[llvm] c46d99e - [AMDGPU] Refine -O0 and -O1 passes.
Stanislav Mekhanoshin via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 15 09:52:02 PDT 2021
Author: Stanislav Mekhanoshin
Date: 2021-07-15T09:51:54-07:00
New Revision: c46d99e4ba16c76e666eb46dce710036ecbcee0d
URL: https://github.com/llvm/llvm-project/commit/c46d99e4ba16c76e666eb46dce710036ecbcee0d
DIFF: https://github.com/llvm/llvm-project/commit/c46d99e4ba16c76e666eb46dce710036ecbcee0d.diff
LOG: [AMDGPU] Refine -O0 and -O1 passes.
Differential Revision: https://reviews.llvm.org/D105579
Added:
Modified:
llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
llvm/test/CodeGen/AMDGPU/control-flow-fastregalloc.ll
llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
llvm/test/CodeGen/AMDGPU/scalar_to_vector_v2x16.ll
llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index f2a27e6452319..6fbcff1e4a6a7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -1105,7 +1105,8 @@ void AMDGPUPassConfig::addCodeGenPrepare() {
EnableLowerKernelArguments)
addPass(createAMDGPULowerKernelArgumentsPass());
- addPass(&AMDGPUPerfHintAnalysisID);
+ if (TM->getOptLevel() > CodeGenOpt::Less)
+ addPass(&AMDGPUPerfHintAnalysisID);
TargetPassConfig::addCodeGenPrepare();
@@ -1120,7 +1121,8 @@ void AMDGPUPassConfig::addCodeGenPrepare() {
}
bool AMDGPUPassConfig::addPreISel() {
- addPass(createFlattenCFGPass());
+ if (TM->getOptLevel() > CodeGenOpt::None)
+ addPass(createFlattenCFGPass());
return false;
}
@@ -1193,14 +1195,13 @@ bool GCNPassConfig::addPreISel() {
if (TM->getOptLevel() > CodeGenOpt::None)
addPass(createAMDGPULateCodeGenPreparePass());
- if (EnableAtomicOptimizations) {
+ if (isPassEnabled(EnableAtomicOptimizations, CodeGenOpt::Less)) {
addPass(createAMDGPUAtomicOptimizerPass());
}
- // FIXME: We need to run a pass to propagate the attributes when calls are
- // supported.
+ if (TM->getOptLevel() > CodeGenOpt::None)
+ addPass(createSinkingPass());
- addPass(createSinkingPass());
// Merge divergent exit nodes. StructurizeCFG won't recognize the multi-exit
// regions formed by them.
addPass(&AMDGPUUnifyDivergentExitNodesID);
@@ -1443,7 +1444,10 @@ void GCNPassConfig::addPreSched2() {
void GCNPassConfig::addPreEmitPass() {
addPass(createSIMemoryLegalizerPass());
addPass(createSIInsertWaitcntsPass());
- addPass(createSIShrinkInstructionsPass());
+
+ if (TM->getOptLevel() > CodeGenOpt::None)
+ addPass(createSIShrinkInstructionsPass());
+
addPass(createSIModeRegisterPass());
if (getOptLevel() > CodeGenOpt::None)
diff --git a/llvm/test/CodeGen/AMDGPU/control-flow-fastregalloc.ll b/llvm/test/CodeGen/AMDGPU/control-flow-fastregalloc.ll
index e25a1d232fc26..25f278f9b5f43 100644
--- a/llvm/test/CodeGen/AMDGPU/control-flow-fastregalloc.ll
+++ b/llvm/test/CodeGen/AMDGPU/control-flow-fastregalloc.ll
