [[COPY]], [[C]](s32)
- ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[ASHR]], [[C1]]
+ ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(s32) = exact G_MUL [[ASHR]], [[C1]]
; CHECK-NEXT: $w0 = COPY [[MUL]](s32)
; CHECK-NEXT: RET_ReallyLR implicit $w0
%0:_(s32) = COPY $w0
@@ -87,13 +87,13 @@ body: |
; CHECK: liveins: $q0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 -991146299
- ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 954437177
- ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[C]](s32), [[C]](s32), [[C]](s32), [[C]](s32)
- ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[C1]](s32), [[C2]](s32), [[C1]](s32), [[C2]](s32)
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 3
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 -991146299
+ ; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 954437177
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = exact G_BUILD_VECTOR [[C]](s32), [[C]](s32), [[C]](s32), [[C]](s32)
+ ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = exact G_BUILD_VECTOR [[C1]](s32), [[C2]](s32), [[C1]](s32), [[C2]](s32)
; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(<4 x s32>) = exact G_ASHR [[COPY]], [[BUILD_VECTOR]](<4 x s32>)
- ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(<4 x s32>) = G_MUL [[ASHR]], [[BUILD_VECTOR1]]
+ ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(<4 x s32>) = exact G_MUL [[ASHR]], [[BUILD_VECTOR1]]
; CHECK-NEXT: $q0 = COPY [[MUL]](<4 x s32>)
; CHECK-NEXT: RET_ReallyLR implicit $q0
%0:_(<4 x s32>) = COPY $q0
@@ -115,12 +115,12 @@ body: |
; CHECK: liveins: $q0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 -991146299
- ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[C]](s32), [[C]](s32), [[C]](s32), [[C]](s32)
- ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[C1]](s32), [[C1]](s32), [[C1]](s32), [[C1]](s32)
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 3
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = exact G_CONSTANT i32 -991146299
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = exact G_BUILD_VECTOR [[C]](s32), [[C]](s32), [[C]](s32), [[C]](s32)
+ ; CHECK-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = exact G_BUILD_VECTOR [[C1]](s32), [[C1]](s32), [[C1]](s32), [[C1]](s32)
; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(<4 x s32>) = exact G_ASHR [[COPY]], [[BUILD_VECTOR]](<4 x s32>)
- ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(<4 x s32>) = G_MUL [[ASHR]], [[BUILD_VECTOR1]]
+ ; CHECK-NEXT: [[MUL:%[0-9]+]]:_(<4 x s32>) = exact G_MUL [[ASHR]], [[BUILD_VECTOR1]]
; CHECK-NEXT: $q0 = COPY [[MUL]](<4 x s32>)
; CHECK-NEXT: RET_ReallyLR implicit $q0
%0:_(<4 x s32>) = COPY $q0
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-ashr-shl-to-sext-inreg.mir b/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-ashr-shl-to-sext-inreg.mir
index 9e540390dd37f43..9176328d873e619 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-ashr-shl-to-sext-inreg.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/prelegalizercombiner-ashr-shl-to-sext-inreg.mir
@@ -15,7 +15,7 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; CHECK-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s16) = G_SEXT_INREG [[TRUNC]], 8
+ ; CHECK-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s16) = exact G_SEXT_INREG [[TRUNC]], 8
; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SEXT_INREG]](s16)
; CHECK-NEXT: $w0 = COPY [[ANYEXT]](s32)
; CHECK-NEXT: RET_ReallyLR implicit $w0
@@ -75,7 +75,7 @@ body: |
; CHECK: liveins: $d0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
- ; CHECK-NEXT: [[SEXT_INREG:%[0-9]+]]:_(<4 x s16>) = G_SEXT_INREG [[COPY]], 8
+ ; CHECK-NEXT: [[SEXT_INREG:%[0-9]+]]:_(<4 x s16>) = exact G_SEXT_INREG [[COPY]], 8
; CHECK-NEXT: $d0 = COPY [[SEXT_INREG]](<4 x s16>)
; CHECK-NEXT: RET_ReallyLR implicit $d0
%0:_(<4 x s16>) = COPY $d0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
index ee775dc674cfecd..90df73e85c9e7c7 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-mul-pre-legalize.mir
