[x86] add tests for PR40243; NFC
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350646 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/test/CodeGen/X86/haddsub-undef.ll b/test/CodeGen/X86/haddsub-undef.ll
index 7ba753b..a5c5991 100644
--- a/test/CodeGen/X86/haddsub-undef.ll
+++ b/test/CodeGen/X86/haddsub-undef.ll
@@ -926,8 +926,6 @@
ret <4 x float> %r
}
-; FIXME: Miscompiles with any AVX.
-
define <8 x float> @v16f32_inputs_v8f32_output_4567(<16 x float> %a, <16 x float> %b) {
; SSE-LABEL: v16f32_inputs_v8f32_output_4567:
; SSE: # %bb.0:
@@ -967,3 +965,24 @@
ret <8 x float> %r
}
+define <8 x float> @PR40243(<8 x float> %a, <8 x float> %b) {
+; SSE-LABEL: PR40243:
+; SSE: # %bb.0:
+; SSE-NEXT: haddps %xmm3, %xmm1
+; SSE-NEXT: retq
+;
+; AVX-LABEL: PR40243:
+; AVX: # %bb.0:
+; AVX-NEXT: vhaddps %ymm0, %ymm0, %ymm0
+; AVX-NEXT: retq
+ %a4 = extractelement <8 x float> %a, i32 4
+ %a5 = extractelement <8 x float> %a, i32 5
+ %add4 = fadd float %a4, %a5
+ %b6 = extractelement <8 x float> %b, i32 6
+ %b7 = extractelement <8 x float> %b, i32 7
+ %add7 = fadd float %b6, %b7
+ %r4 = insertelement <8 x float> undef, float %add4, i32 4
+ %r = insertelement <8 x float> %r4, float %add7, i32 7
+ ret <8 x float> %r
+}
+
diff --git a/test/CodeGen/X86/phaddsub-undef.ll b/test/CodeGen/X86/phaddsub-undef.ll
index 61f4ab4..a2212ed 100644
--- a/test/CodeGen/X86/phaddsub-undef.ll
+++ b/test/CodeGen/X86/phaddsub-undef.ll
@@ -93,6 +93,36 @@
ret <8 x i32> %vecinit5
}
+define <8 x i32> @PR40243_alt(<8 x i32> %a, <8 x i32> %b) {
+; SSE-LABEL: PR40243_alt:
+; SSE: # %bb.0:
+; SSE-NEXT: phaddd %xmm3, %xmm1
+; SSE-NEXT: retq
+;
+; AVX1-LABEL: PR40243_alt:
+; AVX1: # %bb.0:
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: PR40243_alt:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vphaddd %ymm0, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512-LABEL: PR40243_alt:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vphaddd %ymm0, %ymm0, %ymm0
+; AVX512-NEXT: retq
+ %a4 = extractelement <8 x i32> %a, i32 4
+ %a5 = extractelement <8 x i32> %a, i32 5
+ %add4 = add i32 %a4, %a5
+ %b6 = extractelement <8 x i32> %b, i32 6
+ %b7 = extractelement <8 x i32> %b, i32 7
+ %add7 = add i32 %b6, %b7
+ %r4 = insertelement <8 x i32> undef, i32 %add4, i32 4
+ %r = insertelement <8 x i32> %r4, i32 %add7, i32 7
+ ret <8 x i32> %r
+}
+
define <8 x i32> @test16_undef(<8 x i32> %a, <8 x i32> %b) {
; SSE-LABEL: test16_undef:
; SSE: # %bb.0:
