projects / gcc.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: 51e14b0)
AVX-512. 58/n. Add vpmul[u]dq insn patterns.
authorAlexander Ivchenko <alexander.ivchenko@intel.com>
Tue, 14 Oct 2014 08:13:31 +0000 (08:13 +0000)
committerKirill Yukhin <kyukhin@gcc.gnu.org>
Tue, 14 Oct 2014 08:13:31 +0000 (08:13 +0000)
gcc/
* config/i386/sse.md
(define_expand "vec_widen_umult_even_v8si<mask_name>"): Add masking.
(define_insn "*vec_widen_umult_even_v8si<mask_name>"): Ditto.
(define_expand "vec_widen_umult_even_v4si<mask_name>"): Ditto.
(define_insn "*vec_widen_umult_even_v4si<mask_name>"): Ditto.
(define_expand "vec_widen_smult_even_v8si<mask_name>"): Ditto.
(define_insn "*vec_widen_smult_even_v8si<mask_name>"): Ditto.
(define_expand "sse4_1_mulv2siv2di3<mask_name>"): Ditto.
(define_insn "*sse4_1_mulv2siv2di3<mask_name>"): Ditto.
(define_insn "avx512dq_mul<mode>3<mask_name>"): New.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r216176

gcc/ChangeLog patch | blob | history
gcc/config/i386/sse.md patch | blob | history

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 0a4673c98a8e0d35c860339c4957baa6c618350c..604e34c0aef980cdc36d036d07ba35ab4cd24b37 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,23 @@
+2014-10-14  Alexander Ivchenko  <alexander.ivchenko@intel.com>
+           Maxim Kuznetsov  <maxim.kuznetsov@intel.com>
+           Anna Tikhonova  <anna.tikhonova@intel.com>
+           Ilya Tocar  <ilya.tocar@intel.com>
+           Andrey Turetskiy  <andrey.turetskiy@intel.com>
+           Ilya Verbin  <ilya.verbin@intel.com>
+           Kirill Yukhin  <kirill.yukhin@intel.com>
+           Michael Zolotukhin  <michael.v.zolotukhin@intel.com>
+
+       * config/i386/sse.md
+       (define_expand "vec_widen_umult_even_v8si<mask_name>"): Add masking.
+       (define_insn "*vec_widen_umult_even_v8si<mask_name>"): Ditto.
+       (define_expand "vec_widen_umult_even_v4si<mask_name>"): Ditto.
+       (define_insn "*vec_widen_umult_even_v4si<mask_name>"): Ditto.
+       (define_expand "vec_widen_smult_even_v8si<mask_name>"): Ditto.
+       (define_insn "*vec_widen_smult_even_v8si<mask_name>"): Ditto.
+       (define_expand "sse4_1_mulv2siv2di3<mask_name>"): Ditto.
+       (define_insn "*sse4_1_mulv2siv2di3<mask_name>"): Ditto.
+       (define_insn "avx512dq_mul<mode>3<mask_name>"): New.
+
 2014-10-14  Alexander Ivchenko  <alexander.ivchenko@intel.com>
            Maxim Kuznetsov  <maxim.kuznetsov@intel.com>
            Anna Tikhonova  <anna.tikhonova@intel.com>
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index d544ed0676e990954e0000984b43e4b02987ca0a..70cfa6a99ab0cf4366e14a0c596c69b075f358cf 100644 (file)
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -9173,7 +9173,7 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "XI")])
 
-(define_expand "vec_widen_umult_even_v8si"
+(define_expand "vec_widen_umult_even_v8si<mask_name>"
   [(set (match_operand:V4DI 0 "register_operand")
        (mult:V4DI
          (zero_extend:V4DI
@@ -9186,29 +9186,30 @@
              (match_operand:V8SI 2 "nonimmediate_operand")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))))]
-  "TARGET_AVX2"
+  "TARGET_AVX2 && <mask_avx512vl_condition>"
   "ix86_fixup_binary_operands_no_copy (MULT, V8SImode, operands);")
 
