[Clang][BuiltIn][AVX512] Adding intrinsics without mask for VBROADCAST and VPBROADCAST instruction set .

[delena]

include/clang/Basic/BuiltinsX86.def

1996

VL and CD

test/CodeGen/avx512vl-builtins.c

4029

these tests should be in another file VL+CD

[m_zuckerman]

ping

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_cvtd2mask128, "UcV4i","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtd2mask256, "UcV8i","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtmask2d128, "V4iUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtmask2d256, "V8iUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtmask2q128, "V2LLiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtmask2q256, "V4LLiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtq2mask128, "UcV2LLi","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtq2mask256, "UcV4LLi","","avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastmb512, "V8LLiUc","","avx512cd")
+TARGET_BUILTIN(__builtin_ia32_broadcastmw512, "V16iUs","","avx512cd")
+TARGET_BUILTIN(__builtin_ia32_broadcastf32x4_512, "V16fV4fV16fUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcastf64x4_512, "V8dV4dV8dUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcasti32x4_512, "V16iV4iV16iUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcasti64x4_512, "V8LLiV4LLiV8LLiUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcastsd512, "V8dV2dV8dUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcastss512, "V16fV4fV16fUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_broadcastmb128, "V2LLiUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastmb256, "V4LLiUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastmw128, "V4iUs","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_broadcastmw256, "V8iUs","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_512_mask, "V16fV4fV16fUs","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcastf32x8_512_mask, "V16fV8fV16fUs","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_512_mask, "V8dV2dV8dUc","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_512_mask, "V16iV4iV16iUs","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcasti32x8_512_mask, "V16iV8iV16iUs","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcasti64x2_512_mask, "V8LLiV2LLiV8LLiUc","","avx512dq")
 TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_256_mask, "V8fV4fV8fUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_256_mask, "V4dV2dV4dUc","","avx512dq,avx512vl")

lib/Headers/avx512cdintrin.h

 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_maskz_lzcnt_epi64 (__mmask8 __U, __m512i __A)
 {
   return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
              (__v8di) _mm512_setzero_si512 (),
              (__mmask8) __U);
 }
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcastmb_epi64 (__mmask8 __A)
+{
+  return (__m512i) __builtin_ia32_broadcastmb512 (__A);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcastmw_epi32 (__mmask16 __A)
+{
+  return (__m512i) __builtin_ia32_broadcastmw512 (__A);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512fintrin.h

 
 #define _mm_maskz_sqrt_round_ss( __U, __A, __B, __R) __extension__ ({ \
 __builtin_ia32_sqrtss_round_mask ((__v4sf)( __B),\
                  (__v4sf)( __A),(__v4sf) _mm_setzero_ps(),\
                  (__mmask8) __U,\
                  __R);\
 })
 
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_broadcast_f32x4 (__m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
+                 (__v16sf)
+                 _mm512_undefined_ps (),
+                 (__mmask16) -1);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_f32x4 (__m512 __O, __mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
+                 (__v16sf) __O,
+                 __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_f32x4 (__mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
+                 (__v16sf)
+                 _mm512_setzero_ps (),
+                 __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_broadcast_f64x4 (__m256d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
+                  (__v8df)
+                  _mm512_undefined_pd (),
+                  (__mmask8) -1);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_f64x4 (__m512d __O, __mmask8 __M, __m256d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
+                  (__v8df) __O,
+                  __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_f64x4 (__mmask8 __M, __m256d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
+                  (__v8df)
+                  _mm512_setzero_pd (),
+                  __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcast_i32x4 (__m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
+                  (__v16si)
+                  _mm512_undefined_epi32 (),
+                  (__mmask16) -1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_i32x4 (__m512i __O, __mmask16 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
+                  (__v16si) __O,
+                  __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_i32x4 (__mmask16 __M, __m128i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
+                  (__v16si)
+                  _mm512_setzero_si512 (),
+                  __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_broadcast_i64x4 (__m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
+                  (__v8di)
+                  _mm512_undefined_epi32 (),
+                  (__mmask8) -1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_broadcast_i64x4 (__m512i __O, __mmask8 __M, __m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
+                  (__v8di) __O,
+                  __M);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcast_i64x4 (__mmask8 __M, __m256i __A)
+{
+  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
+                  (__v8di)
+                  _mm512_setzero_si512 (),
+                  __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_mask_broadcastsd_pd (__m512d __O, __mmask8 __M, __m128d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastsd512 ((__v2df) __A,
+              (__v8df) __O, __M);
+}
+
+static __inline__ __m512d __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcastsd_pd (__mmask8 __M, __m128d __A)
+{
+  return (__m512d) __builtin_ia32_broadcastsd512 ((__v2df) __A,
+              (__v8df)
+              _mm512_setzero_pd (),
+              __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_broadcastss_ps (__m512 __O, __mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastss512 ((__v4sf) __A,
+             (__v16sf) __O, __M);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_broadcastss_ps (__mmask16 __M, __m128 __A)
+{
+  return (__m512) __builtin_ia32_broadcastss512 ((__v4sf) __A,
+             (__v16sf)
+             _mm512_setzero_ps (),
+             __M);
+}
+
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_cvtsepi32_epi8 (__m512i __A)
 {
   return (__m128i) __builtin_ia32_pmovsdb512_mask ((__v16si) __A,
                (__v16qi) _mm_undefined_si128 (),
                (__mmask16) -1);
 }
 

