AVX512 : VPBROADCASTB lowering

[igorb]

Summary

AVX512: Enabled VPBROADCASTB lowering for v64i8 vectors.

[igorb]

Thanks. Please see the updated diff.

[delena]

LGTM

lib/Target/X86/X86ISelLowering.cpp

if (!Subtarget->useSoftFloat() && Subtarget->hasBWI()) {
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v64i8, Custom);
     setOperationAction(ISD::SELECT,             MVT::v32i1, Custom);
     setOperationAction(ISD::SELECT,             MVT::v64i1, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v32i8, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v32i8, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v32i16, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v32i16, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v32i16, Custom);
+    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v64i8, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v64i8, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v64i8, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v32i1, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v64i1, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v32i16, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v64i8, Custom);
     setOperationAction(ISD::VSELECT,            MVT::v32i16, Legal);
     setOperationAction(ISD::VSELECT,            MVT::v64i8, Legal);

test/CodeGen/X86/avx512-vbroadcast.ll

 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=SKX
 
 define   <16 x i32> @_inreg16xi32(i32 %a) {
 ; CHECK-LABEL: _inreg16xi32:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpbroadcastd %edi, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _inreg16xi32:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastd %edi, %zmm0
+; SKX-NEXT:    retq
   %b = insertelement <16 x i32> undef, i32 %a, i32 0
   %c = shufflevector <16 x i32> %b, <16 x i32> undef, <16 x i32> zeroinitializer
   ret <16 x i32> %c
 }
 
 define   <8 x i64> @_inreg8xi64(i64 %a) {
 ; CHECK-LABEL: _inreg8xi64:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpbroadcastq %rdi, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _inreg8xi64:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastq %rdi, %zmm0
+; SKX-NEXT:    retq
   %b = insertelement <8 x i64> undef, i64 %a, i32 0
   %c = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer
   ret <8 x i64> %c
 }
 
-;CHECK-LABEL: _ss16xfloat_v4
-;CHECK: vbroadcastss %xmm0, %zmm0
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_v4(<4 x float> %a) {
+; CHECK-LABEL: _ss16xfloat_v4:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_v4:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0
+; SKX-NEXT:    retq
   %b = shufflevector <4 x float> %a, <4 x float> undef, <16 x i32> zeroinitializer
   ret <16 x float> %b
 }
 
 define   <16 x float> @_inreg16xfloat(float %a) {
 ; CHECK-LABEL: _inreg16xfloat:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _inreg16xfloat:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0
+; SKX-NEXT:    retq
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   ret <16 x float> %c
 }
 
-;CHECK-LABEL: _ss16xfloat_mask:
-;CHECK: vbroadcastss %xmm0, %zmm1 {%k1}
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_mask(float %a, <16 x float> %i, <16 x i32> %mask1) {
+; CHECK-LABEL: _ss16xfloat_mask:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxord %zmm3, %zmm3, %zmm3
+; CHECK-NEXT:    vpcmpneqd %zmm3, %zmm2, %k1
+; CHECK-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vmovaps %zmm1, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_mask:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxord %zmm3, %zmm3, %zmm3
+; SKX-NEXT:    vpcmpneqd %zmm3, %zmm2, %k1
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
+; SKX-NEXT:    vmovaps %zmm1, %zmm0
+; SKX-NEXT:    retq
   %mask = icmp ne <16 x i32> %mask1, zeroinitializer
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> %i
   ret <16 x float> %r
 }
 
-;CHECK-LABEL: _ss16xfloat_maskz:
-;CHECK: vbroadcastss %xmm0, %zmm0 {%k1} {z}
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_maskz(float %a, <16 x i32> %mask1) {
+; CHECK-LABEL: _ss16xfloat_maskz:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; CHECK-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_maskz:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; SKX-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
+; SKX-NEXT:    retq
   %mask = icmp ne <16 x i32> %mask1, zeroinitializer
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> zeroinitializer
   ret <16 x float> %r
 }
 
-;CHECK-LABEL: _ss16xfloat_load:
-;CHECK: vbroadcastss (%{{.*}}, %zmm
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_load(float* %a.ptr) {
+; CHECK-LABEL: _ss16xfloat_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastss (%rdi), %zmm0
+; SKX-NEXT:    retq
   %a = load float, float* %a.ptr
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   ret <16 x float> %c
 }
 