@@ -31,7 +31,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9-CONTRACT-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -52,7 +52,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9-UNSAFE-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -73,7 +73,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX10-CONTRACT-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX10-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -94,7 +94,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX10-UNSAFE-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX10-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
%0:_(s32) = COPY $vgpr0
@@ -129,7 +129,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9-CONTRACT-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -150,7 +150,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9-UNSAFE-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -171,7 +171,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX10-CONTRACT-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX10-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
;
@@ -192,7 +192,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX10-UNSAFE-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX10-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s32) = reassoc G_FMA [[COPY]], [[COPY1]], [[COPY2]]
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[FMA]](s32)
; GFX10-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
%0:_(s32) = COPY $vgpr0
@@ -234,7 +234,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX9-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX9-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -263,7 +263,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX9-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX9-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -292,7 +292,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX10-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX10-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX10-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -321,7 +321,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX10-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX10-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX10-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -368,7 +368,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX9-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX9-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -397,7 +397,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX9-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX9-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX9-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -426,7 +426,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX10-CONTRACT-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX10-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX10-CONTRACT-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-CONTRACT-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -455,7 +455,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX10-UNSAFE-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; GFX10-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s16) = reassoc G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
; GFX10-UNSAFE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX10-UNSAFE-NEXT: S_SETPC_B64_return $sgpr30_sgpr31, implicit $vgpr0
@@ -509,7 +509,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9-CONTRACT-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -546,7 +546,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9-UNSAFE-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -583,7 +583,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX10-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX10-CONTRACT-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -620,7 +620,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX10-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX10-UNSAFE-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -679,7 +679,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9-CONTRACT-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -716,7 +716,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX9-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX9-UNSAFE-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -753,7 +753,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX10-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX10-CONTRACT-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -790,7 +790,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(s32) = COPY $vgpr4
