[InstCombine] add tests for diff-of-sums; NFC

diff --git a/llvm/test/Transforms/InstCombine/vector-reductions.ll b/llvm/test/Transforms/InstCombine/vector-reductions.ll
new file mode 100644 (file)
index 0000000..5eac0e0
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/vector-reductions.ll
@@ -0,0 +1,75 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+declare float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float, <4 x float>)
+declare float @llvm.experimental.vector.reduce.v2.fadd.f32.v8f32(float, <8 x float>)
+declare void @use_f32(float)
+
+define float @diff_of_sums_v4f32(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
+; CHECK-LABEL: @diff_of_sums_v4f32(
+; CHECK-NEXT:    [[R0:%.*]] = call float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
+; CHECK-NEXT:    [[R1:%.*]] = call float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
+; CHECK-NEXT:    [[R:%.*]] = fsub reassoc nsz float [[R0]], [[R1]]
+; CHECK-NEXT:    ret float [[R]]
+;
+  %r0 = call float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a0, <4 x float> %v0)
+  %r1 = call float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a1, <4 x float> %v1)
+  %r = fsub reassoc nsz float %r0, %r1
+  ret float %r
+}
+
+define float @diff_of_sums_v4f32_fmf(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
+; CHECK-LABEL: @diff_of_sums_v4f32_fmf(
+; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
+; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
+; CHECK-NEXT:    [[R:%.*]] = fsub nnan ninf nsz float [[R0]], [[R1]]
+; CHECK-NEXT:    ret float [[R]]
+;
+  %r0 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a0, <4 x float> %v0)
+  %r1 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a1, <4 x float> %v1)
+  %r = fsub ninf nnan nsz float %r0, %r1
+  ret float %r
+}
+
+define float @diff_of_sums_extra_use1(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
+; CHECK-LABEL: @diff_of_sums_extra_use1(
+; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
+; CHECK-NEXT:    call void @use_f32(float [[R0]])
+; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
+; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
+; CHECK-NEXT:    ret float [[R]]
+;
+  %r0 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a0, <4 x float> %v0)
+  call void @use_f32(float %r0)
+  %r1 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a1, <4 x float> %v1)
+  %r = fsub fast float %r0, %r1
+  ret float %r
+}
+
+define float @diff_of_sums_extra_use2(float %a0, <4 x float> %v0, float %a1, <4 x float> %v1) {
+; CHECK-LABEL: @diff_of_sums_extra_use2(
+; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
+; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A1:%.*]], <4 x float> [[V1:%.*]])
+; CHECK-NEXT:    call void @use_f32(float [[R1]])
+; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
+; CHECK-NEXT:    ret float [[R]]
+;
+  %r0 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a0, <4 x float> %v0)
+  %r1 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a1, <4 x float> %v1)
+  call void @use_f32(float %r1)
+  %r = fsub fast float %r0, %r1
+  ret float %r
+}
+
+define float @diff_of_sums_type_mismatch(float %a0, <4 x float> %v0, float %a1, <8 x float> %v1) {
+; CHECK-LABEL: @diff_of_sums_type_mismatch(
+; CHECK-NEXT:    [[R0:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float [[A0:%.*]], <4 x float> [[V0:%.*]])
+; CHECK-NEXT:    [[R1:%.*]] = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v8f32(float [[A1:%.*]], <8 x float> [[V1:%.*]])
+; CHECK-NEXT:    [[R:%.*]] = fsub fast float [[R0]], [[R1]]
+; CHECK-NEXT:    ret float [[R]]
+;
+  %r0 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v4f32(float %a0, <4 x float> %v0)
+  %r1 = call fast float @llvm.experimental.vector.reduce.v2.fadd.f32.v8f32(float %a1, <8 x float> %v1)
+  %r = fsub fast float %r0, %r1
+  ret float %r
+}