@@ -19,7 +19,7 @@
; Spill load
; GCN: buffer_store_dword [[LOAD0]], off, s[0:3], 0 offset:[[LOAD0_OFFSET:[0-9]+]] ; 4-byte Folded Spill
-; GCN: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]\]]], s{{[0-9]+}}, v0
+; GCN: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]\]]], v0, s{{[0-9]+}}
; Spill saved exec
; GCN: s_mov_b64 s{{\[}}[[SAVEEXEC_LO:[0-9]+]]:[[SAVEEXEC_HI:[0-9]+]]{{\]}}, exec
@@ -88,7 +88,7 @@ endif:
; GCN-DAG: s_mov_b32 m0, -1
; GCN-DAG: v_mov_b32_e32 [[PTR0:v[0-9]+]], 0{{$}}
; GCN: ds_read_b32 [[LOAD0:v[0-9]+]], [[PTR0]]
-; GCN: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], s{{[0-9]+}}, v0
+; GCN: v_cmp_eq_u32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], v0, s{{[0-9]+}}
; Spill load
; GCN: buffer_store_dword [[LOAD0]], off, s[0:3], 0 offset:[[LOAD0_OFFSET:[0-9]+]] ; 4-byte Folded Spill
@@ -166,7 +166,7 @@ end:
; GCN: buffer_store_dword [[LOAD0]], off, s[0:3], 0 offset:[[LOAD0_OFFSET:[0-9]+]] ; 4-byte Folded Spill
; GCN: s_mov_b32 [[ZERO:s[0-9]+]], 0
-; GCN: v_cmp_ne_u32_e64 [[CMP0:s\[[0-9]+:[0-9]\]]], [[ZERO]], v0
+; GCN: v_cmp_ne_u32_e64 [[CMP0:s\[[0-9]+:[0-9]\]]], v0, [[ZERO]]
; GCN: s_mov_b64 s{{\[}}[[SAVEEXEC_LO:[0-9]+]]:[[SAVEEXEC_HI:[0-9]+]]{{\]}}, exec
; GCN: s_and_b64 s{{\[}}[[ANDEXEC_LO:[0-9]+]]:[[ANDEXEC_HI:[0-9]+]]{{\]}}, s{{\[}}[[SAVEEXEC_LO:[0-9]+]]:[[SAVEEXEC_HI:[0-9]+]]{{\]}}, [[CMP0]]
diff --git a/llvm/test/CodeGen/AMDGPU/llc-pipeline.ll b/llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
index 42ea14f303794..17e31d2eecd7b 100644
--- a/llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
+++ b/llvm/test/CodeGen/AMDGPU/llc-pipeline.ll
@@ -54,22 +54,13 @@
; GCN-O0-NEXT: AMDGPU Annotate Kernel Features
; GCN-O0-NEXT: FunctionPass Manager
; GCN-O0-NEXT: AMDGPU Lower Kernel Arguments
-; GCN-O0-NEXT: Analysis if a function is memory bound
-; GCN-O0-NEXT: FunctionPass Manager
; GCN-O0-NEXT: Lazy Value Information Analysis
; GCN-O0-NEXT: Lower SwitchInst's to branches
; GCN-O0-NEXT: Lower invoke and unwind, for unwindless code generators
; GCN-O0-NEXT: Remove unreachable blocks from the CFG
+; GCN-O0-NEXT: Post-Dominator Tree Construction
; GCN-O0-NEXT: Dominator Tree Construction
-; GCN-O0-NEXT: Basic Alias Analysis (stateless AA impl)
-; GCN-O0-NEXT: Function Alias Analysis Results
-; GCN-O0-NEXT: Flatten the CFG
-; GCN-O0-NEXT: Dominator Tree Construction
-; GCN-O0-NEXT: Basic Alias Analysis (stateless AA impl)
-; GCN-O0-NEXT: Function Alias Analysis Results
; GCN-O0-NEXT: Natural Loop Information
-; GCN-O0-NEXT: Code sinking
-; GCN-O0-NEXT: Post-Dominator Tree Construction
; GCN-O0-NEXT: Legacy Divergence Analysis
; GCN-O0-NEXT: Unify divergent function exit nodes
; GCN-O0-NEXT: Lazy Value Information Analysis
@@ -138,7 +129,6 @@