lib/Headers/avx512vlintrin.h

 _mm256_maskz_movedup_pd (__mmask8 __U, __m256d __A)
 {
   return (__m256d) __builtin_ia32_movddup256_mask ((__v4df) __A,
                (__v4df)
               _mm256_setzero_pd (),
                (__mmask8) __U);
 }
 
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_broadcastmb_epi64 (__mmask8 __A)
+{
+  return (__m128i) __builtin_ia32_broadcastmb128 (__A);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_broadcastmb_epi64 (__mmask8 __A)
+{
+  return (__m256i) __builtin_ia32_broadcastmb256 (__A);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_broadcastmw_epi32 (__mmask16 __A)
+{
+  return (__m128i) __builtin_ia32_broadcastmw128 (__A);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_broadcastmw_epi32 (__mmask16 __A)
+{
+  return (__m256i) __builtin_ia32_broadcastmw256 (__A);
+}
 
 #define _mm_mask_set1_epi32( __O, __M, __A) __extension__ ({ \
 __builtin_ia32_pbroadcastd128_gpr_mask (__A, (__v4si)( __O),\
                 ( __M));\
 })
 
 #define _mm_maskz_set1_epi32( __M, __A) __extension__ ({ \
 __builtin_ia32_pbroadcastd128_gpr_mask (__A,\

test/CodeGen/avx512cdintrin.c

__m512i test_mm512_mask_lzcnt_epi64(__m512i __W, __mmask8 __U, __m512i __A) {
   // CHECK: @llvm.x86.avx512.mask.lzcnt.q.512
   return _mm512_mask_lzcnt_epi64(__W,__U,__A); 
 }
 __m512i test_mm512_maskz_lzcnt_epi64(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_lzcnt_epi64
   // CHECK: @llvm.x86.avx512.mask.lzcnt.q.512
   return _mm512_maskz_lzcnt_epi64(__U,__A); 
 }
+
+__m512i test_mm512_broadcastmb_epi64(__mmask8 __A) {
+  // CHECK-LABEL: @test_mm512_broadcastmb_epi64
+  // CHECK: @llvm.x86.avx512.broadcastmb.512
+  return _mm512_broadcastmb_epi64(__A); 
+}
+
+__m512i test_mm512_broadcastmw_epi32(__mmask16 __A) {
+  // CHECK-LABEL: @test_mm512_broadcastmw_epi32
+  // CHECK: @llvm.x86.avx512.broadcastmw.512
+  return _mm512_broadcastmw_epi32(__A); 
+}

test/CodeGen/avx512f-builtins.c

// CHECK: @llvm.x86.avx512.mask.sqrt.ss
     return _mm_maskz_sqrt_ss(__U,__A,__B);
 }
 
 __m128 test_mm_maskz_sqrt_round_ss(__mmask8 __U, __m128 __A, __m128 __B){
   // CHECK: @llvm.x86.avx512.mask.sqrt.ss
     return _mm_maskz_sqrt_round_ss(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
 }
 
