shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

1. Fixing small types issue (PD|PS) (reduce) .

Browse Source 
2. Cosmetic changes

llvm-svn: 285405
This commit is contained in:
Michael Zuckerman 2016-10-28 15:16:03 +00:00
parent cf5bac8766
commit edd99eb07a
2 changed files with 37 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
clang/lib/Headers/avx512fintrin.h
View File

@ -9668,7 +9668,7 @@ _mm512_mask_abs_pd(__m512d __W, __mmask8 __K, __m512d __A)
// This takes log2(n) steps where n is the number of elements in the vector.
// Vec512 - Vector with size 512.
// Operator - Can be one of following: +,*,&&,||
// Operator - Can be one of following: +,*,&,|
// T2 - Can get 'i' for int and 'f' for float.
// T1 - Can get 'i' for int and 'd' for double.
@ -9725,54 +9725,60 @@ static __inline__ double __DEFAULT_FN_ATTRS _mm512_reduce_mul_pd(__m512d __W) {
}
// Vec512 - Vector with size 512.
// Operator - Can be one of following: +,*,&&,||
// Vec512Neutral - All vector elements set to the identity element.
// Identity element: {+,0},{*,1},{&,0xFFFFFFFFFFFFFFFF},{|,0}
// Operator - Can be one of following: +,*,&,|
// Mask - Intrinsic Mask
// Neutral - Identity element: {+,0},{*,1},{&&,0xFFFFFFFFFFFFFFFF},{||,0}
// T2 - Can get 'i' for int and 'f' for float.
// T1 - Can get 'i' for int and 'd' for packed double-precision.
// T3 - Can be Pd for packed double or q for q-word.
#define _mm512_mask_reduce_operator_64bit(Vec512, Operator, Mask, Neutral, \
T2, T1, T3) \
#define _mm512_mask_reduce_operator_64bit(Vec512, Vec512Neutral, Operator, \
Mask, T2, T1, T3) \
__extension__({ \
Vec512 = __builtin_ia32_select##T3##_512( \
(__mmask8)Mask, (__v8d##T2)Vec512, \
(__v8d##T2)_mm512_set1_epi64(Neutral)); \
(__mmask8)Mask, \
(__v8d##T2)Vec512, \
(__v8d##T2)Vec512Neutral); \
_mm512_reduce_operator_64bit(Vec512, Operator, T2, T1); \
})
static __inline__ long long __DEFAULT_FN_ATTRS
_mm512_mask_reduce_add_epi64(__mmask8 __M, __m512i __W) {
_mm512_mask_reduce_operator_64bit(__W, +, __M, 0, i, i, q);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_epi64(0), +, __M, i, i, q);
}
static __inline__ long long __DEFAULT_FN_ATTRS
_mm512_mask_reduce_mul_epi64(__mmask8 __M, __m512i __W) {
_mm512_mask_reduce_operator_64bit(__W, *, __M, 1, i, i, q);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_epi64(1), *, __M, i, i, q);
}
static __inline__ long long __DEFAULT_FN_ATTRS
_mm512_mask_reduce_and_epi64(__mmask8 __M, __m512i __W) {
_mm512_mask_reduce_operator_64bit(__W, &, __M, 0xFFFFFFFFFFFFFFFF, i, i, q);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_epi64(0xFFFFFFFFFFFFFFFF),
&, __M, i, i, q);
}
static __inline__ long long __DEFAULT_FN_ATTRS
_mm512_mask_reduce_or_epi64(__mmask8 __M, __m512i __W) {
_mm512_mask_reduce_operator_64bit(__W, |, __M, 0, i, i, q);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_epi64(0), |, __M,
i, i, q);
}
static __inline__ double __DEFAULT_FN_ATTRS
_mm512_mask_reduce_add_pd(__mmask8 __M, __m512d __W) {
_mm512_mask_reduce_operator_64bit(__W, +, __M, 0, f, d, pd);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_pd(0), +, __M,
f, d, pd);
}
static __inline__ double __DEFAULT_FN_ATTRS
_mm512_mask_reduce_mul_pd(__mmask8 __M, __m512d __W) {
_mm512_mask_reduce_operator_64bit(__W, *, __M, 1, f, d, pd);
_mm512_mask_reduce_operator_64bit(__W, _mm512_set1_pd(1), *, __M,
f, d, pd);
}
// Vec512 - Vector with size 512.
// Operator - Can be one of following: +,*,&&,||
// Operator - Can be one of following: +,*,&,|
// T2 - Can get 'i' for int and ' ' for packed single.
// T1 - Can get 'i' for int and 'f' for float.
@ -9849,50 +9855,53 @@ _mm512_reduce_mul_ps(__m512 __W) {
}
// Vec512 - Vector with size 512.
// Operator - Can be one of following: +,*,&&,||
// Vec512Neutral - All vector elements set to the identity element.
// Identity element: {+,0},{*,1},{&,0xFFFFFFFF},{|,0}
// Operator - Can be one of following: +,*,&,|
// Mask - Intrinsic Mask
// Neutral - Identity element: {+,0},{*,1},{&&,0xFFFFFFFF},{||,0}
// T2 - Can get 'i' for int and 'f' for float.
// T1 - Can get 'i' for int and 'd' for double.
// T3 - Can be Ps for packed single or d for d-word.
#define _mm512_mask_reduce_operator_32bit(Vec512, Operator, Mask, Neutral, \
T2, T1, T3) \
#define _mm512_mask_reduce_operator_32bit(Vec512, Vec512Neutral, Operator, \
Mask, T2, T1, T3) \
__extension__({ \
Vec512 = (__m512##T1)__builtin_ia32_select##T3##_512( \
(__mmask16)Mask, (__v16s##T2)Vec512, \
(__v16s##T2)_mm512_set1_epi32(Neutral)); \
(__mmask16)Mask, \
(__v16s##T2)Vec512, \
(__v16s##T2)Vec512Neutral); \
_mm512_reduce_operator_32bit(Vec512, Operator, T2, T1); \
})
static __inline__ int __DEFAULT_FN_ATTRS
_mm512_mask_reduce_add_epi32( __mmask16 __M, __m512i __W) {
_mm512_mask_reduce_operator_32bit(__W, +, __M, 0, i, i, d);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_epi32(0), +, __M, i, i, d);
}
static __inline__ int __DEFAULT_FN_ATTRS
_mm512_mask_reduce_mul_epi32( __mmask16 __M, __m512i __W) {
_mm512_mask_reduce_operator_32bit(__W, *, __M, 1, i, i, d);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_epi32(1), *, __M, i, i, d);
}
static __inline__ int __DEFAULT_FN_ATTRS
_mm512_mask_reduce_and_epi32( __mmask16 __M, __m512i __W) {
_mm512_mask_reduce_operator_32bit(__W, &, __M, 0xFFFFFFFF, i, i, d);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_epi32(0xFFFFFFFF), &, __M,
i, i, d);
}
static __inline__ int __DEFAULT_FN_ATTRS
_mm512_mask_reduce_or_epi32(__mmask16 __M, __m512i __W) {
_mm512_mask_reduce_operator_32bit(__W, |, __M, 0, i, i, d);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_epi32(0), |, __M, i, i, d);
}
static __inline__ float __DEFAULT_FN_ATTRS
_mm512_mask_reduce_add_ps(__mmask16 __M, __m512 __W) {
_mm512_mask_reduce_operator_32bit(__W, +, __M, 0, f, , ps);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_ps(0), +, __M, f, , ps);
}
static __inline__ float __DEFAULT_FN_ATTRS
_mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W) {
_mm512_mask_reduce_operator_32bit(__W, *, __M, 1, f, , ps);
_mm512_mask_reduce_operator_32bit(__W, _mm512_set1_ps(1), *, __M, f, , ps);
}
#undef __DEFAULT_FN_ATTRS