; GFX10-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(s32) = COPY $vgpr5
; GFX10-UNSAFE-NEXT: [[MV2:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY4]](s32), [[COPY5]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[MV]], [[MV1]], [[MV2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(s64) = reassoc G_FMA [[MV]], [[MV1]], [[MV2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](s64)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -863,7 +863,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
; GFX9-CONTRACT-NEXT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
; GFX9-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -916,7 +916,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
; GFX9-UNSAFE-NEXT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
; GFX9-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -969,7 +969,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
; GFX10-CONTRACT-NEXT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
; GFX10-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1022,7 +1022,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr10
; GFX10-UNSAFE-NEXT: [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr11
; GFX10-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s32>)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1098,7 +1098,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
; GFX9-CONTRACT-NEXT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
; GFX9-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1143,7 +1143,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
; GFX9-UNSAFE-NEXT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
; GFX9-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1188,7 +1188,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
; GFX10-CONTRACT-NEXT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
; GFX10-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1233,7 +1233,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY7:%[0-9]+]]:_(s32) = COPY $vgpr7
; GFX10-UNSAFE-NEXT: [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr8
; GFX10-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s32>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s32>)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1297,7 +1297,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
; GFX9-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
; GFX9-CONTRACT-NEXT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
@@ -1334,7 +1334,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
; GFX9-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
; GFX9-UNSAFE-NEXT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
@@ -1371,7 +1371,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
; GFX10-CONTRACT-NEXT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
; GFX10-CONTRACT-NEXT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
@@ -1408,7 +1408,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[COPY4:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
; GFX10-UNSAFE-NEXT: [[COPY5:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr5
; GFX10-UNSAFE-NEXT: [[CONCAT_VECTORS2:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY4]](<2 x s16>), [[COPY5]](<2 x s16>)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s16>) = reassoc G_FMA [[CONCAT_VECTORS]], [[CONCAT_VECTORS1]], [[CONCAT_VECTORS2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[FMA]](<4 x s16>)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](<2 x s16>)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](<2 x s16>)
@@ -1489,7 +1489,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
; GFX9-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
; GFX9-CONTRACT-NEXT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
; GFX9-CONTRACT-NEXT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
; GFX9-CONTRACT-NEXT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
@@ -1550,7 +1550,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
; GFX9-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
; GFX9-UNSAFE-NEXT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
; GFX9-UNSAFE-NEXT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
