diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll | 1158 |
1 files changed, 1158 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll new file mode 100644 index 00000000000..46349931bf1 --- /dev/null +++ b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll @@ -0,0 +1,1158 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mcpu=skx %s -o - | FileCheck %s + +define <4 x double> @test_double_to_4(double %s) { +; CHECK-LABEL: test_double_to_4: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %res = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mask0(double %s, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $12, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovapd %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 1, i1 1>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mask0(double %s) { +; CHECK-LABEL: test_masked_z_double_to_4_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $12, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 1, i1 1>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mask1(double %s, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovapd %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mask1(double %s) { +; CHECK-LABEL: test_masked_z_double_to_4_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mask2(double %s, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovapd %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mask2(double %s) { +; CHECK-LABEL: test_masked_z_double_to_4_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mask3(double %s, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $3, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovapd %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 0>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mask3(double %s) { +; CHECK-LABEL: test_masked_z_double_to_4_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $3, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 0>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <8 x double> @test_double_to_8(double %s) { +; CHECK-LABEL: test_double_to_8: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %res = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mask0(double %s, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-126, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovapd %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mask0(double %s) { +; CHECK-LABEL: test_masked_z_double_to_8_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-126, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mask1(double %s, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $103, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovapd %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mask1(double %s) { +; CHECK-LABEL: test_masked_z_double_to_8_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $103, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mask2(double %s, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-56, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovapd %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mask2(double %s) { +; CHECK-LABEL: test_masked_z_double_to_8_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-56, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mask3(double %s, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $78, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovapd %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 0>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mask3(double %s) { +; CHECK-LABEL: test_masked_z_double_to_8_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $78, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 0>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <4 x float> @test_float_to_4(float %s) { +; CHECK-LABEL: test_float_to_4: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mask0(float %s, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $7, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm1 {%k1} +; CHECK-NEXT: vmovaps %xmm1, %xmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mask0(float %s) { +; CHECK-LABEL: test_masked_z_float_to_4_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $7, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mask1(float %s, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $8, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm1 {%k1} +; CHECK-NEXT: vmovaps %xmm1, %xmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 0, i1 1>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mask1(float %s) { +; CHECK-LABEL: test_masked_z_float_to_4_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $8, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 0, i1 1>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mask2(float %s, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $11, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm1 {%k1} +; CHECK-NEXT: vmovaps %xmm1, %xmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 1>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mask2(float %s) { +; CHECK-LABEL: test_masked_z_float_to_4_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $11, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 1>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mask3(float %s, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm1 {%k1} +; CHECK-NEXT: vmovaps %xmm1, %xmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mask3(float %s) { +; CHECK-LABEL: test_masked_z_float_to_4_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <8 x float> @test_float_to_8(float %s) { +; CHECK-LABEL: test_float_to_8: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mask0(float %s, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $72, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovaps %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mask0(float %s) { +; CHECK-LABEL: test_masked_z_float_to_8_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $72, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mask1(float %s, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-64, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovaps %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mask1(float %s) { +; CHECK-LABEL: test_masked_z_float_to_8_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-64, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mask2(float %s, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-98, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovaps %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mask2(float %s) { +; CHECK-LABEL: test_masked_z_float_to_8_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-98, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mask3(float %s, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $64, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm1 {%k1} +; CHECK-NEXT: vmovaps %ymm1, %ymm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mask3(float %s) { +; CHECK-LABEL: test_masked_z_float_to_8_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $64, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <16 x float> @test_float_to_16(float %s) { +; CHECK-LABEL: test_float_to_16: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss %xmm0, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mask0(float %s, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-5916, %ax # imm = 0xE8E4 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovaps %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mask0(float %s) { +; CHECK-LABEL: test_masked_z_float_to_16_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-5916, %ax # imm = 0xE8E4 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mask1(float %s, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-1130, %ax # imm = 0xFB96 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovaps %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 1, i1 1, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mask1(float %s) { +; CHECK-LABEL: test_masked_z_float_to_16_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-1130, %ax # imm = 0xFB96 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 1, i1 1, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mask2(float %s, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-12439, %ax # imm = 0xCF69 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovaps %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mask2(float %s) { +; CHECK-LABEL: test_masked_z_float_to_16_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-12439, %ax # imm = 0xCF69 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mask3(float %s, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-6413, %ax # imm = 0xE6F3 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm1 {%k1} +; CHECK-NEXT: vmovaps %zmm1, %zmm0 +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mask3(float %s) { +; CHECK-LABEL: test_masked_z_float_to_16_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-6413, %ax # imm = 0xE6F3 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss %xmm0, %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <4 x double> @test_double_to_4_mem(double* %p) { +; CHECK-LABEL: test_double_to_4_mem: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %res = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mem_mask0(double* %p, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $5, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 0, i1 1, i1 0>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mem_mask0(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_4_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $5, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 0, i1 1, i1 