-;CHECK-LABEL: _ss16xfloat_mask_load:
-;CHECK: vbroadcastss (%rdi), %zmm0 {%k1}
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_mask_load(float* %a.ptr, <16 x float> %i, <16 x i32> %mask1) {
+; CHECK-LABEL: _ss16xfloat_mask_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; CHECK-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_mask_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; SKX-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; SKX-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
+; SKX-NEXT:    retq
   %a = load float, float* %a.ptr
   %mask = icmp ne <16 x i32> %mask1, zeroinitializer
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> %i
   ret <16 x float> %r
 }
 
-;CHECK-LABEL: _ss16xfloat_maskz_load:
-;CHECK: vbroadcastss (%rdi), %zmm0 {%k1} {z}
-;CHECK: ret
 define   <16 x float> @_ss16xfloat_maskz_load(float* %a.ptr, <16 x i32> %mask1) {
+; CHECK-LABEL: _ss16xfloat_maskz_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; CHECK-NEXT:    vpcmpneqd %zmm1, %zmm0, %k1
+; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _ss16xfloat_maskz_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; SKX-NEXT:    vpcmpneqd %zmm1, %zmm0, %k1
+; SKX-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
+; SKX-NEXT:    retq
   %a = load float, float* %a.ptr
   %mask = icmp ne <16 x i32> %mask1, zeroinitializer
   %b = insertelement <16 x float> undef, float %a, i32 0
   %c = shufflevector <16 x float> %b, <16 x float> undef, <16 x i32> zeroinitializer
   %r = select <16 x i1> %mask, <16 x float> %c, <16 x float> zeroinitializer
   ret <16 x float> %r
 }
 
 define   <8 x double> @_inreg8xdouble(double %a) {
 ; CHECK-LABEL: _inreg8xdouble:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _inreg8xdouble:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm0
+; SKX-NEXT:    retq
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   ret <8 x double> %c
 }
 
-;CHECK-LABEL: _sd8xdouble_mask:
-;CHECK: vbroadcastsd %xmm0, %zmm1 {%k1}
-;CHECK: ret
 define   <8 x double> @_sd8xdouble_mask(double %a, <8 x double> %i, <8 x i32> %mask1) {
+; CHECK-LABEL: _sd8xdouble_mask:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; CHECK-NEXT:    vpcmpneqd %zmm3, %zmm2, %k1
+; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vmovaps %zmm1, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _sd8xdouble_mask:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; SKX-NEXT:    vpcmpneqd %ymm3, %ymm2, %k1
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
+; SKX-NEXT:    vmovaps %zmm1, %zmm0
+; SKX-NEXT:    retq
   %mask = icmp ne <8 x i32> %mask1, zeroinitializer
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> %i
   ret <8 x double> %r
 }
 
-;CHECK-LABEL: _sd8xdouble_maskz:
-;CHECK: vbroadcastsd %xmm0, %zmm0 {%k1} {z}
-;CHECK: ret
 define   <8 x double> @_sd8xdouble_maskz(double %a, <8 x i32> %mask1) {
+; CHECK-LABEL: _sd8xdouble_maskz:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxor %ymm2, %ymm2, %ymm2
+; CHECK-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _sd8xdouble_maskz:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxor %ymm2, %ymm2, %ymm2
+; SKX-NEXT:    vpcmpneqd %ymm2, %ymm1, %k1
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
+; SKX-NEXT:    retq
   %mask = icmp ne <8 x i32> %mask1, zeroinitializer
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> zeroinitializer
   ret <8 x double> %r
 }
 
-;CHECK-LABEL: _sd8xdouble_load:
-;CHECK: vbroadcastsd (%rdi), %zmm
-;CHECK: ret
 define   <8 x double> @_sd8xdouble_load(double* %a.ptr) {
+; CHECK-LABEL: _sd8xdouble_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _sd8xdouble_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastsd (%rdi), %zmm0
+; SKX-NEXT:    retq
   %a = load double, double* %a.ptr
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   ret <8 x double> %c
 }
 
