[llvm] [AMDGPU] si-peephole-sdwa: Fix cndmask vcc use for wave32 (PR #139541)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Frederik Harwath (frederik-h)

<details>
<summary>Changes</summary>

Before V_CNDMASK_B32_e64 gets converted to SDWA form, a conversion to V_CNDMASK_B32_e32 occurs.
The vcc use of this instruction must be fixed into a vcc_lo use for wave32. This fix only happens after the final
conversion to the SDWA form. This led to a compiler error in situations where the conversion to SDWA aborts.

Make sure that the vcc-fix gets applied even if the SDWA conversion is not completed.

---
Full diff: https://github.com/llvm/llvm-project/pull/139541.diff


2 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp (+1) 
- (added) llvm/test/CodeGen/AMDGPU/sdwa-peephole-cndmask-fail.ll (+51) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
index 8eb1d7253cd48..bd8baaaa3df20 100644
--- a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
+++ b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
@@ -1105,6 +1105,7 @@ void SIPeepholeSDWA::convertVcndmaskToVOP2(MachineInstr &MI,
                        .add(*TII->getNamedOperand(MI, AMDGPU::OpName::src0))
                        .add(*TII->getNamedOperand(MI, AMDGPU::OpName::src1))
                        .setMIFlags(MI.getFlags());
+  TII->fixImplicitOperands(*Converted);
   LLVM_DEBUG(dbgs() << "Converted to VOP2: " << *Converted);
   (void)Converted;
   MI.eraseFromParent();
diff --git a/llvm/test/CodeGen/AMDGPU/sdwa-peephole-cndmask-fail.ll b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-cndmask-fail.ll
new file mode 100644
index 0000000000000..9ab5a31b52441
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-cndmask-fail.ll
@@ -0,0 +1,51 @@
+; RUN: llc %s -march=amdgcn -mcpu=gfx1030 -o - 2>&1 | FileCheck %s
+
+; In this test, V_CNDMASK_B32_e64 gets converted to V_CNDMASK_B32_e32,
+; but the expected conversion to SDWA does not occur.  This led to a
+; compilation error, because the use of $vcc in the resulting
+; instruction must be fixed to $vcc_lo for wave32 which only happened
+; after the full conversion to SDWA.
+
+
+; CHECK-NOT: {{.*}}V_CNDMASK_B32_e32{{.*}}$vcc
+; CHECK-NOT: {{.*}}Bad machine code: Virtual register defs don't dominate all uses
+; CHECK: {{.*}}v_cndmask_b32_e32{{.*}}vcc_lo
+
+; ModuleID = 'test.ll'
+source_filename = "test.ll"
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
+target triple = "amdgcn-amd-amdhsa"
+
+define amdgpu_kernel void @quux(i32 %arg, i1 %arg1, i1 %arg2) #0 {
+bb:
+  br i1 %arg1, label %bb9, label %bb3
+
+bb3:                                              ; preds = %bb
+  %call = tail call i32 @llvm.amdgcn.workitem.id.x()
+  %mul = mul i32 %call, 5
+  %zext = zext i32 %mul to i64
+  %getelementptr = getelementptr i8, ptr addrspace(1) null, i64 %zext
+  %getelementptr4 = getelementptr i8, ptr addrspace(1) %getelementptr, i64 4
+  %load = load i8, ptr addrspace(1) %getelementptr4, align 1
+  %getelementptr5 = getelementptr i8, ptr addrspace(1) %getelementptr, i64 3
+  %load6 = load i8, ptr addrspace(1) %getelementptr5, align 1
+  %insertelement = insertelement <5 x i8> poison, i8 %load, i64 4
+  %select = select i1 %arg2, <5 x i8> %insertelement, <5 x i8> <i8 poison, i8 poison, i8 poison, i8 poison, i8 0>
+  %insertelement7 = insertelement <5 x i8> %select, i8 %load6, i64 0
+  %icmp = icmp ult i32 0, %arg
+  %select8 = select i1 %icmp, <5 x i8> zeroinitializer, <5 x i8> %insertelement7
+  %shufflevector = shufflevector <5 x i8> zeroinitializer, <5 x i8> %select8, <5 x i32> <i32 0, i32 1, i32 7, i32 8, i32 9>
+  br label %bb9
+
+bb9:                                              ; preds = %bb3, %bb
+  %phi = phi <5 x i8> [ %shufflevector, %bb3 ], [ zeroinitializer, %bb ]
+  %extractelement = extractelement <5 x i8> %phi, i64 0
+  store i8 %extractelement, ptr addrspace(1) null, align 1
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x() #1
+
+attributes #0 = { "target-cpu"="gfx1030" }
+attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1030" }

``````````

</details>


https://github.com/llvm/llvm-project/pull/139541