; GCN-O0-NEXT: SI Memory Legalizer
; GCN-O0-NEXT: MachinePostDominator Tree Construction
; GCN-O0-NEXT: SI insert wait instructions
-; GCN-O0-NEXT: SI Shrink Instructions
; GCN-O0-NEXT: Insert required mode register values
; GCN-O0-NEXT: MachineDominator Tree Construction
; GCN-O0-NEXT: SI Final Branch Preparation
@@ -225,8 +215,6 @@
; GCN-O1-NEXT: AMDGPU Annotate Kernel Features
; GCN-O1-NEXT: FunctionPass Manager
; GCN-O1-NEXT: AMDGPU Lower Kernel Arguments
-; GCN-O1-NEXT: Analysis if a function is memory bound
-; GCN-O1-NEXT: FunctionPass Manager
; GCN-O1-NEXT: Dominator Tree Construction
; GCN-O1-NEXT: Natural Loop Information
; GCN-O1-NEXT: CodeGen Prepare
@@ -495,8 +483,6 @@
; GCN-O1-OPTS-NEXT: AMDGPU Annotate Kernel Features
; GCN-O1-OPTS-NEXT: FunctionPass Manager
; GCN-O1-OPTS-NEXT: AMDGPU Lower Kernel Arguments
-; GCN-O1-OPTS-NEXT: Analysis if a function is memory bound
-; GCN-O1-OPTS-NEXT: FunctionPass Manager
; GCN-O1-OPTS-NEXT: Dominator Tree Construction
; GCN-O1-OPTS-NEXT: Natural Loop Information
; GCN-O1-OPTS-NEXT: CodeGen Prepare
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
index 87812fa0bf78d..18c4afd08e4f3 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
@@ -67,7 +67,7 @@ define amdgpu_kernel void @update_dpp64_test(i64 addrspace(1)* %arg, i64 %in1, i
; GCN-OPT-DAG: v_mov_b32_e32 v[[OLD_LO:[0-9]+]], 0x3afaedd9
; GCN-OPT-DAG: v_mov_b32_e32 v[[OLD_HI:[0-9]+]], 0x7047
; GFX8-NOOPT-DAG: s_mov_b32 s[[SOLD_LO:[0-9]+]], 0x3afaedd9
-; GFX8-NOOPT-DAG: s_movk_i32 s[[SOLD_HI:[0-9]+]], 0x7047
+; GFX8-NOOPT-DAG: s_mov_b32 s[[SOLD_HI:[0-9]+]], 0x7047
; GCN-DAG: load_dwordx2 v{{\[}}[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
; GCN-OPT-DAG: v_mov_b32_dpp v[[OLD_LO]], v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
; GCN-OPT-DAG: v_mov_b32_dpp v[[OLD_HI]], v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
@@ -86,7 +86,7 @@ define amdgpu_kernel void @update_dpp64_imm_old_test(i64 addrspace(1)* %arg, i64
; GCN-OPT-DAG: v_mov_b32_e32 v[[OLD_LO:[0-9]+]], 0x3afaedd9
; GCN-OPT-DAG: v_mov_b32_e32 v[[OLD_HI:[0-9]+]], 0x7047
; GFX8-NOOPT-DAG: s_mov_b32 s[[SOLD_LO:[0-9]+]], 0x3afaedd9
-; GFX8-NOOPT-DAG: s_movk_i32 s[[SOLD_HI:[0-9]+]], 0x7047
+; GFX8-NOOPT-DAG: s_mov_b32 s[[SOLD_HI:[0-9]+]], 0x7047
; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[OLD_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[OLD_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
; GCN-NOOPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
diff --git a/llvm/test/CodeGen/AMDGPU/scalar_to_vector_v2x16.ll b/llvm/test/CodeGen/AMDGPU/scalar_to_vector_v2x16.ll
index 6e7d24cdc3c73..7af841f08d9e9 100644
--- a/llvm/test/CodeGen/AMDGPU/scalar_to_vector_v2x16.ll
+++ b/llvm/test/CodeGen/AMDGPU/scalar_to_vector_v2x16.ll