4
clang/test/CodeGen/avx512-reduceIntrin.c
View File

@ -391,7 +391,7 @@ double test_mm512_mask_reduce_add_pd(__mmask8 __M, __m512d __W){
double test_mm512_mask_reduce_mul_pd(__mmask8 __M, __m512d __W){
// CHECK: {{.*}} = bitcast i8 %__M to <8 x i1>
// CHECK: {{.*}} = select <8 x i1> {{.*}}, <8 x double> %__W, <8 x double> <double 4.940656e-324, double 4.940656e-324, double 4.940656e-324, double 4.940656e-324, double 4.940656e-324, double 4.940656e-324, double 4.940656e-324, double 4.940656e-324>
// CHECK: {{.*}} = select <8 x i1> %0, <8 x double> %__W, <8 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
// CHECK: %shuffle.i = shufflevector <8 x double> {{.*}}, <8 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: %shuffle1.i = shufflevector <8 x double> {{.*}}, <8 x double> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
// CHECK: %mul.i = fmul <4 x double> %shuffle.i, %shuffle1.i
@ -425,7 +425,7 @@ float test_mm512_mask_reduce_add_ps(__mmask16 __M, __m512 __W){
float test_mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W){
// CHECK: {{.*}} = bitcast i16 %__M to <16 x i1>
// CHECK: {{.*}} = select <16 x i1> {{.*}}, <16 x float> %__W, <16 x float> <float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000, float 0x36A0000000000000>
// CHECK: {{.*}} = select <16 x i1> %0, <16 x float> %__W, <16 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
// CHECK: %shuffle.i = shufflevector <16 x float> {{.*}}, <16 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
// CHECK: %shuffle1.i = shufflevector <16 x float> {{.*}}, <16 x float> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
// CHECK: %mul.i = fmul <8 x float> %shuffle.i, %shuffle1.i