[llvm] [SIFoldOperands] Folding immediate into a copy invalidates candidates in the fold list (PR #148187)

[Matt Arsenault via llvm-commits]

Juan Manuel Martinez =?utf-8?q?Caamaño?= <juamarti at amd.com>,
Juan Manuel Martinez =?utf-8?q?Caamaño?= <juamarti at amd.com>,
Juan Manuel Martinez =?utf-8?q?Caamaño?= <juamarti at amd.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/148187 at github.com>


================
@@ -0,0 +1,60 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -O3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a %s -o - | FileCheck %s
+
+define amdgpu_kernel void @kernel() {
+; CHECK-LABEL: kernel:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    ds_read_b64 v[2:3], v0
+; CHECK-NEXT:    ds_write_b32 v0, v0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(1)
+; CHECK-NEXT:    v_cmp_ne_u32_e32 vcc, 2, v2
+; CHECK-NEXT:    s_cbranch_vccnz .LBB0_3
+; CHECK-NEXT:  ; %bb.1: ; %land.rhs49
+; CHECK-NEXT:    ds_read_b64 v[0:1], v0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
+; CHECK-NEXT:    s_cbranch_vccnz .LBB0_3
+; CHECK-NEXT:  ; %bb.2: ; %land.rhs57
+; CHECK-NEXT:    s_mov_b32 s0, 0
+; CHECK-NEXT:    s_mov_b32 s1, s0
+; CHECK-NEXT:    v_mov_b32_e32 v2, 0
+; CHECK-NEXT:    v_pk_mov_b32 v[0:1], s[0:1], s[0:1] op_sel:[0,1]
+; CHECK-NEXT:    s_cmp_lg_u32 s0, 0
+; CHECK-NEXT:    ds_write_b64 v2, v[0:1]
+; CHECK-NEXT:  .LBB0_3: ; %land.end59
+; CHECK-NEXT:    s_endpgm
+entry:
+  %0 = load <2 x i32>, ptr addrspace(3) null, align 8
+  %vecext.i55 = extractelement <2 x i32> %0, i64 0
+  %cmp3.i57 = icmp eq i32 %vecext.i55, 2
+  store i32 0, ptr addrspace(3) null, align 8
+  br i1 %cmp3.i57, label %land.rhs49, label %land.end59
+
+land.rhs49:                                       ; preds = %entry
+  %1 = load <2 x i32>, ptr addrspace(3) null, align 8
+  %vecext.i67 = extractelement <2 x i32> %1, i64 0
+  %cmp3.i69 = icmp eq i32 %vecext.i67, 1
+  br i1 %cmp3.i69, label %land.rhs57, label %land.end59
+
+land.rhs57:                                       ; preds = %land.rhs49
+  %rem.i.i.i = srem <2 x i32> %0, %1
+  %ref.tmp.sroa.0.0.vec.extract.i.i = extractelement <2 x i32> %rem.i.i.i, i64 0
+  store i32 %ref.tmp.sroa.0.0.vec.extract.i.i, ptr addrspace(3) null, align 8
+  store i32 %ref.tmp.sroa.0.0.vec.extract.i.i, ptr addrspace(3) getelementptr inbounds nuw (i8, ptr addrspace(3) null, i32 4), align 4
+  %2 = load <2 x i32>, ptr addrspace(3) null, align 8
----------------
arsenm wrote:

Use names values 

https://github.com/llvm/llvm-project/pull/148187