From c4f782fdcb4e61dff04ae4c8376e2d9d124b896d Mon Sep 17 00:00:00 2001 From: kyukhin Date: Thu, 20 Feb 2014 06:26:38 +0000 Subject: [PATCH] gcc/ * config/i386/avx512erintrin.h (_mm_rcp28_round_sd): Swap operands. (_mm_rcp28_round_ss): Ditto. (_mm_rsqrt28_round_sd): Ditto. (_mm_rsqrt28_round_ss): Ditto. * config/i386/avx512erintrin.h (_mm_rcp14_round_sd): Ditto. (_mm_rcp14_round_ss): Ditto. (_mm_rsqrt14_round_sd): Ditto. (_mm_rsqrt14_round_ss): Ditto. * config/i386/sse.md (rsqrt14): Put nonimmediate operand as the first input operand, get rid of match_dup. (avx512er_exp2): Set type attribute to sse. (avx512er_rcp28): Ditto. (avx512er_vmrcp28): Put nonimmediate operand as the first input operand, set type attribute. (avx512er_rsqrt28): Set type attribute. (avx512er_vmrsqrt28): Put nonimmediate operand as the first input operand, set type attribute. testsuite/gcc/ * gcc.target/i386/avx512er-vrcp28sd-2.c: Distinguish src1 and src2. * gcc.target/i386/avx512er-vrcp28ss-2.c: Call correct intrinsic. * gcc.target/i386/avx512er-vrsqrt28sd-2.c: Distinguish src1 and src2. * gcc.target/i386/avx512er-vrsqrt28ss-2.c: Ditto. * gcc.target/i386/avx512f-vrcp14sd-2.c: Fix reference calculation. * gcc.target/i386/avx512f-vrcp14ss-2.c: Fix reference calculation. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@207932 138bc75d-0d04-0410-961f-82ee72b054a4 --- gcc/ChangeLog | 24 ++++++++++++++++++++++ gcc/config/i386/avx512erintrin.h | 24 +++++++++++----------- gcc/config/i386/avx512fintrin.h | 16 +++++++-------- gcc/config/i386/sse.md | 16 +++++++++------ gcc/testsuite/ChangeLog | 10 +++++++++ ...x512er-vrsqrt28sd-2.c => avx512er-vrcp28sd-2.c} | 11 +++++----- .../gcc.target/i386/avx512er-vrcp28ss-2.c | 11 +++++----- .../gcc.target/i386/avx512er-vrsqrt28sd-2.c | 11 +++++----- .../gcc.target/i386/avx512er-vrsqrt28ss-2.c | 11 +++++----- gcc/testsuite/gcc.target/i386/avx512f-vrcp14sd-2.c | 4 ++-- gcc/testsuite/gcc.target/i386/avx512f-vrcp14ss-2.c | 8 ++++---- 11 files changed, 94 insertions(+), 52 deletions(-) copy gcc/testsuite/gcc.target/i386/{avx512er-vrsqrt28sd-2.c => avx512er-vrcp28sd-2.c} (61%) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 8f84dbdd34a..6748c0fc36a 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,27 @@ +2014-02-20 Ilya Tocar + Kirill Yukhin + + * config/i386/avx512erintrin.h (_mm_rcp28_round_sd): Swap operands. + (_mm_rcp28_round_ss): Ditto. + (_mm_rsqrt28_round_sd): Ditto. + (_mm_rsqrt28_round_ss): Ditto. + * config/i386/avx512erintrin.h (_mm_rcp14_round_sd): Ditto. + (_mm_rcp14_round_ss): Ditto. + (_mm_rsqrt14_round_sd): Ditto. + (_mm_rsqrt14_round_ss): Ditto. + * config/i386/sse.md (rsqrt14): Put nonimmediate operand as + the first input operand, get rid of match_dup. + (avx512er_exp2): Set type + attribute to sse. + (avx512er_rcp28): + Ditto. + (avx512er_vmrcp28): Put nonimmediate + operand as the first input operand, set type attribute. + (avx512er_rsqrt28): + Set type attribute. + (avx512er_vmrsqrt28): Put nonimmediate + operand as the first input operand, set type attribute. + 2014-02-19 Bill Schmidt * config/rs6000/rs6000.c (vspltis_constant): Fix most significant diff --git a/gcc/config/i386/avx512erintrin.h b/gcc/config/i386/avx512erintrin.h index 6fe05bc6608..f6870a5f723 100644 --- a/gcc/config/i386/avx512erintrin.h +++ b/gcc/config/i386/avx512erintrin.h @@ -163,8 +163,8 @@ extern __inline __m128d __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rcp28_round_sd (__m128d __A, __m128d __B, int __R) { - return (__m128d) __builtin_ia32_rcp28sd_round ((__v2df) __A, - (__v2df) __B, + return (__m128d) __builtin_ia32_rcp28sd_round ((__v2df) __B, + (__v2df) __A, __R); } @@ -172,8 +172,8 @@ extern __inline __m128 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rcp28_round_ss (__m128 __A, __m128 __B, int __R) { - return (__m128) __builtin_ia32_rcp28ss_round ((__v4sf) __A, - (__v4sf) __B, + return (__m128) __builtin_ia32_rcp28ss_round ((__v4sf) __B, + (__v4sf) __A, __R); } @@ -237,8 +237,8 @@ extern __inline __m128d __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rsqrt28_round_sd (__m128d __A, __m128d __B, int __R) { - return (__m128d) __builtin_ia32_rsqrt28sd_round ((__v2df) __A, - (__v2df) __B, + return (__m128d) __builtin_ia32_rsqrt28sd_round ((__v2df) __B, + (__v2df) __A, __R); } @@ -246,8 +246,8 @@ extern __inline __m128 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rsqrt28_round_ss (__m128 __A, __m128 __B, int __R) { - return (__m128) __builtin_ia32_rsqrt28ss_round ((__v4sf) __A, - (__v4sf) __B, + return (__m128) __builtin_ia32_rsqrt28ss_round ((__v4sf) __B, + (__v4sf) __A, __R); } @@ -375,16 +375,16 @@ _mm_rsqrt28_round_ss (__m128 __A, __m128 __B, int __R) _mm512_maskz_rsqrt28_round_ps(U, A, _MM_FROUND_CUR_DIRECTION) #define _mm_rcp28_sd(A, B) \ - __builtin_ia32_rcp28sd_round(A, B, _MM_FROUND_CUR_DIRECTION) + __builtin_ia32_rcp28sd_round(B, A, _MM_FROUND_CUR_DIRECTION) #define _mm_rcp28_ss(A, B) \ - __builtin_ia32_rcp28ss_round(A, B, _MM_FROUND_CUR_DIRECTION) + __builtin_ia32_rcp28ss_round(B, A, _MM_FROUND_CUR_DIRECTION) #define _mm_rsqrt28_sd(A, B) \ - __builtin_ia32_rsqrt28sd_round(A, B, _MM_FROUND_CUR_DIRECTION) + __builtin_ia32_rsqrt28sd_round(B, A, _MM_FROUND_CUR_DIRECTION) #define _mm_rsqrt28_ss(A, B) \ - __builtin_ia32_rsqrt28ss_round(A, B, _MM_FROUND_CUR_DIRECTION) + __builtin_ia32_rsqrt28ss_round(B, A, _MM_FROUND_CUR_DIRECTION) #ifdef __DISABLE_AVX512ER__ #undef __DISABLE_AVX512ER__ diff --git a/gcc/config/i386/avx512fintrin.h b/gcc/config/i386/avx512fintrin.h index b3a4f3a41ff..65a49c5c1f3 100644 --- a/gcc/config/i386/avx512fintrin.h +++ b/gcc/config/i386/avx512fintrin.h @@ -1470,16 +1470,16 @@ extern __inline __m128d __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rcp14_sd (__m128d __A, __m128d __B) { - return (__m128d) __builtin_ia32_rcp14sd ((__v2df) __A, - (__v2df) __B); + return (__m128d) __builtin_ia32_rcp14sd ((__v2df) __B, + (__v2df) __A); } extern __inline __m128 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rcp14_ss (__m128 __A, __m128 __B) { - return (__m128) __builtin_ia32_rcp14ss ((__v4sf) __A, - (__v4sf) __B); + return (__m128) __builtin_ia32_rcp14ss ((__v4sf) __B, + (__v4sf) __A); } extern __inline __m512d @@ -1544,16 +1544,16 @@ extern __inline __m128d __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rsqrt14_sd (__m128d __A, __m128d __B) { - return (__m128d) __builtin_ia32_rsqrt14sd ((__v2df) __A, - (__v2df) __B); + return (__m128d) __builtin_ia32_rsqrt14sd ((__v2df) __B, + (__v2df) __A); } extern __inline __m128 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) _mm_rsqrt14_ss (__m128 __A, __m128 __B) { - return (__m128) __builtin_ia32_rsqrt14ss ((__v4sf) __A, - (__v4sf) __B); + return (__m128) __builtin_ia32_rsqrt14ss ((__v4sf) __B, + (__v4sf) __A); } #ifdef __OPTIMIZE__ diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index 