; There is a mismatch between the intrinsic and the actual instruction.
; The actual instruction has a partial update of dest, while the intrinsic
; passes through the upper FP values. Here, we make sure the source and
-; destination of rsqrtss are the same.
-define void @t1(<4 x float> %a) nounwind uwtable ssp {
+; destination of each scalar unary op are the same.
+
+define void @rsqrtss(<4 x float> %a) nounwind uwtable ssp {
entry:
-; CHECK: t1:
+; CHECK-LABEL: rsqrtss:
; CHECK: rsqrtss %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movshdup
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movap
+; CHECK-NEXT: jmp
+
%0 = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %a) nounwind
%a.addr.0.extract = extractelement <4 x float> %0, i32 0
%conv = fpext float %a.addr.0.extract to double
declare void @callee(double, double)
declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone
-define void @t2(<4 x float> %a) nounwind uwtable ssp {
+define void @rcpss(<4 x float> %a) nounwind uwtable ssp {
entry:
-; CHECK: t2:
+; CHECK-LABEL: rcpss:
; CHECK: rcpss %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movshdup
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movap
+; CHECK-NEXT: jmp
+
%0 = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %a) nounwind
%a.addr.0.extract = extractelement <4 x float> %0, i32 0
%conv = fpext float %a.addr.0.extract to double
ret void
}
declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone
+
+define void @sqrtss(<4 x float> %a) nounwind uwtable ssp {
+entry:
+; CHECK-LABEL: sqrtss:
+; CHECK: sqrtss %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movshdup
+; CHECK-NEXT: cvtss2sd %xmm0
+; CHECK-NEXT: movap
+; CHECK-NEXT: jmp
+
+ %0 = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %a) nounwind
+ %a.addr.0.extract = extractelement <4 x float> %0, i32 0
+ %conv = fpext float %a.addr.0.extract to double
+ %a.addr.4.extract = extractelement <4 x float> %0, i32 1
+ %conv3 = fpext float %a.addr.4.extract to double
+ tail call void @callee(double %conv, double %conv3) nounwind
+ ret void
+}
+declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
+
+define void @sqrtsd(<2 x double> %a) nounwind uwtable ssp {
+entry:
+; CHECK-LABEL: sqrtsd:
+; CHECK: sqrtsd %xmm0, %xmm0
+; CHECK-NEXT: cvtsd2ss %xmm0
+; CHECK-NEXT: shufpd
+; CHECK-NEXT: cvtsd2ss %xmm0
+; CHECK-NEXT: movap
+; CHECK-NEXT: jmp
+
+ %0 = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> %a) nounwind
+ %a0 = extractelement <2 x double> %0, i32 0
+ %conv = fptrunc double %a0 to float
+ %a1 = extractelement <2 x double> %0, i32 1
+ %conv3 = fptrunc double %a1 to float
+ tail call void @callee2(float %conv, float %conv3) nounwind
+ ret void
+}
+
+declare void @callee2(float, float)
+declare <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double>) nounwind readnone
+
+define <2 x double> @load_fold_cvtss2sd_int(<4 x float> *%a) {
+; CHECK-LABEL: load_fold_cvtss2sd_int:
+; CHECK: movaps (%rdi), %xmm1
+; CHECK-NEXT: xorps %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd %xmm1, %xmm0
+; CHECK-NEXT: retq
+ %ld = load <4 x float>, <4 x float> *%a
+ %x = call <2 x double> @llvm.x86.sse2.cvtss2sd(<2 x double> <double 0x0, double 0x0>, <4 x float> %ld)
+ ret <2 x double> %x
+}
+
+define <2 x double> @load_fold_cvtss2sd_int_optsize(<4 x float> *%a) optsize {
+; CHECK-LABEL: load_fold_cvtss2sd_int_optsize:
+; CHECK: xorps %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd (%rdi), %xmm0
+; CHECK-NEXT: retq
+ %ld = load <4 x float>, <4 x float> *%a
+ %x = call <2 x double> @llvm.x86.sse2.cvtss2sd(<2 x double> <double 0x0, double 0x0>, <4 x float> %ld)
+ ret <2 x double> %x
+}
+
+define <2 x double> @load_fold_cvtss2sd_int_minsize(<4 x float> *%a) minsize {
+; CHECK-LABEL: load_fold_cvtss2sd_int_minsize:
+; CHECK: xorps %xmm0, %xmm0
+; CHECK-NEXT: cvtss2sd (%rdi), %xmm0
+; CHECK-NEXT: retq
+ %ld = load <4 x float>, <4 x float> *%a
+ %x = call <2 x double> @llvm.x86.sse2.cvtss2sd(<2 x double> <double 0x0, double 0x0>, <4 x float> %ld)
+ ret <2 x double> %x
+}
+
+declare <2 x double> @llvm.x86.sse2.cvtss2sd(<2 x double>, <4 x float>) nounwind readnone
+