[X86] add 3 missing intrinsics: _mm_(mask/maskz)_cvtpbh_ps

[craig.topper]

clang/lib/Headers/avx512vlbf16intrin.h

429

convertion -> conversion

[FreddyYe]

convertion -> conversion.

[craig.topper]

LGTM

clang/lib/Headers/avx512bf16intrin.h

 }
 
 /// Convert Packed BF16 Data to Packed float Data.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __A
 ///    A 256-bit vector of [16 x bfloat].
-/// \returns A 512-bit vector of [16 x float] come from convertion of __A
+/// \returns A 512-bit vector of [16 x float] come from conversion of __A
 static __inline__ __m512 __DEFAULT_FN_ATTRS512 _mm512_cvtpbh_ps(__m256bh __A) {
   return _mm512_castsi512_ps((__m512i)_mm512_slli_epi32(
       (__m512i)_mm512_cvtepi16_epi32((__m256i)__A), 16));
 }
 
 /// Convert Packed BF16 Data to Packed float Data using zeroing mask.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __U
 ///    A 16-bit mask. Elements are zeroed out when the corresponding mask
 ///    bit is not set.
 /// \param __A
 ///    A 256-bit vector of [16 x bfloat].
-/// \returns A 512-bit vector of [16 x float] come from convertion of __A
+/// \returns A 512-bit vector of [16 x float] come from conversion of __A
 static __inline__ __m512 __DEFAULT_FN_ATTRS512
 _mm512_maskz_cvtpbh_ps(__mmask16 __U, __m256bh __A) {
   return _mm512_castsi512_ps((__m512i)_mm512_slli_epi32(
       (__m512i)_mm512_maskz_cvtepi16_epi32((__mmask16)__U, (__m256i)__A), 16));
 }
 
 /// Convert Packed BF16 Data to Packed float Data using merging mask.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __S
 ///    A 512-bit vector of [16 x float]. Elements are copied from __S when
 ///     the corresponding mask bit is not set.
 /// \param __U
 ///    A 16-bit mask.
 /// \param __A
 ///    A 256-bit vector of [16 x bfloat].
-/// \returns A 512-bit vector of [16 x float] come from convertion of __A
+/// \returns A 512-bit vector of [16 x float] come from conversion of __A
 static __inline__ __m512 __DEFAULT_FN_ATTRS512
 _mm512_mask_cvtpbh_ps(__m512 __S, __mmask16 __U, __m256bh __A) {
   return _mm512_castsi512_ps((__m512i)_mm512_mask_slli_epi32(
       (__m512i)__S, (__mmask16)__U,
       (__m512i)_mm512_cvtepi16_epi32((__m256i)__A), 16));
 }
 
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS512
 
 #endif

clang/lib/Headers/avx512vlbf16intrin.h

static __inline__ __bfloat16 __DEFAULT_FN_ATTRS128 _mm_cvtness_sbh(float __A) {
   return __R[0];
 }
 
 /// Convert Packed BF16 Data to Packed float Data.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __A
+///    A 128-bit vector of [4 x bfloat].
+/// \returns A 128-bit vector of [4 x float] come from conversion of __A
+static __inline__ __m128 __DEFAULT_FN_ATTRS128 _mm_cvtpbh_ps(__m128bh __A) {
+  return _mm_castsi128_ps(
+      (__m128i)_mm_slli_epi32((__m128i)_mm_cvtepi16_epi32((__m128i)__A), 16));
+}
+
+/// Convert Packed BF16 Data to Packed float Data.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \param __A
 ///    A 128-bit vector of [8 x bfloat].
-/// \returns A 256-bit vector of [8 x float] come from convertion of __A
+/// \returns A 256-bit vector of [8 x float] come from conversion of __A
 static __inline__ __m256 __DEFAULT_FN_ATTRS256 _mm256_cvtpbh_ps(__m128bh __A) {
   return _mm256_castsi256_ps((__m256i)_mm256_slli_epi32(
       (__m256i)_mm256_cvtepi16_epi32((__m128i)__A), 16));
 }
 
 /// Convert Packed BF16 Data to Packed float Data using zeroing mask.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __U
+///    A 4-bit mask. Elements are zeroed out when the corresponding mask
+///    bit is not set.
+/// \param __A
+///    A 128-bit vector of [4 x bfloat].
+/// \returns A 128-bit vector of [4 x float] come from conversion of __A
+static __inline__ __m128 __DEFAULT_FN_ATTRS128
+_mm_maskz_cvtpbh_ps(__mmask8 __U, __m128bh __A) {
+  return _mm_castsi128_ps((__m128i)_mm_slli_epi32(
+      (__m128i)_mm_maskz_cvtepi16_epi32((__mmask8)__U, (__m128i)__A), 16));
+}
+
+/// Convert Packed BF16 Data to Packed float Data using zeroing mask.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \param __U
 ///    A 8-bit mask. Elements are zeroed out when the corresponding mask
 ///    bit is not set.
 /// \param __A
 ///    A 128-bit vector of [8 x bfloat].
-/// \returns A 256-bit vector of [8 x float] come from convertion of __A
+/// \returns A 256-bit vector of [8 x float] come from conversion of __A
 static __inline__ __m256 __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtpbh_ps(__mmask8 __U, __m128bh __A) {
   return _mm256_castsi256_ps((__m256i)_mm256_slli_epi32(
       (__m256i)_mm256_maskz_cvtepi16_epi32((__mmask8)__U, (__m128i)__A), 16));
 }
 