; GFX9-UNSAFE-NEXT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
@@ -1611,7 +1611,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
; GFX10-CONTRACT-NEXT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
; GFX10-CONTRACT-NEXT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
; GFX10-CONTRACT-NEXT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
; GFX10-CONTRACT-NEXT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
@@ -1672,7 +1672,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[BITCAST4:%[0-9]+]]:_(s96) = G_BITCAST [[CONCAT_VECTORS2]](<6 x s16>)
; GFX10-UNSAFE-NEXT: [[TRUNC2:%[0-9]+]]:_(s48) = G_TRUNC [[BITCAST4]](s96)
; GFX10-UNSAFE-NEXT: [[BITCAST5:%[0-9]+]]:_(<3 x s16>) = G_BITCAST [[TRUNC2]](s48)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s16>) = reassoc G_FMA [[BITCAST1]], [[BITCAST3]], [[BITCAST5]]
; GFX10-UNSAFE-NEXT: [[DEF1:%[0-9]+]]:_(<3 x s16>) = G_IMPLICIT_DEF
; GFX10-UNSAFE-NEXT: [[CONCAT_VECTORS3:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[FMA]](<3 x s16>), [[DEF1]](<3 x s16>)
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS3]](<6 x s16>)
@@ -1805,7 +1805,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
; GFX9-CONTRACT-NEXT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
; GFX9-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -1914,7 +1914,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
; GFX9-UNSAFE-NEXT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
; GFX9-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2023,7 +2023,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
; GFX10-CONTRACT-NEXT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
; GFX10-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2132,7 +2132,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[MV10:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY20]](s32), [[COPY21]](s32)
; GFX10-UNSAFE-NEXT: [[MV11:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY22]](s32), [[COPY23]](s32)
; GFX10-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[MV8]](s64), [[MV9]](s64), [[MV10]](s64), [[MV11]](s64)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<4 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<4 x s64>)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2279,7 +2279,7 @@ body: |
; GFX9-CONTRACT-NEXT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
; GFX9-CONTRACT-NEXT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
; GFX9-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
- ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
; GFX9-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2366,7 +2366,7 @@ body: |
; GFX9-UNSAFE-NEXT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
; GFX9-UNSAFE-NEXT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
; GFX9-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
- ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX9-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX9-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
; GFX9-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX9-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2453,7 +2453,7 @@ body: |
; GFX10-CONTRACT-NEXT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
; GFX10-CONTRACT-NEXT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
; GFX10-CONTRACT-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
- ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-CONTRACT-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-CONTRACT-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
; GFX10-CONTRACT-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-CONTRACT-NEXT: $vgpr1 = COPY [[UV1]](s32)
@@ -2540,7 +2540,7 @@ body: |
; GFX10-UNSAFE-NEXT: [[MV7:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY14]](s32), [[COPY15]](s32)
; GFX10-UNSAFE-NEXT: [[MV8:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[COPY16]](s32), [[COPY17]](s32)
; GFX10-UNSAFE-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s64>) = G_BUILD_VECTOR [[MV6]](s64), [[MV7]](s64), [[MV8]](s64)
- ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
+ ; GFX10-UNSAFE-NEXT: [[FMA:%[0-9]+]]:_(<3 x s64>) = reassoc G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
; GFX10-UNSAFE-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA]](<3 x s64>)
