From 8e6d41aeb26d43da802d01c5c7fdc4e6b02ef93a Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Tue, 2 Apr 2019 20:24:06 +0000
Subject: [PATCH] [x86] add more tests for FP scalarization; NFC

llvm-svn: 357523
---
 llvm/test/CodeGen/X86/scalarize-fp.ll | 184 ++++++++++++++++++++++++++++++++++
 1 file changed, 184 insertions(+)

diff --git a/llvm/test/CodeGen/X86/scalarize-fp.ll b/llvm/test/CodeGen/X86/scalarize-fp.ll
index 5cf526d..5bc574b 100644
--- a/llvm/test/CodeGen/X86/scalarize-fp.ll
+++ b/llvm/test/CodeGen/X86/scalarize-fp.ll
@@ -376,3 +376,187 @@ define <4 x double> @load_fdiv_op0_constant_v4f64(double* %p) nounwind {
   ret <4 x double> %b
 }
 
+define <2 x double> @fadd_splat_splat_v2f64(<2 x double> %vx, <2 x double> %vy) {
+; SSE-LABEL: fadd_splat_splat_v2f64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
+; SSE-NEXT:    addpd %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: fadd_splat_splat_v2f64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
+; AVX-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %splatx = shufflevector <2 x double> %vx, <2 x double> undef, <2 x i32> zeroinitializer
+  %splaty = shufflevector <2 x double> %vy, <2 x double> undef, <2 x i32> zeroinitializer
+  %r = fadd <2 x double> %splatx, %splaty
+  ret <2 x double> %r
+}
+
+define <4 x double> @fsub_splat_splat_v4f64(double %x, double %y) {
+; SSE-LABEL: fsub_splat_splat_v4f64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0,0]
+; SSE-NEXT:    subpd %xmm1, %xmm0
+; SSE-NEXT:    movapd %xmm0, %xmm1
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: fsub_splat_splat_v4f64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm1
+; AVX-NEXT:    vsubpd %ymm1, %ymm0, %ymm0
+; AVX-NEXT:    retq
+  %vx = insertelement <4 x double> undef, double %x, i32 0
+  %vy = insertelement <4 x double> undef, double %y, i32 0
+  %splatx = shufflevector <4 x double> %vx, <4 x double> undef, <4 x i32> zeroinitializer
+  %splaty = shufflevector <4 x double> %vy, <4 x double> undef, <4 x i32> zeroinitializer
+  %r = fsub <4 x double> %splatx, %splaty
+  ret <4 x double> %r
+}
+
+define <4 x float> @fmul_splat_splat_v4f32(<4 x float> %vx, <4 x float> %vy) {
+; SSE-LABEL: fmul_splat_splat_v4f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    mulps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: fmul_splat_splat_v4f32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; AVX-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX-NEXT:    vmulps %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %splatx = shufflevector <4 x float> %vx, <4 x float> undef, <4 x i32> zeroinitializer
+  %splaty = shufflevector <4 x float> %vy, <4 x float> undef, <4 x i32> zeroinitializer
+  %r = fmul fast <4 x float> %splatx, %splaty
+  ret <4 x float> %r
+}
+
+define <8 x float> @fdiv_splat_splat_v8f32(<8 x float> %vx, <8 x float> %vy) {
+; SSE-LABEL: fdiv_splat_splat_v8f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    rcpps %xmm2, %xmm3
+; SSE-NEXT:    mulps %xmm3, %xmm2
+; SSE-NEXT:    movaps {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
+; SSE-NEXT:    subps %xmm2, %xmm1
+; SSE-NEXT:    mulps %xmm3, %xmm1
+; SSE-NEXT:    addps %xmm3, %xmm1
+; SSE-NEXT:    mulps %xmm0, %xmm1
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: fdiv_splat_splat_v8f32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm1
+; AVX-NEXT:    vrcpps %ymm1, %ymm2
+; AVX-NEXT:    vmulps %ymm2, %ymm1, %ymm1
+; AVX-NEXT:    vmovaps {{.*#+}} ymm3 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
+; AVX-NEXT:    vsubps %ymm1, %ymm3, %ymm1
+; AVX-NEXT:    vmulps %ymm1, %ymm2, %ymm1
+; AVX-NEXT:    vaddps %ymm1, %ymm2, %ymm1
+; AVX-NEXT:    vmulps %ymm1, %ymm0, %ymm0
+; AVX-NEXT:    retq
+  %splatx = shufflevector <8 x float> %vx, <8 x float> undef, <8 x i32> zeroinitializer
+  %splaty = shufflevector <8 x float> %vy, <8 x float> undef, <8 x i32> zeroinitializer
+  %r = fdiv fast <8 x float> %splatx, %splaty
+  ret <8 x float> %r
+}
+
+define <2 x double> @splat0_fadd_v2f64(<2 x double> %vx, <2 x double> %vy) {
+; SSE-LABEL: splat0_fadd_v2f64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    addpd %xmm1, %xmm0
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: splat0_fadd_v2f64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX-NEXT:    retq
+  %b = fadd <2 x double> %vx, %vy
+  %r = shufflevector <2 x double> %b, <2 x double> undef, <2 x i32> zeroinitializer
+  ret <2 x double> %r
+}
+
+define <4 x double> @splat0_fsub_v4f64(double %x, double %y) {
+; SSE-LABEL: splat0_fsub_v4f64:
+; SSE:       # %bb.0:
+; SSE-NEXT:    subpd %xmm1, %xmm0
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
+; SSE-NEXT:    movapd %xmm0, %xmm1
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: splat0_fsub_v4f64:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vsubpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX-NEXT:    retq
+  %vx = insertelement <4 x double> undef, double %x, i32 0
+  %vy = insertelement <4 x double> undef, double %y, i32 0
+  %b = fsub <4 x double> %vx, %vy
+  %r = shufflevector <4 x double> %b, <4 x double> undef, <4 x i32> zeroinitializer
+  ret <4 x double> %r
+}
+
+define <4 x float> @splat0_fmul_v4f32(<4 x float> %vx, <4 x float> %vy) {
+; SSE-LABEL: splat0_fmul_v4f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    mulps %xmm1, %xmm0
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: splat0_fmul_v4f32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmulps %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; AVX-NEXT:    retq
+  %b = fmul fast <4 x float> %vx, %vy
+  %r = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> zeroinitializer
+  ret <4 x float> %r
+}
+
+define <8 x float> @splat0_fdiv_v8f32(<8 x float> %vx, <8 x float> %vy) {
+; SSE-LABEL: splat0_fdiv_v8f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    rcpps %xmm2, %xmm3
+; SSE-NEXT:    mulps %xmm3, %xmm2
+; SSE-NEXT:    movaps {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
+; SSE-NEXT:    subps %xmm2, %xmm1
+; SSE-NEXT:    mulps %xmm3, %xmm1
+; SSE-NEXT:    addps %xmm3, %xmm1
+; SSE-NEXT:    mulps %xmm0, %xmm1
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: splat0_fdiv_v8f32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vrcpps %ymm1, %ymm2
+; AVX-NEXT:    vmulps %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vmovaps {{.*#+}} xmm3 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
+; AVX-NEXT:    vsubps %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vmulps %xmm1, %xmm2, %xmm1
+; AVX-NEXT:    vaddps %xmm1, %xmm2, %xmm1
+; AVX-NEXT:    vmulps %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX-NEXT:    retq
+  %b = fdiv fast <8 x float> %vx, %vy
+  %r = shufflevector <8 x float> %b, <8 x float> undef, <8 x i32> zeroinitializer
+  ret <8 x float> %r
+}
-- 
2.7.4