@@ -14,7 +14,7 @@ define void @scalar_to_vector_i16() {
}
; GCN-LABEL: {{^}}scalar_to_vector_f16:
-; GCN-NOOPT: s_movk_i32 [[S:s[0-9]+]], 0x3c00
+; GCN-NOOPT: s_mov_b32 [[S:s[0-9]+]], 0x3c00
; GCN-NOOPT: v_mov_b32_e32 [[V:v[0-9]+]], [[S]]
; GCN-OPT: v_mov_b32_e32 [[V:v[0-9]+]], 0x3c00
; GCN: buffer_store_short [[V]],
diff --git a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
index f27593f3a24f9..335181da67d28 100644
--- a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
+++ b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
@@ -36,20 +36,20 @@ define amdgpu_gfx void @strict_wwm_no_cfg(<4 x i32> inreg %tmp14) {
; GFX9-O0-NEXT: v_mov_b32_e32 v2, s8
; GFX9-O0-NEXT: s_nop 1
; GFX9-O0-NEXT: v_mov_b32_dpp v2, v0 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-O0-NEXT: v_add_u32_e32 v0, v0, v2
+; GFX9-O0-NEXT: v_add_u32_e64 v0, v0, v2
; GFX9-O0-NEXT: s_mov_b64 exec, s[10:11]
; GFX9-O0-NEXT: v_mov_b32_e32 v3, v0
; GFX9-O0-NEXT: s_or_saveexec_b64 s[10:11], -1
; GFX9-O0-NEXT: v_mov_b32_e32 v0, s8
; GFX9-O0-NEXT: s_nop 1
; GFX9-O0-NEXT: v_mov_b32_dpp v0, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-O0-NEXT: v_add_u32_e32 v0, v1, v0
+; GFX9-O0-NEXT: v_add_u32_e64 v0, v1, v0
; GFX9-O0-NEXT: s_mov_b64 exec, s[10:11]
; GFX9-O0-NEXT: v_mov_b32_e32 v4, v0
; GFX9-O0-NEXT: v_cmp_eq_u32_e64 s[10:11], v3, v4
; GFX9-O0-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[10:11]
; GFX9-O0-NEXT: s_mov_b32 s9, 1
-; GFX9-O0-NEXT: v_lshlrev_b32_e32 v3, s9, v3
+; GFX9-O0-NEXT: v_lshlrev_b32_e64 v3, s9, v3
; GFX9-O0-NEXT: s_mov_b32 s9, 2
; GFX9-O0-NEXT: v_mov_b32_e32 v4, s9
; GFX9-O0-NEXT: v_and_b32_e32 v3, v3, v4
@@ -170,11 +170,11 @@ define amdgpu_gfx void @strict_wwm_cfg(<4 x i32> inreg %tmp14, i32 %arg) {
; GFX9-O0-NEXT: v_mov_b32_e32 v2, s4
; GFX9-O0-NEXT: s_nop 1
; GFX9-O0-NEXT: v_mov_b32_dpp v2, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-O0-NEXT: v_add_u32_e32 v1, v1, v2
+; GFX9-O0-NEXT: v_add_u32_e64 v1, v1, v2
; GFX9-O0-NEXT: s_mov_b64 exec, s[6:7]
; GFX9-O0-NEXT: v_mov_b32_e32 v3, v1
; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], s4, v0
+; GFX9-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, s4
; GFX9-O0-NEXT: v_mov_b32_e32 v0, s4
; GFX9-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
; GFX9-O0-NEXT: s_mov_b64 s[4:5], exec
@@ -198,7 +198,7 @@ define amdgpu_gfx void @strict_wwm_cfg(<4 x i32> inreg %tmp14, i32 %arg) {
; GFX9-O0-NEXT: s_not_b64 exec, exec
; GFX9-O0-NEXT: s_or_saveexec_b64 s[4:5], -1
; GFX9-O0-NEXT: v_mov_b32_dpp v1, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-O0-NEXT: v_add_u32_e32 v1, v2, v1
+; GFX9-O0-NEXT: v_add_u32_e64 v1, v2, v1
; GFX9-O0-NEXT: s_mov_b64 exec, s[4:5]
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v1
; GFX9-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
