r271373 - [Clang][Intrinsics][avx512] Adding round intrinsics fot max/min/sqrt instruction set to clang
Michael Zuckerman via cfe-commits
cfe-commits at lists.llvm.org
Wed Jun 1 01:34:03 PDT 2016
Author: mzuckerm
Date: Wed Jun 1 03:34:03 2016
New Revision: 271373
URL: http://llvm.org/viewvc/llvm-project?rev=271373&view=rev
Log:
[Clang][Intrinsics][avx512] Adding round intrinsics fot max/min/sqrt instruction set to clang
Differential Revision: http://reviews.llvm.org/D20812
Modified:
cfe/trunk/lib/Headers/avx512fintrin.h
cfe/trunk/test/CodeGen/avx512f-builtins.c
Modified: cfe/trunk/lib/Headers/avx512fintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/avx512fintrin.h?rev=271373&r1=271372&r2=271373&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/avx512fintrin.h (original)
+++ cfe/trunk/lib/Headers/avx512fintrin.h Wed Jun 1 03:34:03 2016
@@ -813,6 +813,29 @@ _mm512_maskz_sub_epi32 (__mmask16 __U, _
(__mmask16) __U);
}
+#define _mm512_mask_max_round_pd( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)( __W),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_max_round_pd( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)\
+ _mm512_setzero_pd (),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_max_round_pd( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)\
+ _mm512_undefined_pd (),\
+ (__mmask8) -1,( __R));\
+})
+
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_max_pd(__m512d __A, __m512d __B)
{
@@ -845,6 +868,29 @@ _mm512_maskz_max_pd (__mmask8 __U, __m51
_MM_FROUND_CUR_DIRECTION);
}
+#define _mm512_mask_max_round_ps( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)( __W),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_max_round_ps( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)\
+ _mm512_setzero_ps (),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_max_round_ps( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_maxps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)\
+ _mm512_undefined_ps (),\
+ (__mmask16) -1,( __R));\
+})
+
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_max_ps(__m512 __A, __m512 __B)
{
@@ -1062,6 +1108,29 @@ _mm512_maskz_max_epu64 (__mmask8 __M, __
__M);
}
+#define _mm512_mask_min_round_pd( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)( __W),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_min_round_pd( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)\
+ _mm512_setzero_pd (),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_min_round_pd( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minpd512_mask ((__v8df)( __A),\
+ (__v8df)( __B),\
+ (__v8df)\
+ _mm512_undefined_pd (),\
+ (__mmask8) -1,( __R));\
+})
+
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_min_pd(__m512d __A, __m512d __B)
{
@@ -1083,6 +1152,29 @@ _mm512_mask_min_pd (__m512d __W, __mmask
_MM_FROUND_CUR_DIRECTION);
}
+#define _mm512_mask_min_round_ps( __W, __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)( __W),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_min_round_ps( __U, __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)\
+ _mm512_setzero_ps (),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_min_round_ps( __A, __B, __R) __extension__ ({ \
+__builtin_ia32_minps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __B),\
+ (__v16sf)\
+ _mm512_undefined_ps (),\
+ (__mmask16) -1,( __R));\
+})
+
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_min_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
@@ -1391,6 +1483,26 @@ _mm512_mask_mullo_epi32 (__m512i __W, __
(__v16si) __W, __M);
}
+#define _mm512_mask_sqrt_round_pd( __W, __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+ (__v8df)( __W),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_maskz_sqrt_round_pd( __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+ (__v8df)\
