; RUN: llc < %s -march=x86 -mcpu=pentium4 -mattr=+sse2 | FileCheck %s -check-prefix=SSE2
; RUN: llc < %s -march=x86 -mcpu=pentium4 -mattr=+sse3 | FileCheck %s -check-prefix=SSE3
+; RUN: llc < %s -march=x86-64 -mattr=+avx | FileCheck %s -check-prefix=AVX
define void @test_v4sf(<4 x float>* %P, <4 x float>* %Q, float %X) nounwind {
%tmp = insertelement <4 x float> zeroinitializer, float %X, i32 0 ; <<4 x float>> [#uses=1]
; Fold extract of a load into the load's address computation. This avoids spilling to the stack.
define <4 x float> @load_extract_splat(<4 x float>* nocapture readonly %ptr, i64 %i, i64 %j) nounwind {
+ %1 = getelementptr inbounds <4 x float>* %ptr, i64 %i
+ %2 = load <4 x float>* %1, align 16
+ %3 = trunc i64 %j to i32
+ %4 = extractelement <4 x float> %2, i32 %3
+ %5 = insertelement <4 x float> undef, float %4, i32 0
+ %6 = insertelement <4 x float> %5, float %4, i32 1
+ %7 = insertelement <4 x float> %6, float %4, i32 2
+ %8 = insertelement <4 x float> %7, float %4, i32 3
+ ret <4 x float> %8
+
+; AVX-LABEL: load_extract_splat
+; AVX-NOT: rsp
+; AVX: vbroadcastss
+}
+
+; Fold extract of a load into the load's address computation. This avoids spilling to the stack.
+define <4 x float> @load_extract_splat1(<4 x float>* nocapture readonly %ptr, i64 %i, i64 %j) nounwind {
%1 = getelementptr inbounds <4 x float>* %ptr, i64 %i
%2 = load <4 x float>* %1, align 16
%3 = extractelement <4 x float> %2, i64 %j
%7 = insertelement <4 x float> %6, float %3, i32 3
ret <4 x float> %7
-; AVX-LABEL: load_extract_splat
+; AVX-LABEL: load_extract_splat1
; AVX-NOT: movs
; AVX: vbroadcastss
}