@@ -218,7 +218,7 @@ define amdgpu_gfx void @strict_wwm_cfg(<4 x i32> inreg %tmp14, i32 %arg) {
; GFX9-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v0, v3
; GFX9-O0-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[6:7]
; GFX9-O0-NEXT: s_mov_b32 s6, 1
-; GFX9-O0-NEXT: v_lshlrev_b32_e32 v0, s6, v0
+; GFX9-O0-NEXT: v_lshlrev_b32_e64 v0, s6, v0
; GFX9-O0-NEXT: s_mov_b32 s6, 2
; GFX9-O0-NEXT: v_mov_b32_e32 v3, s6
; GFX9-O0-NEXT: v_and_b32_e32 v0, v0, v3
@@ -321,11 +321,11 @@ define hidden i32 @strict_wwm_called(i32 %a) noinline {
; GFX9-O0-LABEL: strict_wwm_called:
; GFX9-O0: ; %bb.0:
; GFX9-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-O0-NEXT: v_add_u32_e32 v1, v0, v0
+; GFX9-O0-NEXT: v_add_u32_e64 v1, v0, v0
; GFX9-O0-NEXT: ; implicit-def: $sgpr4
; GFX9-O0-NEXT: ; implicit-def: $sgpr4
; GFX9-O0-NEXT: v_mul_lo_u32 v0, v1, v0
-; GFX9-O0-NEXT: v_sub_u32_e32 v0, v0, v1
+; GFX9-O0-NEXT: v_sub_u32_e64 v0, v0, v1
; GFX9-O0-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-O3-LABEL: strict_wwm_called:
@@ -352,7 +352,7 @@ define amdgpu_gfx void @strict_wwm_call(<4 x i32> inreg %tmp14, i32 inreg %arg)
; GFX9-O0-NEXT: s_mov_b64 exec, s[10:11]
; GFX9-O0-NEXT: v_writelane_b32 v3, s33, 7
; GFX9-O0-NEXT: s_mov_b32 s33, s32
-; GFX9-O0-NEXT: s_addk_i32 s32, 0x400
+; GFX9-O0-NEXT: s_add_i32 s32, s32, 0x400
; GFX9-O0-NEXT: v_writelane_b32 v3, s30, 0
; GFX9-O0-NEXT: v_writelane_b32 v3, s31, 1
; GFX9-O0-NEXT: v_writelane_b32 v3, s8, 2
@@ -389,11 +389,11 @@ define amdgpu_gfx void @strict_wwm_call(<4 x i32> inreg %tmp14, i32 inreg %arg)
; GFX9-O0-NEXT: v_readlane_b32 s30, v3, 0
; GFX9-O0-NEXT: v_readlane_b32 s31, v3, 1
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v0
-; GFX9-O0-NEXT: v_add_u32_e32 v1, v1, v2
+; GFX9-O0-NEXT: v_add_u32_e64 v1, v1, v2
; GFX9-O0-NEXT: s_mov_b64 exec, s[10:11]
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v1
; GFX9-O0-NEXT: buffer_store_dword v0, off, s[4:7], s8 offset:4
-; GFX9-O0-NEXT: s_addk_i32 s32, 0xfc00
+; GFX9-O0-NEXT: s_add_i32 s32, s32, 0xfffffc00
; GFX9-O0-NEXT: v_readlane_b32 s33, v3, 7
; GFX9-O0-NEXT: s_or_saveexec_b64 s[4:5], -1
; GFX9-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
@@ -553,7 +553,7 @@ define amdgpu_gfx void @strict_wwm_call_i64(<4 x i32> inreg %tmp14, i64 inreg %a
; GFX9-O0-NEXT: s_mov_b64 exec, s[10:11]
; GFX9-O0-NEXT: v_writelane_b32 v11, s33, 9
; GFX9-O0-NEXT: s_mov_b32 s33, s32
-; GFX9-O0-NEXT: s_addk_i32 s32, 0xc00
+; GFX9-O0-NEXT: s_add_i32 s32, s32, 0xc00
; GFX9-O0-NEXT: v_writelane_b32 v11, s30, 0
; GFX9-O0-NEXT: v_writelane_b32 v11, s31, 1
; GFX9-O0-NEXT: v_writelane_b32 v11, s9, 2
@@ -619,7 +619,7 @@ define amdgpu_gfx void @strict_wwm_call_i64(<4 x i32> inreg %tmp14, i64 inreg %a
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v3