-;CHECK-LABEL: _sd8xdouble_mask_load:
-;CHECK: vbroadcastsd (%rdi), %zmm0 {%k1}
-;CHECK: ret
 define   <8 x double> @_sd8xdouble_mask_load(double* %a.ptr, <8 x double> %i, <8 x i32> %mask1) {
+; CHECK-LABEL: _sd8xdouble_mask_load:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxor %ymm2, %ymm2, %ymm2
+; CHECK-NEXT:    vpcmpneqd %zmm2, %zmm1, %k1
+; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _sd8xdouble_mask_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxor %ymm2, %ymm2, %ymm2
+; SKX-NEXT:    vpcmpneqd %ymm2, %ymm1, %k1
+; SKX-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
+; SKX-NEXT:    retq
   %a = load double, double* %a.ptr
   %mask = icmp ne <8 x i32> %mask1, zeroinitializer
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> %i
   ret <8 x double> %r
 }
 
 define   <8 x double> @_sd8xdouble_maskz_load(double* %a.ptr, <8 x i32> %mask1) {
 ; CHECK-LABEL: _sd8xdouble_maskz_load:
-; CHECK:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
-; CHECK:    ret
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpxor %ymm1, %ymm1, %ymm1
+; CHECK-NEXT:    vpcmpneqd %zmm1, %zmm0, %k1
+; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _sd8xdouble_maskz_load:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpxor %ymm1, %ymm1, %ymm1
+; SKX-NEXT:    vpcmpneqd %ymm1, %ymm0, %k1
+; SKX-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
+; SKX-NEXT:    retq
   %a = load double, double* %a.ptr
   %mask = icmp ne <8 x i32> %mask1, zeroinitializer
   %b = insertelement <8 x double> undef, double %a, i32 0
   %c = shufflevector <8 x double> %b, <8 x double> undef, <8 x i32> zeroinitializer
   %r = select <8 x i1> %mask, <8 x double> %c, <8 x double> zeroinitializer
   ret <8 x double> %r
 }
 
 define   <16 x i32> @_xmm16xi32(<16 x i32> %a) {
 ; CHECK-LABEL: _xmm16xi32:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpbroadcastd %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _xmm16xi32:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastd %xmm0, %zmm0
+; SKX-NEXT:    retq
   %b = shufflevector <16 x i32> %a, <16 x i32> undef, <16 x i32> zeroinitializer
   ret <16 x i32> %b
 }
 
 define   <16 x float> @_xmm16xfloat(<16 x float> %a) {
 ; CHECK-LABEL: _xmm16xfloat:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _xmm16xfloat:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0
+; SKX-NEXT:    retq
   %b = shufflevector <16 x float> %a, <16 x float> undef, <16 x i32> zeroinitializer
   ret <16 x float> %b
 }
 
 define <16 x i32> @test_vbroadcast() {
 ; CHECK-LABEL: test_vbroadcast:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vpxord %zmm0, %zmm0, %zmm0
 ; CHECK-NEXT:    vcmpunordps %zmm0, %zmm0, %k1
 ; CHECK-NEXT:    vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z}
 ; CHECK-NEXT:    knotw %k1, %k1
 ; CHECK-NEXT:    vmovdqu32 %zmm0, %zmm0 {%k1} {z}
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_vbroadcast:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vpxord %zmm0, %zmm0, %zmm0
+; SKX-NEXT:    vcmpunordps %zmm0, %zmm0, %k0
+; SKX-NEXT:    vpmovm2d %k0, %zmm0
+; SKX-NEXT:    knotw %k0, %k1
+; SKX-NEXT:    vmovdqu32 %zmm0, %zmm0 {%k1} {z}
+; SKX-NEXT:    retq
 entry:
   %0 = sext <16 x i1> zeroinitializer to <16 x i32>
   %1 = fcmp uno <16 x float> undef, zeroinitializer
   %2 = sext <16 x i1> %1 to <16 x i32>
   %3 = select <16 x i1> %1, <16 x i32> %0, <16 x i32> %2
   ret <16 x i32> %3
 }
 
