[AMDGPU] PeepholeSDWA: Don't assume inst srcs are registers #69576
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
To fix that ticket we only needed to address the V_LSHLREV_B16 case, but I did it for all insts just in case. Fixes llvm#66899
|
@llvm/pr-subscribers-backend-amdgpu Author: Pierre van Houtryve (Pierre-vh) ChangesTo fix that ticket we only needed to address the V_LSHLREV_B16 case, but I did it for all insts just in case. Fixes #66899 Full diff: https://github.com/llvm/llvm-project/pull/69576.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
index 97b3161c7f98b13..53fc2c0686245f9 100644
--- a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
+++ b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
@@ -546,7 +546,8 @@ SIPeepholeSDWA::matchSDWAOperand(MachineInstr &MI) {
MachineOperand *Src1 = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
MachineOperand *Dst = TII->getNamedOperand(MI, AMDGPU::OpName::vdst);
- if (Src1->getReg().isPhysical() || Dst->getReg().isPhysical())
+ if (!Src1->isReg() || Src1->getReg().isPhysical() ||
+ Dst->getReg().isPhysical())
break;
if (Opcode == AMDGPU::V_LSHLREV_B32_e32 ||
@@ -584,7 +585,8 @@ SIPeepholeSDWA::matchSDWAOperand(MachineInstr &MI) {
MachineOperand *Src1 = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
MachineOperand *Dst = TII->getNamedOperand(MI, AMDGPU::OpName::vdst);
- if (Src1->getReg().isPhysical() || Dst->getReg().isPhysical())
+ if (!Src1->isReg() || Src1->getReg().isPhysical() ||
+ Dst->getReg().isPhysical())
break;
if (Opcode == AMDGPU::V_LSHLREV_B16_e32 ||
@@ -647,7 +649,8 @@ SIPeepholeSDWA::matchSDWAOperand(MachineInstr &MI) {
MachineOperand *Src0 = TII->getNamedOperand(MI, AMDGPU::OpName::src0);
MachineOperand *Dst = TII->getNamedOperand(MI, AMDGPU::OpName::vdst);
- if (Src0->getReg().isPhysical() || Dst->getReg().isPhysical())
+ if (!Src0->isReg() || Src0->getReg().isPhysical() ||
+ Dst->getReg().isPhysical())
break;
return std::make_unique<SDWASrcOperand>(
@@ -675,7 +678,8 @@ SIPeepholeSDWA::matchSDWAOperand(MachineInstr &MI) {
MachineOperand *Dst = TII->getNamedOperand(MI, AMDGPU::OpName::vdst);
- if (ValSrc->getReg().isPhysical() || Dst->getReg().isPhysical())
+ if (!ValSrc->isReg() || ValSrc->getReg().isPhysical() ||
+ Dst->getReg().isPhysical())
break;
return std::make_unique<SDWASrcOperand>(
diff --git a/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll b/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
index 79e96b45a901c4c..086fab3f5217589 100644
--- a/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
@@ -2098,6 +2098,113 @@ bb11: ; preds = %bb10, %bb2
br label %bb1
}
+define void @crash_lshlrevb16_not_reg_op() {
+; NOSDWA-LABEL: crash_lshlrevb16_not_reg_op:
+; NOSDWA: ; %bb.0: ; %bb0
+; NOSDWA-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; NOSDWA-NEXT: s_mov_b64 s[4:5], 0
+; NOSDWA-NEXT: s_and_b64 vcc, exec, -1
+; NOSDWA-NEXT: v_lshlrev_b16_e64 v3, 8, 1
+; NOSDWA-NEXT: .LBB22_1: ; %bb1
+; NOSDWA-NEXT: ; =>This Inner Loop Header: Depth=1
+; NOSDWA-NEXT: v_mov_b32_e32 v0, s4
+; NOSDWA-NEXT: v_mov_b32_e32 v2, 0xff
+; NOSDWA-NEXT: s_lshl_b32 s6, s4, 3
+; NOSDWA-NEXT: v_mov_b32_e32 v1, s5
+; NOSDWA-NEXT: s_mov_b64 s[4:5], 1
+; NOSDWA-NEXT: v_and_b32_e32 v2, s4, v2
+; NOSDWA-NEXT: v_or_b32_e32 v2, v2, v3
+; NOSDWA-NEXT: v_lshrrev_b16_e32 v2, s6, v2
+; NOSDWA-NEXT: flat_store_byte v[0:1], v2
+; NOSDWA-NEXT: s_mov_b64 vcc, vcc
+; NOSDWA-NEXT: s_cbranch_vccnz .LBB22_1