; GFX10-UNSAFE-NEXT: $vgpr0 = COPY [[UV]](s32)
; GFX10-UNSAFE-NEXT: $vgpr1 = COPY [[UV1]](s32)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-unmerge-values.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-unmerge-values.mir
index 291f478e08faed9..dccf970a364a1bd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-unmerge-values.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-unmerge-values.mir
@@ -79,9 +79,9 @@ body: |
; GFX10-NEXT: %ptr:_(p1) = COPY $vgpr0_vgpr1
; GFX10-NEXT: %vec:_(<2 x s32>) = G_LOAD %ptr(p1) :: (load (<2 x s32>), addrspace 1)
; GFX10-NEXT: %el0:_(s32), %el1:_(s32) = G_UNMERGE_VALUES %vec(<2 x s32>)
- ; GFX10-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
- ; GFX10-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
- ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], %el1
+ ; GFX10-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FPEXT [[TRUNC]](s16)
+ ; GFX10-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FPEXT [[TRUNC1]](s16)
+ ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[FPEXT]], [[FPEXT1]], %el1
; GFX10-NEXT: $vgpr0 = COPY [[FMA]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s16) = G_TRUNC %0(s32)
@@ -116,9 +116,9 @@ body: |
; GFX10-NEXT: %ptr:_(p1) = COPY $vgpr0_vgpr1
; GFX10-NEXT: %vec:_(<2 x s32>) = G_LOAD %ptr(p1) :: (load (<2 x s32>), addrspace 1)
; GFX10-NEXT: %el0:_(s32), %el1:_(s32) = G_UNMERGE_VALUES %vec(<2 x s32>)
- ; GFX10-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
- ; GFX10-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
- ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], %el1
+ ; GFX10-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FPEXT [[TRUNC]](s16)
+ ; GFX10-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FPEXT [[TRUNC1]](s16)
+ ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[FPEXT]], [[FPEXT1]], %el1
; GFX10-NEXT: $vgpr0 = COPY [[FMA]](s32)
%0:_(s32) = COPY $sgpr0
%1:_(s16) = G_TRUNC %0(s32)
@@ -149,8 +149,8 @@ body: |
; GFX10-NEXT: %ptr:_(p1) = COPY $vgpr4_vgpr5
; GFX10-NEXT: %vec:_(<2 x s32>) = G_LOAD %ptr(p1) :: (load (<2 x s32>), addrspace 1)
; GFX10-NEXT: %el0:_(s32), %el1:_(s32) = G_UNMERGE_VALUES %vec(<2 x s32>)
- ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY2]], [[COPY3]], %el1
- ; GFX10-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[FMA]]
+ ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[COPY2]], [[COPY3]], %el1
+ ; GFX10-NEXT: [[FMA1:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[COPY]], [[COPY1]], [[FMA]]
; GFX10-NEXT: $vgpr0 = COPY [[FMA1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
@@ -181,8 +181,8 @@ body: |
; GFX10-NEXT: %ptr:_(p1) = COPY $vgpr4_vgpr5
; GFX10-NEXT: %vec:_(<2 x s32>) = G_LOAD %ptr(p1) :: (load (<2 x s32>), addrspace 1)
; GFX10-NEXT: %el0:_(s32), %el1:_(s32) = G_UNMERGE_VALUES %vec(<2 x s32>)
- ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY2]], [[COPY3]], %el1
- ; GFX10-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[FMA]]
+ ; GFX10-NEXT: [[FMA:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[COPY2]], [[COPY3]], %el1
+ ; GFX10-NEXT: [[FMA1:%[0-9]+]]:_(s32) = nnan ninf nsz arcp contract afn reassoc G_FMA [[COPY]], [[COPY1]], [[FMA]]
; GFX10-NEXT: $vgpr0 = COPY [[FMA1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fold-binop-into-select.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fold-binop-into-select.mir
index 9f3ad8b44444625..8a16ab53804a3b2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fold-binop-into-select.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fold-binop-into-select.mir
@@ -1034,8 +1034,8 @@ body: |
; CHECK-NEXT: %reg:_(s32) = COPY $vgpr0
; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
; CHECK-NEXT: %cond:_(s1) = G_ICMP intpred(eq), %reg(s32), %zero
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.800000e+01
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 2.000000e+01
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 1.800000e+01
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 2.000000e+01
; CHECK-NEXT: %fadd:_(s32) = nnan G_SELECT %cond(s1), [[C]], [[C1]]
; CHECK-NEXT: S_ENDPGM 0, implicit %fadd(s32)
%reg:_(s32) = COPY $vgpr0
@@ -1061,8 +1061,8 @@ body: |
; CHECK-NEXT: %reg:_(s32) = COPY $vgpr0
; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