 ; We implement the set1 intrinsics with vector initializers.  Verify that the
 ; IR generated will produce broadcasts at the end.
 define <8 x double> @test_set1_pd(double %d) #2 {
 ; CHECK-LABEL: test_set1_pd:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_set1_pd:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm0
+; SKX-NEXT:    retq
 entry:
   %vecinit.i = insertelement <8 x double> undef, double %d, i32 0
   %vecinit1.i = insertelement <8 x double> %vecinit.i, double %d, i32 1
   %vecinit2.i = insertelement <8 x double> %vecinit1.i, double %d, i32 2
   %vecinit3.i = insertelement <8 x double> %vecinit2.i, double %d, i32 3
   %vecinit4.i = insertelement <8 x double> %vecinit3.i, double %d, i32 4
   %vecinit5.i = insertelement <8 x double> %vecinit4.i, double %d, i32 5
   %vecinit6.i = insertelement <8 x double> %vecinit5.i, double %d, i32 6
   %vecinit7.i = insertelement <8 x double> %vecinit6.i, double %d, i32 7
   ret <8 x double> %vecinit7.i
 }
 
 define <8 x i64> @test_set1_epi64(i64 %d) #2 {
 ; CHECK-LABEL: test_set1_epi64:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vpbroadcastq %rdi, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_set1_epi64:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vpbroadcastq %rdi, %zmm0
+; SKX-NEXT:    retq
 entry:
   %vecinit.i = insertelement <8 x i64> undef, i64 %d, i32 0
   %vecinit1.i = insertelement <8 x i64> %vecinit.i, i64 %d, i32 1
   %vecinit2.i = insertelement <8 x i64> %vecinit1.i, i64 %d, i32 2
   %vecinit3.i = insertelement <8 x i64> %vecinit2.i, i64 %d, i32 3
   %vecinit4.i = insertelement <8 x i64> %vecinit3.i, i64 %d, i32 4
   %vecinit5.i = insertelement <8 x i64> %vecinit4.i, i64 %d, i32 5
   %vecinit6.i = insertelement <8 x i64> %vecinit5.i, i64 %d, i32 6
   %vecinit7.i = insertelement <8 x i64> %vecinit6.i, i64 %d, i32 7
   ret <8 x i64> %vecinit7.i
 }
 
 define <16 x float> @test_set1_ps(float %f) #2 {
 ; CHECK-LABEL: test_set1_ps:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_set1_ps:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0
+; SKX-NEXT:    retq
 entry:
   %vecinit.i = insertelement <16 x float> undef, float %f, i32 0
   %vecinit1.i = insertelement <16 x float> %vecinit.i, float %f, i32 1
   %vecinit2.i = insertelement <16 x float> %vecinit1.i, float %f, i32 2
   %vecinit3.i = insertelement <16 x float> %vecinit2.i, float %f, i32 3
   %vecinit4.i = insertelement <16 x float> %vecinit3.i, float %f, i32 4
   %vecinit5.i = insertelement <16 x float> %vecinit4.i, float %f, i32 5
   %vecinit6.i = insertelement <16 x float> %vecinit5.i, float %f, i32 6
entry:
   ret <16 x float> %vecinit15.i
 }
 
 define <16 x i32> @test_set1_epi32(i32 %f) #2 {
 ; CHECK-LABEL: test_set1_epi32:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vpbroadcastd %edi, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_set1_epi32:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vpbroadcastd %edi, %zmm0
+; SKX-NEXT:    retq
 entry:
   %vecinit.i = insertelement <16 x i32> undef, i32 %f, i32 0
   %vecinit1.i = insertelement <16 x i32> %vecinit.i, i32 %f, i32 1
   %vecinit2.i = insertelement <16 x i32> %vecinit1.i, i32 %f, i32 2
   %vecinit3.i = insertelement <16 x i32> %vecinit2.i, i32 %f, i32 3
   %vecinit4.i = insertelement <16 x i32> %vecinit3.i, i32 %f, i32 4
   %vecinit5.i = insertelement <16 x i32> %vecinit4.i, i32 %f, i32 5
   %vecinit6.i = insertelement <16 x i32> %vecinit5.i, i32 %f, i32 6