; GFX9-O0-NEXT: s_mov_b32 s8, 0
; GFX9-O0-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], s8 offset:4
-; GFX9-O0-NEXT: s_addk_i32 s32, 0xf400
+; GFX9-O0-NEXT: s_add_i32 s32, s32, 0xfffff400
; GFX9-O0-NEXT: v_readlane_b32 s33, v11, 9
; GFX9-O0-NEXT: s_or_saveexec_b64 s[4:5], -1
; GFX9-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
@@ -729,7 +729,7 @@ define amdgpu_gfx void @strict_wwm_amdgpu_cs_main(<4 x i32> inreg %desc, i32 %in
; GFX9-O0-NEXT: s_mov_b32 s7, s8
; GFX9-O0-NEXT: ; kill: def $sgpr8_sgpr9_sgpr10_sgpr11 killed $sgpr4_sgpr5_sgpr6_sgpr7
; GFX9-O0-NEXT: s_mov_b32 s8, 5
-; GFX9-O0-NEXT: v_lshlrev_b32_e32 v0, s8, v0
+; GFX9-O0-NEXT: v_lshlrev_b32_e64 v0, s8, v0
; GFX9-O0-NEXT: s_mov_b32 s8, 0
; GFX9-O0-NEXT: buffer_load_dwordx4 v[10:13], v0, s[4:7], s8 offen
; GFX9-O0-NEXT: buffer_load_dwordx2 v[3:4], v0, s[4:7], s8 offen offset:16
@@ -738,7 +738,7 @@ define amdgpu_gfx void @strict_wwm_amdgpu_cs_main(<4 x i32> inreg %desc, i32 %in
; GFX9-O0-NEXT: v_mov_b32_e32 v5, v10
; GFX9-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6 killed $exec
; GFX9-O0-NEXT: v_mov_b32_e32 v6, v7
-; GFX9-O0-NEXT: s_brev_b32 s9, -2
+; GFX9-O0-NEXT: s_mov_b32 s9, 0x7fffffff
; GFX9-O0-NEXT: s_mov_b32 s10, -1
; GFX9-O0-NEXT: ; kill: def $sgpr10 killed $sgpr10 def $sgpr10_sgpr11
; GFX9-O0-NEXT: s_mov_b32 s11, s9
diff --git a/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll b/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
index b0212bab0f9b6..340e7e56b6584 100644
--- a/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
+++ b/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
@@ -15,14 +15,16 @@ define amdgpu_cs void @no_cfg(<4 x i32> inreg %tmp14) {
; GFX9: s_or_saveexec_b64 s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, -1
; GFX9-DAG: v_mov_b32_dpp v[[FIRST_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DAG: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O3-DAG: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O0-DAG: v_add_u32_e64 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
; GFX9-DAG: v_mov_b32_e32 v[[FIRST:[0-9]+]], v[[FIRST_ADD]]
%tmp120 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp105, i32 323, i32 12, i32 15, i1 false)
%tmp121 = add i32 %tmp105, %tmp120
%tmp122 = tail call i32 @llvm.amdgcn.wwm.i32(i32 %tmp121)
; GFX9-DAG: v_mov_b32_dpp v[[SECOND_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DAG: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O3-DAG: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O0-DAG: v_add_u32_e64 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
; GFX9-DAG: v_mov_b32_e32 v[[SECOND:[0-9]+]], v[[SECOND_ADD]]
%tmp135 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp107, i32 323, i32 12, i32 15, i1 false)
