From: Hongtao Liu <crazylht@gmail.com>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: [PATCH] [AVX512] [PR87767] Optimize memory broadcast for constant vector under AVX512
Date: Thu, 9 Jul 2020 16:33:46 +0800 [thread overview]
Message-ID: <CAMZc-bxdgUxqd=ZRyQ0ciPpLr=Deg=U8vhERov0f+B2Rgm17yA@mail.gmail.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 1200 bytes --]
Hi:
For a constant vector having one duplicated value, there's no need
to put the whole vector in the constant pool, using embedded broadcast
instead.
Bootstrap test is Ok, regression test for i386/x86-64 backend is ok.
gcc/ChangeLog:
PR target/87767
* config/i386/i386-features.c
(replace_constant_pool_with_broadcast): New function.
(constant_pool_broadcast): Ditto.
(class pass_constant_pool_broadcast): New pass.
(make_pass_constant_pool_broadcast): Ditto.
* config/i386/i386-passes.def: Insert new pass after combine.
* config/i386/i386-protos.h
(make_pass_constant_pool_broadcast): Declare.
* config/i386/sse.md (*avx512dq_mul<mode>3<mask_name>_bcst,
*avx512f_mul<mode>3<mask_name>_bcst): New define_insn.
gcc/testsuite/ChangeLog:
PR target/87767
* gcc.target/i386/avx2-broadcast-pr87767-1.c: New test.
* gcc.target/i386/avx512f-broadcast-pr87767-1.c: New test.
* gcc.target/i386/avx512f-broadcast-pr87767-2.c: New test.
* gcc.target/i386/avx512vl-broadcast-pr87767-1.c: New test.
* gcc.target/i386/pr92865-1.c: Adjust testcase.
--
BR,
Hongtao
[-- Attachment #2: 0001-Optimize-memory-broadcast-for-constant-vector-under-.patch --]
[-- Type: text/x-patch, Size: 17898 bytes --]
From b8f49299e3d23f927a659cd394e3099e3291a76f Mon Sep 17 00:00:00 2001
From: liuhongt <hongtao.liu@intel.com>
Date: Wed, 8 Jul 2020 17:14:36 +0800
Subject: [PATCH] Optimize memory broadcast for constant vector under AVX512.
For constant vector having one duplicated value, there's no need to put
whole vector in the constant pool, using embedded broadcast instead.
2020-07-09 Hongtao Liu <hongtao.liu@intel.com>
gcc/ChangeLog:
PR target/87767
* config/i386/i386-features.c
(replace_constant_pool_with_broadcast): New function.
(constant_pool_broadcast): Ditto.
(class pass_constant_pool_broadcast): New pass.
(make_pass_constant_pool_broadcast): Ditto.
* config/i386/i386-passes.def: Insert new pass after combine.
* config/i386/i386-protos.h
(make_pass_constant_pool_broadcast): Declare.
* config/i386/sse.md (*avx512dq_mul<mode>3<mask_name>_bcst,
*avx512f_mul<mode>3<mask_name>_bcst): New define_insn.
gcc/testsuite/ChangeLog:
PR target/87767
* gcc.target/i386/avx2-broadcast-pr87767-1.c: New test.
* gcc.target/i386/avx512f-broadcast-pr87767-1.c: New test.
* gcc.target/i386/avx512f-broadcast-pr87767-2.c: New test.
* gcc.target/i386/avx512vl-broadcast-pr87767-1.c: New test.
* gcc.target/i386/pr92865-1.c: Adjust testcase.