 /// Convert Packed BF16 Data to Packed float Data using merging mask.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \param __S
+///    A 128-bit vector of [4 x float]. Elements are copied from __S when
+///     the corresponding mask bit is not set.
+/// \param __U
+///    A 4-bit mask. Elements are zeroed out when the corresponding mask
+///    bit is not set.
+/// \param __A
+///    A 128-bit vector of [4 x bfloat].
+/// \returns A 128-bit vector of [4 x float] come from conversion of __A
+static __inline__ __m128 __DEFAULT_FN_ATTRS128
+_mm_mask_cvtpbh_ps(__m128 __S, __mmask8 __U, __m128bh __A) {
+  return _mm_castsi128_ps((__m128i)_mm_mask_slli_epi32(
+      (__m128i)__S, (__mmask8)__U, (__m128i)_mm_cvtepi16_epi32((__m128i)__A),
+      16));
+}
+
+/// Convert Packed BF16 Data to Packed float Data using merging mask.
+///
+/// \headerfile <x86intrin.h>
+///
+/// \param __S
 ///    A 256-bit vector of [8 x float]. Elements are copied from __S when
 ///     the corresponding mask bit is not set.
 /// \param __U
 ///    A 8-bit mask. Elements are zeroed out when the corresponding mask
 ///    bit is not set.
 /// \param __A
 ///    A 128-bit vector of [8 x bfloat].
-/// \returns A 256-bit vector of [8 x float] come from convertion of __A
+/// \returns A 256-bit vector of [8 x float] come from conversion of __A
 static __inline__ __m256 __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtpbh_ps(__m256 __S, __mmask8 __U, __m128bh __A) {
   return _mm256_castsi256_ps((__m256i)_mm256_mask_slli_epi32(
       (__m256i)__S, (__mmask8)__U, (__m256i)_mm256_cvtepi16_epi32((__m128i)__A),
       16));
 }
 
 #undef __DEFAULT_FN_ATTRS128
 #undef __DEFAULT_FN_ATTRS256
 
 #endif

clang/test/CodeGen/X86/avx512vlbf16-builtins.c

 
 __bfloat16 test_mm_cvtness_sbh(float A) {
   // CHECK-LABEL: @test_mm_cvtness_sbh
   // CHECK: @llvm.x86.avx512bf16.mask.cvtneps2bf16.128
   // CHECK: ret i16 %{{.*}}
   return _mm_cvtness_sbh(A);
 }
 
+__m128 test_mm_cvtpbh_ps(__m128bh A) {
+  // CHECK-LABEL: @test_mm_cvtpbh_ps
+  // CHECK: sext <4 x i16> %{{.*}} to <4 x i32>
+  // CHECK: @llvm.x86.sse2.pslli.d
+  // CHECK: bitcast <2 x i64> %{{.*}} to <4 x float>
+  // CHECK: ret <4 x float> %{{.*}}
+  return _mm_cvtpbh_ps(A);
+}
+
 __m256 test_mm256_cvtpbh_ps(__m128bh A) {
   // CHECK-LABEL: @test_mm256_cvtpbh_ps
   // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
   // CHECK: @llvm.x86.avx2.pslli.d
   // CHECK: bitcast <4 x i64> %{{.*}} to <8 x float>
   // CHECK: ret <8 x float> %{{.*}}
   return _mm256_cvtpbh_ps(A);
 }
 
+__m128 test_mm_maskz_cvtpbh_ps(__mmask8 M, __m128bh A) {
+  // CHECK-LABEL: @test_mm_maskz_cvtpbh_ps
+  // CHECK: sext <4 x i16> %{{.*}} to <4 x i32>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
+  // CHECK: @llvm.x86.sse2.pslli.d
+  // CHECK: bitcast <2 x i64> %{{.*}} to <4 x float>
+  // CHECK: ret <4 x float> %{{.*}}
+  return _mm_maskz_cvtpbh_ps(M, A);
+}
+
 __m256 test_mm256_maskz_cvtpbh_ps(__mmask8 M, __m128bh A) {
   // CHECK-LABEL: @test_mm256_maskz_cvtpbh_ps
   // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   // CHECK: @llvm.x86.avx2.pslli.d
   // CHECK: bitcast <4 x i64> %{{.*}} to <8 x float>
   // CHECK: ret <8 x float> %{{.*}}
   return _mm256_maskz_cvtpbh_ps(M, A);
 }
 
+__m128 test_mm_mask_cvtpbh_ps(__m128 S, __mmask8 M, __m128bh A) {
+  // CHECK-LABEL: @test_mm_mask_cvtpbh_ps
+  // CHECK: sext <4 x i16> %{{.*}} to <4 x i32>
+  // CHECK: @llvm.x86.sse2.pslli.d
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
+  // CHECK: bitcast <2 x i64> %{{.*}} to <4 x float>
+  // CHECK: ret <4 x float> %{{.*}}
+  return _mm_mask_cvtpbh_ps(S, M, A);
+}
+
 __m256 test_mm256_mask_cvtpbh_ps(__m256 S, __mmask8 M, __m128bh A) {
   // CHECK-LABEL: @test_mm256_mask_cvtpbh_ps
   // CHECK: sext <8 x i16> %{{.*}} to <8 x i32>
   // CHECK: @llvm.x86.avx2.pslli.d
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   // CHECK: bitcast <4 x i64> %{{.*}} to <8 x float>
   // CHECK: ret <8 x float> %{{.*}}
   return _mm256_mask_cvtpbh_ps(S, M, A);
 }