%tmp136 = add i32 %tmp107, %tmp135
@@ -48,7 +50,8 @@ entry:
%tmp105 = tail call i32 @llvm.amdgcn.set.inactive.i32(i32 %tmp102, i32 0)
; GFX9: v_mov_b32_dpp v[[FIRST_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O3: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O0: v_add_u32_e64 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
; GFX9: v_mov_b32_e32 v[[FIRST:[0-9]+]], v[[FIRST_ADD]]
; GFX9-O0: buffer_store_dword v[[FIRST]], off, s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, 0 offset:[[FIRST_IMM_OFFSET:[0-9]+]]
%tmp120 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp105, i32 323, i32 12, i32 15, i1 false)
@@ -62,7 +65,8 @@ if:
%tmp107 = tail call i32 @llvm.amdgcn.set.inactive.i32(i32 %tmp103, i32 0)
; GFX9: v_mov_b32_dpp v[[SECOND_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O3: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O0: v_add_u32_e64 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
; GFX9: v_mov_b32_e32 v[[SECOND:[0-9]+]], v[[SECOND_ADD]]
; GFX9-O0: buffer_store_dword v[[SECOND]], off, s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, 0 offset:[[SECOND_IMM_OFFSET:[0-9]+]]
%tmp135 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp107, i32 323, i32 12, i32 15, i1 false)
@@ -87,11 +91,13 @@ merge:
; GFX9-LABEL: {{^}}called:
define hidden i32 @called(i32 %a) noinline {
-; GFX9: v_add_u32_e32 v1, v0, v0
+; GFX9-O3: v_add_u32_e32 v1, v0, v0
+; GFX9-O0: v_add_u32_e64 v1, v0, v0
%add = add i32 %a, %a
; GFX9: v_mul_lo_u32 v0, v1, v0
%mul = mul i32 %add, %a
-; GFX9: v_sub_u32_e32 v0, v0, v1
+; GFX9-O3: v_sub_u32_e32 v0, v0, v1
+; GFX9-O0: v_sub_u32_e64 v0, v0, v1
%sub = sub i32 %mul, %add
ret i32 %sub
}
@@ -114,7 +120,8 @@ define amdgpu_kernel void @call(<4 x i32> inreg %tmp14, i32 inreg %arg) {
; GFX9: s_swappc_b64
%tmp134 = call i32 @called(i32 %tmp107)
; GFX9: v_mov_b32_e32 v1, v0
-; GFX9: v_add_u32_e32 v1, v1, v2
+; GFX9-O3: v_add_u32_e32 v1, v1, v2
+; GFX9-O0: v_add_u32_e64 v1, v1, v2
%tmp136 = add i32 %tmp134, %tmp107
%tmp137 = tail call i32 @llvm.amdgcn.wwm.i32(i32 %tmp136)
; GFX9: buffer_store_dword v0
@@ -202,14 +209,16 @@ define amdgpu_cs void @strict_wwm_no_cfg(<4 x i32> inreg %tmp14) {
; GFX9: s_or_saveexec_b64 s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, -1
; GFX9-DAG: v_mov_b32_dpp v[[FIRST_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DAG: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O3-DAG: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O0-DAG: v_add_u32_e64 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
; GFX9-DAG: v_mov_b32_e32 v[[FIRST:[0-9]+]], v[[FIRST_ADD]]
%tmp120 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp105, i32 323, i32 12, i32 15, i1 false)
%tmp121 = add i32 %tmp105, %tmp120