-(define_insn "*vec_widen_umult_even_v8si"
-  [(set (match_operand:V4DI 0 "register_operand" "=x")
+(define_insn "*vec_widen_umult_even_v8si<mask_name>"
+  [(set (match_operand:V4DI 0 "register_operand" "=v")
        (mult:V4DI
          (zero_extend:V4DI
            (vec_select:V4SI
-             (match_operand:V8SI 1 "nonimmediate_operand" "%x")
+             (match_operand:V8SI 1 "nonimmediate_operand" "%v")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))
          (zero_extend:V4DI
            (vec_select:V4SI
-             (match_operand:V8SI 2 "nonimmediate_operand" "xm")
+             (match_operand:V8SI 2 "nonimmediate_operand" "vm")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))))]
-  "TARGET_AVX2 && ix86_binary_operator_ok (MULT, V8SImode, operands)"
-  "vpmuludq\t{%2, %1, %0|%0, %1, %2}"
+  "TARGET_AVX2 && <mask_avx512vl_condition>
+   && ix86_binary_operator_ok (MULT, V8SImode, operands)"
+  "vpmuludq\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
   [(set_attr "type" "sseimul")
-   (set_attr "prefix" "vex")
+   (set_attr "prefix" "maybe_evex")
    (set_attr "mode" "OI")])
 
-(define_expand "vec_widen_umult_even_v4si"
+(define_expand "vec_widen_umult_even_v4si<mask_name>"
   [(set (match_operand:V2DI 0 "register_operand")
        (mult:V2DI
          (zero_extend:V2DI
@@ -9219,28 +9220,29 @@
            (vec_select:V2SI
              (match_operand:V4SI 2 "nonimmediate_operand")
              (parallel [(const_int 0) (const_int 2)])))))]
-  "TARGET_SSE2"
+  "TARGET_SSE2 && <mask_avx512vl_condition>"
   "ix86_fixup_binary_operands_no_copy (MULT, V4SImode, operands);")
 
-(define_insn "*vec_widen_umult_even_v4si"
-  [(set (match_operand:V2DI 0 "register_operand" "=x,x")
+(define_insn "*vec_widen_umult_even_v4si<mask_name>"
+  [(set (match_operand:V2DI 0 "register_operand" "=x,v")
        (mult:V2DI
          (zero_extend:V2DI
            (vec_select:V2SI
-             (match_operand:V4SI 1 "nonimmediate_operand" "%0,x")
+             (match_operand:V4SI 1 "nonimmediate_operand" "%0,v")
              (parallel [(const_int 0) (const_int 2)])))
          (zero_extend:V2DI
            (vec_select:V2SI
-             (match_operand:V4SI 2 "nonimmediate_operand" "xm,xm")
+             (match_operand:V4SI 2 "nonimmediate_operand" "xm,vm")
              (parallel [(const_int 0) (const_int 2)])))))]
-  "TARGET_SSE2 && ix86_binary_operator_ok (MULT, V4SImode, operands)"
+  "TARGET_SSE2 && <mask_avx512vl_condition>
+   && ix86_binary_operator_ok (MULT, V4SImode, operands)"
   "@
    pmuludq\t{%2, %0|%0, %2}
-   vpmuludq\t{%2, %1, %0|%0, %1, %2}"
+   vpmuludq\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
   [(set_attr "isa" "noavx,avx")
    (set_attr "type" "sseimul")
    (set_attr "prefix_data16" "1,*")
-   (set_attr "prefix" "orig,vex")
+   (set_attr "prefix" "orig,maybe_evex")
    (set_attr "mode" "TI")])
 
 (define_expand "vec_widen_smult_even_v16si<mask_name>"
@@ -9288,7 +9290,7 @@
    (set_attr "prefix" "evex")
    (set_attr "mode" "XI")])
 
-(define_expand "vec_widen_smult_even_v8si"
+(define_expand "vec_widen_smult_even_v8si<mask_name>"
   [(set (match_operand:V4DI 0 "register_operand")
        (mult:V4DI
          (sign_extend:V4DI
@@ -9301,30 +9303,31 @@
              (match_operand:V8SI 2 "nonimmediate_operand")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))))]
-  "TARGET_AVX2"
+  "TARGET_AVX2 && <mask_avx512vl_condition>"
   "ix86_fixup_binary_operands_no_copy (MULT, V8SImode, operands);")
 