+ _mm512_setzero_pd (),\
+ (__mmask8)( __U),( __R));\
+})
+
+#define _mm512_sqrt_round_pd( __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtpd512_mask ((__v8df)( __A),\
+ (__v8df)\
+ _mm512_undefined_pd (),\
+ (__mmask8) -1,( __R));\
+})
+
static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_sqrt_pd(__m512d __a)
{
@@ -1419,6 +1531,26 @@ _mm512_maskz_sqrt_pd (__mmask8 __U, __m5
_MM_FROUND_CUR_DIRECTION);
}
+#define _mm512_mask_sqrt_round_ps( __W, __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+ (__v16sf)( __W),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_maskz_sqrt_round_ps( __U, __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+ (__v16sf)\
+ _mm512_setzero_ps (),\
+ (__mmask16)( __U),( __R));\
+})
+
+#define _mm512_sqrt_round_ps( __A, __R) __extension__ ({ \
+__builtin_ia32_sqrtps512_mask ((__v16sf)( __A),\
+ (__v16sf)\
+ _mm512_undefined_ps (),\
+ (__mmask16) -1,( __R));\
+})
+
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_sqrt_ps(__m512 __a)
{
@@ -1679,7 +1811,6 @@ _mm512_mask_floor_ps (__m512 __W, __mmas
_MM_FROUND_CUR_DIRECTION);
}
-
static __inline __m512d __DEFAULT_FN_ATTRS
_mm512_floor_pd(__m512d __A)
{
@@ -1698,7 +1829,6 @@ _mm512_mask_floor_pd (__m512d __W, __mma
_MM_FROUND_CUR_DIRECTION);
}
-
static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_ceil_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
@@ -3679,8 +3809,7 @@ _mm512_maskz_cvtpd_ps (__mmask8 __U, __m
(__m256i)__builtin_ia32_vcvtps2ph512_mask((__v16sf)(__m512)(A), (int)(I), \
(__v16hi)_mm256_setzero_si256(), \
(__mmask16)(W)); })
-
-
+
static __inline __m512 __DEFAULT_FN_ATTRS
_mm512_cvtph_ps(__m256i __A)
{
Modified: cfe/trunk/test/CodeGen/avx512f-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/avx512f-builtins.c?rev=271373&r1=271372&r2=271373&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/avx512f-builtins.c (original)
+++ cfe/trunk/test/CodeGen/avx512f-builtins.c Wed Jun 1 03:34:03 2016
@@ -26,6 +26,27 @@ __m512d test_mm512_maskz_sqrt_pd (__mmas
return _mm512_maskz_sqrt_pd (__U,__A);
}
+__m512d test_mm512_mask_sqrt_round_pd(__m512d __W,__mmask8 __U,__m512d __A)
+{
+ // CHECK-LABEL: @test_mm512_mask_sqrt_round_pd
+ // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+ return _mm512_mask_sqrt_round_pd(__W,__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_sqrt_round_pd(__mmask8 __U,__m512d __A)
+{
+ // CHECK-LABEL: @test_mm512_maskz_sqrt_round_pd
+ // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+ return _mm512_maskz_sqrt_round_pd(__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_sqrt_round_pd(__m512d __A)
+{
+ // CHECK-LABEL: @test_mm512_sqrt_round_pd
+ // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
+ return _mm512_sqrt_round_pd(__A,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512 test_mm512_sqrt_ps(__m512 a)
{
// CHECK-LABEL: @test_mm512_sqrt_ps
@@ -47,6 +68,27 @@ __m512 test_mm512_maskz_sqrt_ps( __mmask
return _mm512_maskz_sqrt_ps(__U ,__A);
}
+__m512 test_mm512_mask_sqrt_round_ps(__m512 __W,__mmask16 __U,__m512 __A)
+{
+ // CHECK-LABEL: @test_mm512_mask_sqrt_round_ps
+ // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+ return _mm512_mask_sqrt_round_ps(__W,__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_sqrt_round_ps(__mmask16 __U,__m512 __A)
+{
+ // CHECK-LABEL: @test_mm512_maskz_sqrt_round_ps
+ // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+ return _mm512_maskz_sqrt_round_ps(__U,__A,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_sqrt_round_ps(__m512 __A)
+{
+ // CHECK-LABEL: @test_mm512_sqrt_round_ps
+ // CHECK: @llvm.x86.avx512.mask.sqrt.ps.512
+ return _mm512_sqrt_round_ps(__A,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512d test_mm512_rsqrt14_pd(__m512d a)
{