0>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mem_mask1(double* %p, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mem_mask1(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_4_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mem_mask2(double* %p, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $11, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mem_mask2(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_4_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $11, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 1, i1 0, i1 1>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <4 x double> @test_masked_double_to_4_mem_mask3(double* %p, <4 x double> %default) { +; CHECK-LABEL: test_masked_double_to_4_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $8, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 0, i1 1>, <4 x double> %shuf, <4 x double> %default + ret <4 x double> %res +} + +define <4 x double> @test_masked_z_double_to_4_mem_mask3(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_4_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $8, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 0, i1 0, i1 1>, <4 x double> %shuf, <4 x double> zeroinitializer + ret <4 x double> %res +} +define <8 x double> @test_double_to_8_mem(double* %p) { +; CHECK-LABEL: test_double_to_8_mem: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %res = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mem_mask0(double* %p, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $120, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mem_mask0(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_8_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $120, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 0, i1 1, i1 1, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mem_mask1(double* %p, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $26, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mem_mask1(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_8_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $26, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mem_mask2(double* %p, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $111, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mem_mask2(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_8_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $111, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 0>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <8 x double> @test_masked_double_to_8_mem_mask3(double* %p, <8 x double> %default) { +; CHECK-LABEL: test_masked_double_to_8_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-100, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1>, <8 x double> %shuf, <8 x double> %default + ret <8 x double> %res +} + +define <8 x double> @test_masked_z_double_to_8_mem_mask3(double* %p) { +; CHECK-LABEL: test_masked_z_double_to_8_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-100, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load double, double* %p + %vec = insertelement <2 x double> undef, double %s, i32 0 + %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 0, i1 1>, <8 x double> %shuf, <8 x double> zeroinitializer + ret <8 x double> %res +} +define <4 x float> @test_float_to_4_mem(float* %p) { +; CHECK-LABEL: test_float_to_4_mem: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mem_mask0(float* %p, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $13, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 0, i1 1, i1 1>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mem_mask0(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_4_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $13, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 1, i1 0, i1 1, i1 1>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mem_mask1(float* %p, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $14, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 1>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mem_mask1(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_4_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $14, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 1>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mem_mask2(float* %p, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mem_mask2(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_4_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $6, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 1, i1 0>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <4 x float> @test_masked_float_to_4_mem_mask3(float* %p, <4 x float> %default) { +; CHECK-LABEL: test_masked_float_to_4_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x float> %shuf, <4 x float> %default + ret <4 x float> %res +} + +define <4 x float> @test_masked_z_float_to_4_mem_mask3(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_4_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $10, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %xmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + %res = select <4 x i1> <i1 0, i1 1, i1 0, i1 1>, <4 x float> %shuf, <4 x float> zeroinitializer + ret <4 x float> %res +} +define <8 x float> @test_float_to_8_mem(float* %p) { +; CHECK-LABEL: test_float_to_8_mem: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mem_mask0(float* %p, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $67, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mem_mask0(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_8_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movb $67, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 0>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mem_mask1(float* %p, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-51, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mem_mask1(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_8_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-51, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mem_mask2(float* %p, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-116, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0, i1 1>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mem_mask2(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_8_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movb $-116, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0, i1 1>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <8 x float> @test_masked_float_to_8_mem_mask3(float* %p, <8 x float> %default) { +; CHECK-LABEL: test_masked_float_to_8_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $4, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0>, <8 x float> %shuf, <8 x float> %default + ret <8 x float> %res +} + +define <8 x float> @test_masked_z_float_to_8_mem_mask3(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_8_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movb $4, %al +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %ymm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <8 x i1> <i1 0, i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0>, <8 x float> %shuf, <8 x float> zeroinitializer + ret <8 x float> %res +} +define <16 x float> @test_float_to_16_mem(float* %p) { +; CHECK-LABEL: test_float_to_16_mem: +; CHECK: # BB#0: +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %res = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mem_mask0(float* %p, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-18370, %ax # imm = 0xB83E +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 1, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mem_mask0(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_16_mem_mask0: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-18370, %ax # imm = 0xB83E +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 1, i1 1, i1 1, i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mem_mask1(float* %p, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movw $26137, %ax # imm = 0x6619 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mem_mask1(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_16_mem_mask1: +; CHECK: # BB#0: +; CHECK-NEXT: movw $26137, %ax # imm = 0x6619 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mem_mask2(float* %p, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-11480, %ax # imm = 0xD328 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mem_mask2(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_16_mem_mask2: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-11480, %ax # imm = 0xD328 +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 0, i1 0, i1 0, i1 1, i1 0, i1 1, i1 0, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 0, i1 1, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} +define <16 x float> @test_masked_float_to_16_mem_mask3(float* %p, <16 x float> %default) { +; CHECK-LABEL: test_masked_float_to_16_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-21749, %ax # imm = 0xAB0B +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 1, i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 0, i1 1, i1 0, i1 1, i1 0, i1 1>, <16 x float> %shuf, <16 x float> %default + ret <16 x float> %res +} + +define <16 x float> @test_masked_z_float_to_16_mem_mask3(float* %p) { +; CHECK-LABEL: test_masked_z_float_to_16_mem_mask3: +; CHECK: # BB#0: +; CHECK-NEXT: movw $-21749, %ax # imm = 0xAB0B +; CHECK-NEXT: kmovd %eax, %k1 +; CHECK-NEXT: vbroadcastss (%rdi), %zmm0 {%k1} {z} +; CHECK-NEXT: retq + %s = load float, float* %p + %vec = insertelement <2 x float> undef, float %s, i32 0 + %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> + %res = select <16 x i1> <i1 1, i1 1, i1 0, i1 1, i1 0, i1 0, i1 0, i1 0, i1 1, i1 1, i1 0, i1 1, i1 0, i1 1, i1 0, i1 1>, <16 x float> %shuf, <16 x float> zeroinitializer + ret <16 x float> %res +} |