-(define_insn "*vec_widen_smult_even_v8si"
-  [(set (match_operand:V4DI 0 "register_operand" "=x")
+(define_insn "*vec_widen_smult_even_v8si<mask_name>"
+  [(set (match_operand:V4DI 0 "register_operand" "=v")
        (mult:V4DI
          (sign_extend:V4DI
            (vec_select:V4SI
-             (match_operand:V8SI 1 "nonimmediate_operand" "x")
+             (match_operand:V8SI 1 "nonimmediate_operand" "v")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))
          (sign_extend:V4DI
            (vec_select:V4SI
-             (match_operand:V8SI 2 "nonimmediate_operand" "xm")
+             (match_operand:V8SI 2 "nonimmediate_operand" "vm")
              (parallel [(const_int 0) (const_int 2)
                         (const_int 4) (const_int 6)])))))]
-  "TARGET_AVX2 && ix86_binary_operator_ok (MULT, V8SImode, operands)"
-  "vpmuldq\t{%2, %1, %0|%0, %1, %2}"
+  "TARGET_AVX2
+   && ix86_binary_operator_ok (MULT, V8SImode, operands)"
+  "vpmuldq\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
   [(set_attr "type" "sseimul")
    (set_attr "prefix_extra" "1")
    (set_attr "prefix" "vex")
    (set_attr "mode" "OI")])
 
-(define_expand "sse4_1_mulv2siv2di3"
+(define_expand "sse4_1_mulv2siv2di3<mask_name>"
   [(set (match_operand:V2DI 0 "register_operand")
        (mult:V2DI
          (sign_extend:V2DI
@@ -9335,24 +9338,25 @@
            (vec_select:V2SI
              (match_operand:V4SI 2 "nonimmediate_operand")
              (parallel [(const_int 0) (const_int 2)])))))]
-  "TARGET_SSE4_1"
+  "TARGET_SSE4_1 && <mask_avx512vl_condition>"
   "ix86_fixup_binary_operands_no_copy (MULT, V4SImode, operands);")
 
-(define_insn "*sse4_1_mulv2siv2di3"
-  [(set (match_operand:V2DI 0 "register_operand" "=x,x")
+(define_insn "*sse4_1_mulv2siv2di3<mask_name>"
+  [(set (match_operand:V2DI 0 "register_operand" "=x,v")
        (mult:V2DI
          (sign_extend:V2DI
            (vec_select:V2SI
-             (match_operand:V4SI 1 "nonimmediate_operand" "%0,x")
+             (match_operand:V4SI 1 "nonimmediate_operand" "%0,v")
              (parallel [(const_int 0) (const_int 2)])))
          (sign_extend:V2DI
            (vec_select:V2SI
-             (match_operand:V4SI 2 "nonimmediate_operand" "xm,xm")
+             (match_operand:V4SI 2 "nonimmediate_operand" "xm,vm")
              (parallel [(const_int 0) (const_int 2)])))))]
-  "TARGET_SSE4_1 && ix86_binary_operator_ok (MULT, V4SImode, operands)"
+  "TARGET_SSE4_1 && <mask_avx512vl_condition>
+   && ix86_binary_operator_ok (MULT, V4SImode, operands)"
   "@
    pmuldq\t{%2, %0|%0, %2}
-   vpmuldq\t{%2, %1, %0|%0, %1, %2}"
+   vpmuldq\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
   [(set_attr "isa" "noavx,avx")
    (set_attr "type" "sseimul")
    (set_attr "prefix_data16" "1,*")
@@ -9491,6 +9495,17 @@
    (set_attr "prefix" "orig,vex")
    (set_attr "mode" "TI")])
 
+(define_insn "avx512dq_mul<mode>3<mask_name>"
+  [(set (match_operand:VI8 0 "register_operand" "=v")
+       (mult:VI8
+         (match_operand:VI8 1 "register_operand" "v")
+         (match_operand:VI8 2 "nonimmediate_operand" "vm")))]
+  "TARGET_AVX512DQ && <mask_mode512bit_condition>"
+  "vpmullq\t{%2, %1, %0<mask_operand3>|%0<mask_operand3>, %1, %2}"
+  [(set_attr "type" "sseimul")
+   (set_attr "prefix" "evex")
+   (set_attr "mode" "<sseinsnmode>")])
+
 (define_expand "mul<mode>3<mask_name>"
   [(set (match_operand:VI4_AVX512F 0 "register_operand")
        (mult:VI4_AVX512F