+; NOSDWA-NEXT: ; %bb.2: ; %DummyReturnBlock
+; NOSDWA-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; NOSDWA-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX89-LABEL: crash_lshlrevb16_not_reg_op:
+; GFX89: ; %bb.0: ; %bb0
+; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX89-NEXT: s_mov_b64 s[4:5], 0
+; GFX89-NEXT: s_and_b64 vcc, exec, -1
+; GFX89-NEXT: v_lshlrev_b16_e64 v0, 8, 1
+; GFX89-NEXT: .LBB22_1: ; %bb1
+; GFX89-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX89-NEXT: v_mov_b32_e32 v3, s4
+; GFX89-NEXT: s_lshl_b32 s6, s4, 3
+; GFX89-NEXT: v_mov_b32_e32 v1, s4
+; GFX89-NEXT: v_or_b32_sdwa v3, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX89-NEXT: v_mov_b32_e32 v2, s5
+; GFX89-NEXT: s_mov_b64 s[4:5], 1
+; GFX89-NEXT: v_lshrrev_b16_e32 v3, s6, v3
+; GFX89-NEXT: flat_store_byte v[1:2], v3
+; GFX89-NEXT: s_mov_b64 vcc, vcc
+; GFX89-NEXT: s_cbranch_vccnz .LBB22_1
+; GFX89-NEXT: ; %bb.2: ; %DummyReturnBlock
+; GFX89-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX89-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: crash_lshlrevb16_not_reg_op:
+; GFX9: ; %bb.0: ; %bb0
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-NEXT: v_lshlrev_b16_e64 v0, 8, 1
+; GFX9-NEXT: s_and_b64 vcc, exec, -1
+; GFX9-NEXT: v_or_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT: .LBB22_1: ; %bb1
+; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-NEXT: s_lshl_b32 s6, s4, 3
+; GFX9-NEXT: v_mov_b32_e32 v1, s4
+; GFX9-NEXT: v_mov_b32_e32 v2, s5
+; GFX9-NEXT: s_mov_b64 s[4:5], 1
+; GFX9-NEXT: v_lshrrev_b16_e32 v3, s6, v0
+; GFX9-NEXT: flat_store_byte v[1:2], v3
+; GFX9-NEXT: s_mov_b64 vcc, vcc
+; GFX9-NEXT: s_cbranch_vccnz .LBB22_1
+; GFX9-NEXT: ; %bb.2: ; %DummyReturnBlock
+; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-LABEL: crash_lshlrevb16_not_reg_op:
+; GFX10: ; %bb.0: ; %bb0
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_lshlrev_b16 v0, 8, 1
+; GFX10-NEXT: s_mov_b32 vcc_lo, exec_lo
+; GFX10-NEXT: v_or_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX10-NEXT: s_mov_b64 s[4:5], 0
+; GFX10-NEXT: .LBB22_1: ; %bb1
+; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX10-NEXT: s_lshl_b32 s6, s4, 3
+; GFX10-NEXT: v_mov_b32_e32 v1, s4
+; GFX10-NEXT: v_mov_b32_e32 v2, s5
+; GFX10-NEXT: v_lshrrev_b16 v3, s6, v0
+; GFX10-NEXT: s_mov_b64 s[4:5], 1
+; GFX10-NEXT: flat_store_byte v[1:2], v3
+; GFX10-NEXT: s_cbranch_vccnz .LBB22_1
+; GFX10-NEXT: ; %bb.2: ; %DummyReturnBlock
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_setpc_b64 s[30:31]
+ %1 = alloca [2 x i8], align 1, addrspace(5)
+ %2 = getelementptr [2 x i8], ptr addrspace(5) %1, i32 0, i32 1
+ br label %bb0
+
+bb0:
+ store i8 1, ptr addrspace(5) %2, align 1
+ br label %bb1
+
+bb1:
+ %3 = phi i64 [ 1, %bb1 ], [ 0, %bb0 ]
+ %4 = trunc i64 %3 to i32
+ %5 = getelementptr i8, ptr addrspace(5) %1, i32 %4
+ %6 = load i8, ptr addrspace(5) %5, align 1
+ %7 = getelementptr i8, ptr null, i64 %3
+ store i8 %6, ptr %7, align 1
+ br i1 false, label %bb2, label %bb1
+
+bb2:
+ br label %bb0
+}
+
declare i32 @llvm.amdgcn.workitem.id.x()
attributes #0 = { "denormal-fp-math"="preserve-sign,preserve-sign" }
|
jayfoad
approved these changes
Oct 19, 2023
|
As a follow up, it would be nice it see why the V_LSHLREV_B16 did not get folded and if we can fix it. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
To fix that ticket we only needed to address the V_LSHLREV_B16 case, but I did it for all insts just in case.
Fixes #66899