---
gcc/config/i386/i386-features.c | 146 ++++++++++++++++++
gcc/config/i386/i386-passes.def | 1 +
gcc/config/i386/i386-protos.h | 1 +
gcc/config/i386/sse.md | 25 +++
.../i386/avx2-broadcast-pr87767-1.c | 40 +++++
.../i386/avx512f-broadcast-pr87767-1.c | 66 ++++++++
.../i386/avx512f-broadcast-pr87767-2.c | 54 +++++++
.../i386/avx512vl-broadcast-pr87767-1.c | 40 +++++
gcc/testsuite/gcc.target/i386/pr92865-1.c | 9 +-
9 files changed, 378 insertions(+), 4 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/avx2-broadcast-pr87767-1.c
create mode 100644 gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-1.c
create mode 100644 gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-2.c
create mode 100644 gcc/testsuite/gcc.target/i386/avx512vl-broadcast-pr87767-1.c
diff --git a/gcc/config/i386/i386-features.c b/gcc/config/i386/i386-features.c
index 535fc7e981d..8f81d101382 100644
--- a/gcc/config/i386/i386-features.c
+++ b/gcc/config/i386/i386-features.c
@@ -2379,6 +2379,152 @@ make_pass_remove_partial_avx_dependency (gcc::context *ctxt)
return new pass_remove_partial_avx_dependency (ctxt);
}
+/* Replace all one-value const vector that are referenced by SYMBOL_REFs in x
+ with embedded broadcast. i.e.transform
+
+ vpaddq .LC0(%rip), %zmm0, %zmm0
+ ret
+ .LC0:
+ .quad 3
+ .quad 3
+ .quad 3
+ .quad 3
+ .quad 3
+ .quad 3
+ .quad 3
+ .quad 3
+
+ to
+
+ vpaddq .LC0(%rip){1to8}, %zmm0, %zmm0
+ ret
+ .LC0:
+ .quad 3 */
+static void
+replace_constant_pool_with_broadcast (rtx_insn* insn)
+{
+ subrtx_ptr_iterator::array_type array;
+ FOR_EACH_SUBRTX_PTR (iter, array, &PATTERN (insn), ALL)
+ {
+ rtx *loc = *iter;
+ rtx x = *loc;
+ rtx broadcast_mem, vec_dup, constant, first;
+ machine_mode mode;
+ if (GET_CODE (x) != MEM
+ || GET_CODE (XEXP (x, 0)) != SYMBOL_REF
+ || !CONSTANT_POOL_ADDRESS_P (XEXP (x, 0)))
+ continue;
+
+ mode = GET_MODE (x);
+ if (!VECTOR_MODE_P (mode))
+ return;
+
+ constant = get_pool_constant (XEXP (x, 0));
+ first = XVECEXP (constant, 0, 0);
+ /* There could be some rtx like
+ (mem/u/c:V16QI (symbol_ref/u:DI ("*.LC1")))
+ but with "*.LC1" refer to V2DI constant vector. */
+ if (GET_MODE (constant) != mode)
+ return;
+
+ for (int i = 1; i < GET_MODE_NUNITS (mode); ++i)
+ {
+ rtx tmp = XVECEXP (constant, 0, i);
+ /* Only handle one-value const vector. */
+ if (!rtx_equal_p (tmp, first))
+ return;
+ }
+
+ broadcast_mem = force_const_mem (GET_MODE_INNER (mode), first);
+ vec_dup = gen_rtx_VEC_DUPLICATE (mode, broadcast_mem);
+ *loc = vec_dup;
+ INSN_CODE (insn) = -1;
+ /* Revert change if there's no corresponding pattern. */
+ if (recog_memoized (insn) < 0)
+ {
+ *loc = x;
+ recog_memoized (insn);
+ }
+ /* At most 1 memory_operand in an insn. */
+ return;
+ }
+}
+
+/* For const vector having one duplicated value, there's no need to put
+ whole vector in the constant pool when target supports embedded broadcast. */
+static unsigned int
+constant_pool_broadcast (void)
+{
+ timevar_push (TV_MACH_DEP);
+ rtx_insn *insn;
+
+ for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
+ {
+ if (!INSN_P (insn))
+ continue;
+
+ /* Insns may appear inside a SEQUENCE. Only check the patterns of
+ insns, not any notes that may be attached. We don't want to mark
+ a constant just because it happens to appear in a REG_EQUIV note. */
+ if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (PATTERN (insn)))
+ {
