[llvm] [AMDGPU] Always emit SI_KILL_I1_PSEUDO for uniform floating point branches. (PR #124028)

[Konstantina Mitropoulou via llvm-commits]

================
@@ -0,0 +1,58 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -stop-after=amdgpu-isel < %s 2>&1 | FileCheck %s
+
+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--amdpal"
+
+define amdgpu_ps void @_amdgpu_ps_main() {
+  ; CHECK-LABEL: name: _amdgpu_ps_main
+  ; CHECK: bb.0..entry:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
+  ; CHECK-NEXT:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_]], %subreg.sub1, [[S_MOV_B32_]], %subreg.sub2, [[S_MOV_B32_]], %subreg.sub3
+  ; CHECK-NEXT:   [[S_BUFFER_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM killed [[REG_SEQUENCE]], 0, 0 :: (dereferenceable invariant load (s32))
+  ; CHECK-NEXT:   [[S_MOV_B32_1:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
+  ; CHECK-NEXT:   nofpexcept S_CMP_NLT_F32 [[S_BUFFER_LOAD_DWORD_IMM]], [[S_MOV_B32_1]], implicit-def $scc, implicit $mode
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:sreg_32_xm0_xexec = COPY $scc
+  ; CHECK-NEXT:   SI_KILL_I1_PSEUDO killed [[COPY]], 0, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+  ; CHECK-NEXT:   nofpexcept S_CMP_LT_F32 [[S_BUFFER_LOAD_DWORD_IMM]], [[S_MOV_B32_1]], implicit-def $scc, implicit $mode
+  ; CHECK-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit $scc
+  ; CHECK-NEXT:   S_BRANCH %bb.1
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1 (%ir-block.3):
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2 (%ir-block.5):
+  ; CHECK-NEXT:   S_ENDPGM 0
+.entry:
+  %0 = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> zeroinitializer, i32 0, i32 0)
+  %1 = bitcast i32 %0 to float
+  %2 = fcmp uge float %1, 0.000000e+00
+  call void @llvm.amdgcn.kill(i1 %2)
+  br i1 %2, label %3, label %8
+
+3:                                                ; preds = %.entry
+  %4 = call i64 @llvm.amdgcn.s.getpc()
+  %5 = and i64 %4, 1
+  %6 = inttoptr i64 %5 to ptr addrspace(4)
+  %7 = getelementptr i8, ptr addrspace(4) %6, i64 32
+  br label %8
+
+8:                                                ; preds = %3, %.entry
+  ret void
+}
+
+; Function Attrs: nocallback nofree nounwind
+declare void @llvm.amdgcn.kill(i1) #0
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare noundef i64 @llvm.amdgcn.s.getpc() #1
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)
+declare i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32>, i32, i32 immarg) #2
+
+attributes #0 = { nocallback nofree nounwind }
+attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+attributes #2 = { nocallback nofree nosync nounwind willreturn memory(none) }
+
----------------
kmitropoulou wrote:

Done


https://github.com/llvm/llvm-project/pull/124028