; CHECK-NEXT: %cond:_(s1) = G_ICMP intpred(eq), %reg(s32), %zero
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.800000e+01
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 2.000000e+01
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 1.800000e+01
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 2.000000e+01
; CHECK-NEXT: %fadd:_(s32) = nnan G_SELECT %cond(s1), [[C]], [[C1]]
; CHECK-NEXT: S_ENDPGM 0, implicit %fadd(s32)
%reg:_(s32) = COPY $vgpr0
@@ -1088,8 +1088,8 @@ body: |
; CHECK-NEXT: %reg:_(s32) = COPY $vgpr0
; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
; CHECK-NEXT: %cond:_(s1) = G_ICMP intpred(eq), %reg(s32), %zero
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float -1.400000e+01
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float -1.200000e+01
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float -1.400000e+01
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float -1.200000e+01
; CHECK-NEXT: %fsub:_(s32) = nnan G_SELECT %cond(s1), [[C]], [[C1]]
; CHECK-NEXT: S_ENDPGM 0, implicit %fsub(s32)
%reg:_(s32) = COPY $vgpr0
@@ -1115,8 +1115,8 @@ body: |
; CHECK-NEXT: %reg:_(s32) = COPY $vgpr0
; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
; CHECK-NEXT: %cond:_(s1) = G_ICMP intpred(eq), %reg(s32), %zero
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 3.200000e+01
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 6.400000e+01
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 3.200000e+01
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 6.400000e+01
; CHECK-NEXT: %fmul:_(s32) = nnan G_SELECT %cond(s1), [[C]], [[C1]]
; CHECK-NEXT: S_ENDPGM 0, implicit %fmul(s32)
%reg:_(s32) = COPY $vgpr0
@@ -1142,8 +1142,8 @@ body: |
; CHECK-NEXT: %reg:_(s32) = COPY $vgpr0
; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
; CHECK-NEXT: %cond:_(s1) = G_ICMP intpred(eq), %reg(s32), %zero
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.250000e-01
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_FCONSTANT float 2.500000e-01
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 1.250000e-01
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = nnan G_FCONSTANT float 2.500000e-01
; CHECK-NEXT: %fdiv:_(s32) = nnan G_SELECT %cond(s1), [[C]], [[C1]]
; CHECK-NEXT: S_ENDPGM 0, implicit %fdiv(s32)
%reg:_(s32) = COPY $vgpr0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fsub-fneg.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fsub-fneg.mir
index 2bce20573529900..bb7014ae4546b25 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fsub-fneg.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fsub-fneg.mir
@@ -13,8 +13,8 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK-NEXT: %input:_(s16) = G_TRUNC [[COPY]](s32)
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s16) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s16) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: %res:_(s32) = G_ANYEXT [[FCANONICALIZE]](s16)
; CHECK-NEXT: $vgpr0 = COPY %res(s32)
%0:_(s32) = COPY $vgpr0
@@ -84,8 +84,8 @@ body: |
; CHECK: liveins: $vgpr0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(s32) = COPY $vgpr0
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s32) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s32) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0 = COPY [[FCANONICALIZE]](s32)
%input:_(s32) = COPY $vgpr0
%cst:_(s32) = G_FCONSTANT float 0.0
@@ -144,8 +144,8 @@ body: |
; CHECK: liveins: $vgpr0_vgpr1
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(s64) = COPY $vgpr0_vgpr1
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s64) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(s64) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(s64) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0_vgpr1 = COPY [[FCANONICALIZE]](s64)
%input:_(s64) = COPY $vgpr0_vgpr1
%cst:_(s64) = G_FCONSTANT double 0.0
@@ -204,8 +204,8 @@ body: |
; CHECK: liveins: $vgpr0_vgpr1
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(<4 x s16>) = COPY $vgpr0_vgpr1
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s16>) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s16>) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s16>) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s16>) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0_vgpr1 = COPY [[FCANONICALIZE]](<4 x s16>)