%tmp122 = tail call i32 @llvm.amdgcn.strict.wwm.i32(i32 %tmp121)
; GFX9-DAG: v_mov_b32_dpp v[[SECOND_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DAG: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O3-DAG: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O0-DAG: v_add_u32_e64 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
; GFX9-DAG: v_mov_b32_e32 v[[SECOND:[0-9]+]], v[[SECOND_ADD]]
%tmp135 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp107, i32 323, i32 12, i32 15, i1 false)
%tmp136 = add i32 %tmp107, %tmp135
@@ -235,7 +244,8 @@ entry:
%tmp105 = tail call i32 @llvm.amdgcn.set.inactive.i32(i32 %tmp102, i32 0)
; GFX9: v_mov_b32_dpp v[[FIRST_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O3: v_add_u32_e32 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
+; GFX9-O0: v_add_u32_e64 v[[FIRST_ADD:[0-9]+]], v{{[0-9]+}}, v[[FIRST_MOV]]
; GFX9: v_mov_b32_e32 v[[FIRST:[0-9]+]], v[[FIRST_ADD]]
; GFX9-O0: buffer_store_dword v[[FIRST]], off, s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, 0 offset:[[FIRST_IMM_OFFSET:[0-9]+]]
%tmp120 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp105, i32 323, i32 12, i32 15, i1 false)
@@ -249,7 +259,8 @@ if:
%tmp107 = tail call i32 @llvm.amdgcn.set.inactive.i32(i32 %tmp103, i32 0)
; GFX9: v_mov_b32_dpp v[[SECOND_MOV:[0-9]+]], v{{[0-9]+}} row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O3: v_add_u32_e32 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
+; GFX9-O0: v_add_u32_e64 v[[SECOND_ADD:[0-9]+]], v{{[0-9]+}}, v[[SECOND_MOV]]
; GFX9: v_mov_b32_e32 v[[SECOND:[0-9]+]], v[[SECOND_ADD]]
; GFX9-O0: buffer_store_dword v[[SECOND]], off, s{{\[}}{{[0-9]+}}:{{[0-9]+}}{{\]}}, 0 offset:[[SECOND_IMM_OFFSET:[0-9]+]]
%tmp135 = tail call i32 @llvm.amdgcn.update.dpp.i32(i32 0, i32 %tmp107, i32 323, i32 12, i32 15, i1 false)
@@ -274,11 +285,13 @@ merge:
; GFX9-LABEL: {{^}}strict_wwm_called:
define hidden i32 @strict_wwm_called(i32 %a) noinline {
-; GFX9: v_add_u32_e32 v1, v0, v0
+; GFX9-O3: v_add_u32_e32 v1, v0, v0
+; GFX9-O0: v_add_u32_e64 v1, v0, v0
%add = add i32 %a, %a
; GFX9: v_mul_lo_u32 v0, v1, v0
%mul = mul i32 %add, %a
-; GFX9: v_sub_u32_e32 v0, v0, v1
+; GFX9-O3: v_sub_u32_e32 v0, v0, v1
+; GFX9-O0: v_sub_u32_e64 v0, v0, v1
%sub = sub i32 %mul, %add
ret i32 %sub
}
@@ -301,7 +314,8 @@ define amdgpu_kernel void @strict_wwm_call(<4 x i32> inreg %tmp14, i32 inreg %ar
; GFX9: s_swappc_b64
%tmp134 = call i32 @strict_wwm_called(i32 %tmp107)
; GFX9: v_mov_b32_e32 v1, v0
-; GFX9: v_add_u32_e32 v1, v1, v2
+; GFX9-O3: v_add_u32_e32 v1, v1, v2
+; GFX9-O0: v_add_u32_e64 v1, v1, v2
%tmp136 = add i32 %tmp134, %tmp107
%tmp137 = tail call i32 @llvm.amdgcn.strict.wwm.i32(i32 %tmp136)
; GFX9: buffer_store_dword v0
More information about the llvm-commits
mailing list