// CHECK-LABEL: @test_mm512_rsqrt14_pd
@@ -6576,6 +6618,27 @@ __m512 test_mm512_mask_max_ps (__m512 __
return _mm512_mask_max_ps (__W,__U,__A,__B);
}
+__m512d test_mm512_mask_max_round_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_mask_max_round_pd
+ // CHECK: @llvm.x86.avx512.mask.max.pd.512
+ return _mm512_mask_max_round_pd(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_max_round_pd(__mmask8 __U,__m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_maskz_max_round_pd
+ // CHECK: @llvm.x86.avx512.mask.max.pd.512
+ return _mm512_maskz_max_round_pd(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_max_round_pd(__m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_max_round_pd
+ // CHECK: @llvm.x86.avx512.mask.max.pd.512
+ return _mm512_max_round_pd(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512 test_mm512_maskz_max_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
// CHECK-LABEL: @test_mm512_maskz_max_ps
@@ -6583,6 +6646,27 @@ __m512 test_mm512_maskz_max_ps (__mmask1
return _mm512_maskz_max_ps (__U,__A,__B);
}
+__m512 test_mm512_mask_max_round_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_mask_max_round_ps
+ // CHECK: @llvm.x86.avx512.mask.max.ps.512
+ return _mm512_mask_max_round_ps(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_max_round_ps(__mmask16 __U,__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_maskz_max_round_ps
+ // CHECK: @llvm.x86.avx512.mask.max.ps.512
+ return _mm512_maskz_max_round_ps(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_max_round_ps(__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_max_round_ps
+ // CHECK: @llvm.x86.avx512.mask.max.ps.512
+ return _mm512_max_round_ps(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512d test_mm512_mask_min_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
// CHECK-LABEL: @test_mm512_mask_min_pd
@@ -6597,6 +6681,27 @@ __m512d test_mm512_maskz_min_pd (__mmask
return _mm512_maskz_min_pd (__U,__A,__B);
}
+__m512d test_mm512_mask_min_round_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_mask_min_round_pd
+ // CHECK: @llvm.x86.avx512.mask.min.pd.512
+ return _mm512_mask_min_round_pd(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_maskz_min_round_pd(__mmask8 __U,__m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_maskz_min_round_pd
+ // CHECK: @llvm.x86.avx512.mask.min.pd.512
+ return _mm512_maskz_min_round_pd(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512d test_mm512_min_round_pd( __m512d __A,__m512d __B)
+{
+ // CHECK-LABEL: @test_mm512_min_round_pd
+ // CHECK: @llvm.x86.avx512.mask.min.pd.512
+ return _mm512_min_round_pd(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512 test_mm512_mask_min_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
// CHECK-LABEL: @test_mm512_mask_min_ps
@@ -6611,6 +6716,27 @@ __m512 test_mm512_maskz_min_ps (__mmask1
return _mm512_maskz_min_ps (__U,__A,__B);
}
+__m512 test_mm512_mask_min_round_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_mask_min_round_ps
+ // CHECK: @llvm.x86.avx512.mask.min.ps.512
+ return _mm512_mask_min_round_ps(__W,__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_maskz_min_round_ps(__mmask16 __U,__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_maskz_min_round_ps
+ // CHECK: @llvm.x86.avx512.mask.min.ps.512
+ return _mm512_maskz_min_round_ps(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
+__m512 test_mm512_min_round_ps(__m512 __A,__m512 __B)
+{
+ // CHECK-LABEL: @test_mm512_min_round_ps
+ // CHECK: @llvm.x86.avx512.mask.min.ps.512
+ return _mm512_min_round_ps(__A,__B,_MM_FROUND_CUR_DIRECTION);
+}
+
__m512 test_mm512_mask_floor_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
// CHECK-LABEL: @test_mm512_mask_floor_ps
More information about the cfe-commits
mailing list