%input:_(<4 x s16>) = COPY $vgpr0_vgpr1
%cst:_(s16) = G_FCONSTANT half 0.0
@@ -268,8 +268,8 @@ body: |
; CHECK: liveins: $vgpr0_vgpr1_vgpr2_vgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s32>) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s32>) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s32>) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s32>) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FCANONICALIZE]](<4 x s32>)
%input:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%cst:_(s32) = G_FCONSTANT float 0.0
@@ -332,8 +332,8 @@ body: |
; CHECK: liveins: $vgpr0_vgpr1_vgpr2_vgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s32>) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s32>) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<4 x s32>) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<4 x s32>) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FCANONICALIZE]](<4 x s32>)
%input:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%cst:_(s32) = G_FCONSTANT float 0.0
@@ -354,8 +354,8 @@ body: |
; CHECK: liveins: $vgpr0_vgpr1_vgpr2_vgpr3
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %input:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
- ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<2 x s64>) = G_FNEG %input
- ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<2 x s64>) = G_FCANONICALIZE [[FNEG]]
+ ; CHECK-NEXT: [[FNEG:%[0-9]+]]:_(<2 x s64>) = nsz G_FNEG %input
+ ; CHECK-NEXT: [[FCANONICALIZE:%[0-9]+]]:_(<2 x s64>) = nsz G_FCANONICALIZE [[FNEG]]
; CHECK-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FCANONICALIZE]](<2 x s64>)
%input:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%cst:_(s64) = G_FCONSTANT double 0.0
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
index 0f20e75d38a5d27..3319dae71127b83 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
@@ -93,12 +93,12 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: bool GenMyCombiner::tryCombineAll(MachineInstr &I) const {
// CHECK-NEXT: const TargetSubtargetInfo &ST = MF.getSubtarget();
// CHECK-NEXT: const PredicateBitset AvailableFeatures = getAvailableFeatures();
-// CHECK-NEXT: NewMIVector OutMIs;
+// CHECK-NEXT: B.setInstrAndDebugLoc(I);
// CHECK-NEXT: State.MIs.clear();
// CHECK-NEXT: State.MIs.push_back(&I);
// CHECK-NEXT: MatchInfos = MatchInfosTy();
// CHECK-EMPTY:
-// CHECK-NEXT: if (executeMatchTable(*this, OutMIs, State, ExecInfo, getMatchTable(), *ST.getInstrInfo(), MRI, *MRI.getTargetRegisterInfo(), *ST.getRegBankInfo(), AvailableFeatures, /*CoverageInfo*/ nullptr, &Observer))
+// CHECK-NEXT: if (executeMatchTable(*this, State, ExecInfo, B, getMatchTable(), *ST.getInstrInfo(), MRI, *MRI.getTargetRegisterInfo(), *ST.getRegBankInfo(), AvailableFeatures, /*CoverageInfo*/ nullptr))
// CHECK-NEXT: return true;
// CHECK-NEXT: }
// CHECK-EMPTY:
@@ -111,7 +111,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GICXXCustomAction_CombineApplyGICombiner1,
// CHECK-NEXT: GICXXCustomAction_CombineApplyGICombiner2,
// CHECK-NEXT: };
-// CHECK-NEXT: void GenMyCombiner::runCustomAction(unsigned ApplyID, const MatcherState &State, NewMIVector &OutMIs) const {
+// CHECK-NEXT: void GenMyCombiner::runCustomAction(unsigned ApplyID, const MatcherState &State, ArrayRef<MachineInstrBuilder> OutMIs) const {
// CHECK-NEXT: switch(ApplyID) {
// CHECK-NEXT: case GICXXCustomAction_CombineApplyGICombiner0:{
// CHECK-NEXT: APPLY
diff --git a/llvm/test/TableGen/GlobalISelEmitter.td b/llvm/test/TableGen/GlobalISelEmitter.td
index 7cca2d52e4062e8..70602c8fa3ded15 100644
--- a/llvm/test/TableGen/GlobalISelEmitter.td
+++ b/llvm/test/TableGen/GlobalISelEmitter.td
@@ -82,7 +82,7 @@ def HasC : Predicate<"Subtarget->hasC()"> { let RecomputePerFunction = 1; }
// CHECK-NEXT: const int64_t *getMatchTable() const override;
// CHECK-NEXT: bool testMIPredicate_MI(unsigned PredicateID, const MachineInstr &MI, const MatcherState &State) const override;
// CHECK-NEXT: bool testSimplePredicate(unsigned PredicateID) const override;
-// CHECK-NEXT: void runCustomAction(unsigned FnID, const MatcherState &State, NewMIVector &OutMIs) const override;
+// CHECK-NEXT: void runCustomAction(unsigned FnID, const MatcherState &State, ArrayRef<MachineInstrBuilder> OutMIs) const override;
// CHECK-NEXT: #endif // ifdef GET_GLOBALISEL_TEMPORARIES_DECL