 
 ; We implement the scalar broadcast intrinsics with vector initializers.
 ; Verify that the IR generated will produce the broadcast at the end.
 define <8 x double> @test_mm512_broadcastsd_pd(<2 x double> %a) {
 ; CHECK-LABEL: test_mm512_broadcastsd_pd:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
 ; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test_mm512_broadcastsd_pd:
+; SKX:       ## BB#0: ## %entry
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm0
+; SKX-NEXT:    retq
 entry:
   %0 = extractelement <2 x double> %a, i32 0
   %vecinit.i = insertelement <8 x double> undef, double %0, i32 0
   %vecinit1.i = insertelement <8 x double> %vecinit.i, double %0, i32 1
   %vecinit2.i = insertelement <8 x double> %vecinit1.i, double %0, i32 2
   %vecinit3.i = insertelement <8 x double> %vecinit2.i, double %0, i32 3
   %vecinit4.i = insertelement <8 x double> %vecinit3.i, double %0, i32 4
   %vecinit5.i = insertelement <8 x double> %vecinit4.i, double %0, i32 5
   %vecinit6.i = insertelement <8 x double> %vecinit5.i, double %0, i32 6
   %vecinit7.i = insertelement <8 x double> %vecinit6.i, double %0, i32 7
   ret <8 x double> %vecinit7.i
 }
 
-; CHECK-LABEL: test1
-; CHECK: vbroadcastss
 define <16 x float> @test1(<8 x float>%a)  {
+; CHECK-LABEL: test1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test1:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastss %xmm0, %zmm0
+; SKX-NEXT:    retq
   %res = shufflevector <8 x float> %a, <8 x float> undef, <16 x i32> zeroinitializer
   ret <16 x float>%res
 }
 
-; CHECK-LABEL: test2
-; CHECK: vbroadcastsd
 define <8 x double> @test2(<4 x double>%a)  {
+; CHECK-LABEL: test2:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: test2:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vbroadcastsd %xmm0, %zmm0
+; SKX-NEXT:    retq
   %res = shufflevector <4 x double> %a, <4 x double> undef, <8 x i32> zeroinitializer
   ret <8 x double>%res
 }
 
-; CHECK-LABEL: test3
-; CHECK: vpbroadcastd
-define <16 x i32> @test3(<8 x i32>%a)  {
+define <64 x i8> @_invec32xi8(<32 x i8>%a)  {
+; CHECK-LABEL: _invec32xi8:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpbroadcastb %xmm0, %ymm0
+; CHECK-NEXT:    vmovaps %zmm0, %zmm1
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _invec32xi8:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastb %xmm0, %zmm0
+; SKX-NEXT:    retq
+  %res = shufflevector <32 x i8> %a, <32 x i8> undef, <64 x i32> zeroinitializer
+  ret <64 x i8>%res
+}
+
+define <32 x i16> @_invec16xi16(<16 x i16>%a)  {
+; CHECK-LABEL: _invec16xi16:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpbroadcastw %xmm0, %ymm0
+; CHECK-NEXT:    vmovaps %zmm0, %zmm1
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _invec16xi16:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastw %xmm0, %zmm0
+; SKX-NEXT:    retq
+  %res = shufflevector <16 x i16> %a, <16 x i16> undef, <32 x i32> zeroinitializer
+  ret <32 x i16>%res
+}
+
+define <16 x i32> @_invec8xi32(<8 x i32>%a)  {
+; CHECK-LABEL: _invec8xi32:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpbroadcastd %xmm0, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _invec8xi32:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastd %xmm0, %zmm0
+; SKX-NEXT:    retq
   %res = shufflevector <8 x i32> %a, <8 x i32> undef, <16 x i32> zeroinitializer
   ret <16 x i32>%res
 }
 
-; CHECK-LABEL: test4
-; CHECK: vpbroadcastq
-define <8 x i64> @test4(<4 x i64>%a)  {
+define <8 x i64> @_invec4xi64(<4 x i64>%a)  {
+; CHECK-LABEL: _invec4xi64:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpbroadcastq %xmm0, %zmm0
+; CHECK-NEXT:    retq
+;
+; SKX-LABEL: _invec4xi64:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpbroadcastq %xmm0, %zmm0
+; SKX-NEXT:    retq
   %res = shufflevector <4 x i64> %a, <4 x i64> undef, <8 x i32> zeroinitializer
   ret <8 x i64>%res
 }
+