5595767bd98..487c9174716 100644 --- a/gcc/config/i386/sse.md +++ b/gcc/config/i386/sse.md @@ -1551,13 +1551,12 @@ [(set (match_operand:VF_128 0 "register_operand" "=v") (vec_merge:VF_128 (unspec:VF_128 - [(match_operand:VF_128 1 "register_operand" "v") - (match_operand:VF_128 2 "nonimmediate_operand" "vm")] + [(match_operand:VF_128 1 "nonimmediate_operand" "vm")] UNSPEC_RSQRT14) - (match_dup 1) + (match_operand:VF_128 2 "register_operand" "v") (const_int 1)))] "TARGET_AVX512F" - "vrsqrt14\t{%2, %1, %0|%0, %1, %2}" + "vrsqrt14\t{%1, %2, %0|%0, %2, %1}" [(set_attr "type" "sse") (set_attr "prefix" "evex") (set_attr "mode" "")]) @@ -12804,6 +12803,7 @@ "TARGET_AVX512ER" "vexp2\t{%1, %0|%0, %1}" [(set_attr "prefix" "evex") + (set_attr "type" "sse") (set_attr "mode" "")]) (define_insn "avx512er_rcp28" @@ -12814,6 +12814,7 @@ "TARGET_AVX512ER" "vrcp28\t{%1, %0|%0, %1}" [(set_attr "prefix" "evex") + (set_attr "type" "sse") (set_attr "mode" "")]) (define_insn "avx512er_vmrcp28" @@ -12825,9 +12826,10 @@ (match_operand:VF_128 2 "register_operand" "v") (const_int 1)))] "TARGET_AVX512ER" - "vrcp28\t{%2, %1, %0|%0, %1, %2}" + "vrcp28\t{%1, %2, %0|%0, %2, %1}" [(set_attr "length_immediate" "1") (set_attr "prefix" "evex") + (set_attr "type" "sse") (set_attr "mode" "")]) (define_insn "avx512er_rsqrt28" @@ -12838,6 +12840,7 @@ "TARGET_AVX512ER" "vrsqrt28\t{%1, %0|%0, %1}" [(set_attr "prefix" "evex") + (set_attr "type" "sse") (set_attr "mode" "")]) (define_insn "avx512er_vmrsqrt28" @@ -12849,8 +12852,9 @@ (match_operand:VF_128 2 "register_operand" "v") (const_int 1)))] "TARGET_AVX512ER" - "vrsqrt28\t{%2, %1, %0|%0, %1, %2}" + "vrsqrt28\t{%1, %2, %0|%0, %2, %1}" [(set_attr "length_immediate" "1") + (set_attr "type" "sse") (set_attr "prefix" "evex") (set_attr "mode" "")]) diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index fd796d7179f..5aad2d5649d 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,13 @@ +2014-02-20 Ilya Tocar + Kirill Yukhin + + * gcc.target/i386/avx512er-vrcp28sd-2.c: Distinguish src1 and src2. + * gcc.target/i386/avx512er-vrcp28ss-2.c: Call correct intrinsic. + * gcc.target/i386/avx512er-vrsqrt28sd-2.c: Distinguish src1 and src2. + * gcc.target/i386/avx512er-vrsqrt28ss-2.c: Ditto. + * gcc.target/i386/avx512f-vrcp14sd-2.c: Fix reference calculation. + * gcc.target/i386/avx512f-vrcp14ss-2.c: Fix reference calculation. + 2014-02-19 Jakub Jelinek PR c/37743 diff --git a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c b/gcc/testsuite/gcc.target/i386/avx512er-vrcp28sd-2.c similarity index 61% copy from gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c copy to gcc/testsuite/gcc.target/i386/avx512er-vrcp28sd-2.c index 1537a5932ca..889f990acfe 100644 --- a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512er-vrcp28sd-2.c @@ -10,19 +10,20 @@ void static avx512er_test (void) { - union128d src, res; + union128d src1, src2, res; double res_ref[2]; int i; for (i = 0; i < 2; i++) { - src.a[i] = 179.345 - 6.5645 * i; - res_ref[i] = src.a[i]; + src1.a[i] = 179.345 - 6.5645 * i; + src2.a[i] = 204179.345 + 6.5645 * i; + res_ref[i] = src1.a[i]; } - res_ref[0] = 1.0 / sqrt (src.a[0]); + res_ref[0] = 1.0 / src2.a[0]; - res.x = _mm_rsqrt28_round_sd (src.x, src.x, _MM_FROUND_NO_EXC); + res.x = _mm_rcp28_round_sd (src1.x, src2.x, _MM_FROUND_NO_EXC); if (checkVd (res.a, res_ref, 2)) abort (); diff --git a/gcc/testsuite/gcc.target/i386/avx512er-vrcp28ss-2.c