+__m512 test_mm512_broadcast_f32x4(__m128 __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm512_broadcast_f32x4(__A); 
+}
+
+__m512 test_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm512_mask_broadcast_f32x4(__O, __M, __A); 
+}
+
+__m512 test_mm512_maskz_broadcast_f32x4(__mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_f32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+  return _mm512_maskz_broadcast_f32x4(__M, __A); 
+}
+
+__m512d test_mm512_broadcast_f64x4(__m256d __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_f64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
+  return _mm512_broadcast_f64x4(__A); 
+}
+
+__m512d test_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, __m256d __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_f64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
+  return _mm512_mask_broadcast_f64x4(__O, __M, __A); 
+}
+
+__m512d test_mm512_maskz_broadcast_f64x4(__mmask8 __M, __m256d __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_f64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
+  return _mm512_maskz_broadcast_f64x4(__M, __A); 
+}
+
+__m512i test_mm512_broadcast_i32x4(__m128i __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm512_broadcast_i32x4(__A); 
+}
+
+__m512i test_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm512_mask_broadcast_i32x4(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_i32x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
+  return _mm512_maskz_broadcast_i32x4(__M, __A); 
+}
+
+__m512i test_mm512_broadcast_i64x4(__m256i __A) {
+  // CHECK-LABEL: @test_mm512_broadcast_i64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
+  return _mm512_broadcast_i64x4(__A); 
+}
+
+__m512i test_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcast_i64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
+  return _mm512_mask_broadcast_i64x4(__O, __M, __A); 
+}
+
+__m512i test_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcast_i64x4
+  // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
+  return _mm512_maskz_broadcast_i64x4(__M, __A); 
+}
+
+__m512d test_mm512_mask_broadcastsd_pd(__m512d __O, __mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcastsd_pd
+  // CHECK: @llvm.x86.avx512.mask.broadcast.sd.pd.512
+  return _mm512_mask_broadcastsd_pd(__O, __M, __A); 
+}
+
+__m512d test_mm512_maskz_broadcastsd_pd(__mmask8 __M, __m128d __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcastsd_pd
+  // CHECK: @llvm.x86.avx512.mask.broadcast.sd.pd.512
+  return _mm512_maskz_broadcastsd_pd(__M, __A); 
+}
+
+__m512 test_mm512_mask_broadcastss_ps(__m512 __O, __mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_mask_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.512
+  return _mm512_mask_broadcastss_ps(__O, __M, __A); 
+}
+
+__m512 test_mm512_maskz_broadcastss_ps(__mmask16 __M, __m128 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_broadcastss_ps
+  // CHECK: @llvm.x86.avx512.mask.broadcast.ss.ps.512
+  return _mm512_maskz_broadcastss_ps(__M, __A); 
+}
+
 __m128i test_mm512_cvtsepi32_epi8(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvtsepi32_epi8
   // CHECK: @llvm.x86.avx512.mask.pmovs.db.512
   return _mm512_cvtsepi32_epi8(__A); 
 }
 
 __m128i test_mm512_mask_cvtsepi32_epi8(__m128i __O, __mmask16 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtsepi32_epi8

test/CodeGen/avx512vl-builtins.c

 }
 
 __m256d test_mm256_mask_movedup_pd(__m256d __W, __mmask8 __U, __m256d __A) {
  // CHECK-LABEL: @test_mm256_mask_movedup_pd
   // CHECK: @llvm.x86.avx512.mask.movddup.256
   return _mm256_mask_movedup_pd(__W, __U, __A); 
 }
 
+__m128i test_mm_broadcastmb_epi64(__mmask8 __A) {
+  // CHECK-LABEL: @test_mm_broadcastmb_epi64
+  // CHECK: @llvm.x86.avx512.broadcastmb.128
+  return _mm_broadcastmb_epi64(__A); 
+}
+
+__m256i test_mm256_broadcastmb_epi64(__mmask8 __A) {
+  // CHECK-LABEL: @test_mm256_broadcastmb_epi64
+  // CHECK: @llvm.x86.avx512.broadcastmb.256
+  return _mm256_broadcastmb_epi64(__A); 
+}
+
+__m128i test_mm_broadcastmw_epi32(__mmask16 __A) {
+  // CHECK-LABEL: @test_mm_broadcastmw_epi32
+  // CHECK: @llvm.x86.avx512.broadcastmw.128
+  return _mm_broadcastmw_epi32(__A); 
+}
+
+__m256i test_mm256_broadcastmw_epi32(__mmask16 __A) {
+  // CHECK-LABEL: @test_mm256_broadcastmw_epi32
+  // CHECK: @llvm.x86.avx512.broadcastmw.256
+  return _mm256_broadcastmw_epi32(__A); 
+}
+
 __m256d test_mm256_maskz_movedup_pd(__mmask8 __U, __m256d __A) {
  // CHECK-LABEL: @test_mm256_maskz_movedup_pd
   // CHECK: @llvm.x86.avx512.mask.movddup.256
   return _mm256_maskz_movedup_pd(__U, __A); 
 }
 
 __m128i test_mm_mask_set1_epi32(__m128i __O, __mmask8 __M) {
   // CHECK-LABEL: @test_mm_mask_set1_epi32