// CHECK-LABEL: #ifdef GET_GLOBALISEL_TEMPORARIES_INIT
@@ -216,11 +216,11 @@ def HasC : Predicate<"Subtarget->hasC()"> { let RecomputePerFunction = 1; }
// CHECK: bool MyTargetInstructionSelector::selectImpl(MachineInstr &I, CodeGenCoverage &CoverageInfo) const {
// CHECK-NEXT: const PredicateBitset AvailableFeatures = getAvailableFeatures();
-// CHECK-NEXT: NewMIVector OutMIs;
+// CHECK-NEXT: MachineIRBuilder B(I);
// CHECK-NEXT: State.MIs.clear();
// CHECK-NEXT: State.MIs.push_back(&I);
-// CHECK: if (executeMatchTable(*this, OutMIs, State, ExecInfo, getMatchTable(), TII, MF->getRegInfo(), TRI, RBI, AvailableFeatures, &CoverageInfo)) {
+// CHECK: if (executeMatchTable(*this, State, ExecInfo, B, getMatchTable(), TII, MF->getRegInfo(), TRI, RBI, AvailableFeatures, &CoverageInfo)) {
// CHECK-NEXT: return true;
// CHECK-NEXT: }
diff --git a/llvm/utils/TableGen/GlobalISelCombinerEmitter.cpp b/llvm/utils/TableGen/GlobalISelCombinerEmitter.cpp
index 08a4db2fb3dec91..e32ef8b9a7986b1 100644
--- a/llvm/utils/TableGen/GlobalISelCombinerEmitter.cpp
+++ b/llvm/utils/TableGen/GlobalISelCombinerEmitter.cpp
@@ -3465,15 +3465,15 @@ void GICombinerEmitter::emitAdditionalImpl(raw_ostream &OS) {
<< " const TargetSubtargetInfo &ST = MF.getSubtarget();\n"
<< " const PredicateBitset AvailableFeatures = "
"getAvailableFeatures();\n"
- << " NewMIVector OutMIs;\n"
+ << " B.setInstrAndDebugLoc(I);\n"
<< " State.MIs.clear();\n"
<< " State.MIs.push_back(&I);\n"
<< " " << MatchDataInfo::StructName << " = "
<< MatchDataInfo::StructTypeName << "();\n\n"
- << " if (executeMatchTable(*this, OutMIs, State, ExecInfo"
+ << " if (executeMatchTable(*this, State, ExecInfo, B"
<< ", getMatchTable(), *ST.getInstrInfo(), MRI, "
"*MRI.getTargetRegisterInfo(), *ST.getRegBankInfo(), AvailableFeatures"
- << ", /*CoverageInfo*/ nullptr, &Observer)) {\n"
+ << ", /*CoverageInfo*/ nullptr)) {\n"
<< " return true;\n"
<< " }\n\n"
<< " return false;\n"
@@ -3550,7 +3550,7 @@ void GICombinerEmitter::emitRunCustomAction(raw_ostream &OS) {
OS << "void " << getClassName()
<< "::runCustomAction(unsigned ApplyID, const MatcherState &State, "
- "NewMIVector &OutMIs) const "
+ "ArrayRef<MachineInstrBuilder> OutMIs) const "
"{\n";
if (!ApplyCode.empty()) {
OS << " switch(ApplyID) {\n";
diff --git a/llvm/utils/TableGen/GlobalISelEmitter.cpp b/llvm/utils/TableGen/GlobalISelEmitter.cpp
index 1dbd821f20fdc55..d98a55f717d2d86 100644
--- a/llvm/utils/TableGen/GlobalISelEmitter.cpp
+++ b/llvm/utils/TableGen/GlobalISelEmitter.cpp
@@ -2263,10 +2263,10 @@ void GlobalISelEmitter::emitAdditionalImpl(raw_ostream &OS) {
"&CoverageInfo) const {\n"
<< " const PredicateBitset AvailableFeatures = "
"getAvailableFeatures();\n"
- << " NewMIVector OutMIs;\n"
+ << " MachineIRBuilder B(I);\n"
<< " State.MIs.clear();\n"
<< " State.MIs.push_back(&I);\n\n"
- << " if (executeMatchTable(*this, OutMIs, State, ExecInfo"
+ << " if (executeMatchTable(*this, State, ExecInfo, B"
<< ", getMatchTable(), TII, MF->getRegInfo(), TRI, RBI, AvailableFeatures"
<< ", &CoverageInfo)) {\n"
<< " return true;\n"
@@ -2348,7 +2348,7 @@ void GlobalISelEmitter::emitTestSimplePredicate(raw_ostream &OS) {
void GlobalISelEmitter::emitRunCustomAction(raw_ostream &OS) {
OS << "void " << getClassName()
- << "::runCustomAction(unsigned, const MatcherState&, NewMIVector &) const "
+ << "::runCustomAction(unsigned, const MatcherState&, ArrayRef<MachineInstrBuilder>) const "
"{\n"
<< " llvm_unreachable(\"" + getClassName() +
" does not support custom C++ actions!\");\n"
diff --git a/llvm/utils/TableGen/GlobalISelMatchTableExecutorEmitter.cpp b/llvm/utils/TableGen/GlobalISelMatchTableExecutorEmitter.cpp
index c6cd3240a94e7b0..b6f9dbeb336ee7d 100644
--- a/llvm/utils/TableGen/GlobalISelMatchTableExecutorEmitter.cpp
+++ b/llvm/utils/TableGen/GlobalISelMatchTableExecutorEmitter.cpp
@@ -232,7 +232,7 @@ void GlobalISelMatchTableExecutorEmitter::emitTemporariesDecl(
"const override;\n"
<< " bool testSimplePredicate(unsigned PredicateID) const override;\n"
<< " void runCustomAction(unsigned FnID, const MatcherState &State, "
- "NewMIVector &OutMIs) "
+ "ArrayRef<MachineInstrBuilder> OutMIs) "
"const override;\n";
emitAdditionalTemporariesDecl(OS, " ");
OS << "#endif // ifdef " << IfDefName << "\n\n";
More information about the llvm-commits
mailing list