b/gcc/testsuite/gcc.target/i386/avx512er-vrcp28ss-2.c index 499a9771e78..3280879107e 100644 --- a/gcc/testsuite/gcc.target/i386/avx512er-vrcp28ss-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512er-vrcp28ss-2.c @@ -10,19 +10,20 @@ void static avx512er_test (void) { - union128 src, res; + union128 src1, src2, res; float res_ref[4]; int i; for (i = 0; i < 4; i++) { - src.a[i] = 179.345 - 6.5645 * i; - res_ref[i] = src.a[i]; + src1.a[i] = 179.345 - 6.5645 * i; + src2.a[i] = 179345.006 + 6.5645 * i; + res_ref[i] = src1.a[i]; } - res_ref[0] = 1.0 / src.a[0]; + res_ref[0] = 1.0 / src2.a[0]; - res.x = _mm_rsqrt28_round_ss (src.x, src.x, _MM_FROUND_NO_EXC); + res.x = _mm_rcp28_round_ss (src1.x, src2.x, _MM_FROUND_NO_EXC); if (checkVf (res.a, res_ref, 4)) abort (); diff --git a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c b/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c index 1537a5932ca..bd217e8228f 100644 --- a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28sd-2.c @@ -10,19 +10,20 @@ void static avx512er_test (void) { - union128d src, res; + union128d src1, src2, res; double res_ref[2]; int i; for (i = 0; i < 2; i++) { - src.a[i] = 179.345 - 6.5645 * i; - res_ref[i] = src.a[i]; + src1.a[i] = 179.345 - 6.5645 * i; + src2.a[i] = 45 - 6.5645 * i; + res_ref[i] = src1.a[i]; } - res_ref[0] = 1.0 / sqrt (src.a[0]); + res_ref[0] = 1.0 / sqrt (src2.a[0]); - res.x = _mm_rsqrt28_round_sd (src.x, src.x, _MM_FROUND_NO_EXC); + res.x = _mm_rsqrt28_round_sd (src1.x, src2.x, _MM_FROUND_NO_EXC); if (checkVd (res.a, res_ref, 2)) abort (); diff --git a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28ss-2.c b/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28ss-2.c index f88422ea5ed..f7bfff5a50d 100644 --- a/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28ss-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512er-vrsqrt28ss-2.c @@ -10,19 +10,20 @@ void static avx512er_test (void) { - union128 src, res; + union128 src1, src2, res; float res_ref[4]; int i; for (i = 0; i < 4; i++) { - src.a[i] = 179.345 - 6.5645 * i; - res_ref[i] = src.a[i]; + src1.a[i] = 179.345 - 6.5645 * i; + src2.a[i] = 179221345 + 6.5645 * i; + res_ref[i] = src1.a[i]; } - res_ref[0] = 1.0 / sqrt (src.a[0]); + res_ref[0] = 1.0 / sqrt (src2.a[0]); - res.x = _mm_rsqrt28_round_ss (src.x, src.x, _MM_FROUND_NO_EXC); + res.x = _mm_rsqrt28_round_ss (src1.x, src2.x, _MM_FROUND_NO_EXC); if (checkVf (res.a, res_ref, 4)) abort (); diff --git a/gcc/testsuite/gcc.target/i386/avx512f-vrcp14sd-2.c b/gcc/testsuite/gcc.target/i386/avx512f-vrcp14sd-2.c index 0c9211a9a63..f94460036c0 100644 --- a/gcc/testsuite/gcc.target/i386/avx512f-vrcp14sd-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512f-vrcp14sd-2.c @@ -8,8 +8,8 @@ static void compute_vrcp14sd (double *s1, double *s2, double *r) { - r[0] = 1.0 / s1[0]; - r[1] = s2[1]; + r[0] = 1.0 / s2[0]; + r[1] = s1[1]; } static void diff --git a/gcc/testsuite/gcc.target/i386/avx512f-vrcp14ss-2.c b/gcc/testsuite/gcc.target/i386/avx512f-vrcp14ss-2.c index 3344dadb8ef..7aca591bfca 100644 --- a/gcc/testsuite/gcc.target/i386/avx512f-vrcp14ss-2.c +++ b/gcc/testsuite/gcc.target/i386/avx512f-vrcp14ss-2.c @@ -8,10 +8,10 @@ static void compute_vrcp14ss (float *s1, float *s2, float *r) { - r[0] = 1.0 / s1[0]; - r[1] = s2[1]; - r[2] = s2[2]; - r[3] = s2[3]; + r[0] = 1.0 / s2[0]; + r[1] = s1[1]; + r[2] = s1[2]; + r[3] = s1[3]; } static void -- 2.11.4.GIT