+ int i, n = seq->len ();
+ for (i = 0; i < n; ++i)
+ {
+ rtx subinsn = seq->element (i);
+ if (INSN_P (subinsn))
+ replace_constant_pool_with_broadcast (dyn_cast <rtx_insn *> (subinsn));
+ }
+ }
+ else
+ replace_constant_pool_with_broadcast (insn);
+ }
+ timevar_pop (TV_MACH_DEP);
+ return 0;
+}
+
+namespace {
+
+const pass_data pass_data_constant_pool_broadcast =
+{
+ RTL_PASS, /* type */
+ "cpb", /* name */
+ OPTGROUP_NONE, /* optinfo_flags */
+ TV_MACH_DEP, /* tv_id */
+ 0, /* properties_required */
+ 0, /* properties_provided */
+ 0, /* properties_destroyed */
+ 0, /* todo_flags_start */
+ TODO_df_finish, /* todo_flags_finish */
+};
+
+class pass_constant_pool_broadcast : public rtl_opt_pass
+{
+public:
+ pass_constant_pool_broadcast (gcc::context *ctxt)
+ : rtl_opt_pass (pass_data_constant_pool_broadcast, ctxt)
+ {}
+
+ /* opt_pass methods: */
+ virtual bool gate (function *)
+ {
+ return TARGET_AVX512F;
+ }
+
+ virtual unsigned int execute (function *)
+ {
+ return constant_pool_broadcast ();
+ }
+}; // class pass_cpb
+
+} // anon namespace
+
+rtl_opt_pass *
+make_pass_constant_pool_broadcast (gcc::context *ctxt)
+{
+ return new pass_constant_pool_broadcast (ctxt);
+}
+
/* This compares the priority of target features in function DECL1
and DECL2. It returns positive value if DECL1 is higher priority,
negative value if DECL2 is higher priority and 0 if they are the
diff --git a/gcc/config/i386/i386-passes.def b/gcc/config/i386/i386-passes.def
index d83c7b956b1..07ecf8e790f 100644
--- a/gcc/config/i386/i386-passes.def
+++ b/gcc/config/i386/i386-passes.def
@@ -33,3 +33,4 @@ along with GCC; see the file COPYING3. If not see
INSERT_PASS_BEFORE (pass_shorten_branches, 1, pass_insert_endbr_and_patchable_area);
INSERT_PASS_AFTER (pass_combine, 1, pass_remove_partial_avx_dependency);
+ INSERT_PASS_AFTER (pass_combine, 1, pass_constant_pool_broadcast);
diff --git a/gcc/config/i386/i386-protos.h b/gcc/config/i386/i386-protos.h
index 7c2ce618f3f..6c6909b41dd 100644
--- a/gcc/config/i386/i386-protos.h
+++ b/gcc/config/i386/i386-protos.h
@@ -386,3 +386,4 @@ extern rtl_opt_pass *make_pass_insert_endbr_and_patchable_area
(gcc::context *);
extern rtl_opt_pass *make_pass_remove_partial_avx_dependency
(gcc::context *);
+extern rtl_opt_pass *make_pass_constant_pool_broadcast (gcc::context *);
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 431571a4bc1..fbfb459c5bf 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -12127,6 +12127,19 @@
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
+(define_insn "*avx512dq_mul<mode>3<mask_name>_bcst"
+ [(set (match_operand:VI8_AVX512VL 0 "register_operand" "=v")
+ (mult:VI8_AVX512VL
+ (vec_duplicate:VI8_AVX512VL
+ (match_operand:<ssescalarmode> 1 "memory_operand" "m"))
+ (match_operand:VI8_AVX512VL 2 "register_operand" "v")
+))]
+ "TARGET_AVX512DQ"
+ "vpmullq\t{%1<avx512bcst>, %2, %0<mask_operand3>|%0<mask_operand3>, %2, %1<avx512bcst>}"
+ [(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
@@ -12167,6 +12180,18 @@
(set_attr "btver2_decode" "vector,vector,vector")
(set_attr "mode" "<sseinsnmode>")])
+(define_insn "*avx512f_mul<mode>3<mask_name>_bcst"
+ [(set (match_operand:VI4_AVX512VL 0 "register_operand" "=v")
+ (mult:VI4_AVX512VL
+ (vec_duplicate:VI4_AVX512VL
+ (match_operand:<ssescalarmode> 1 "memory_operand" "m"))
+ (match_operand:VI4_AVX512VL 2 "register_operand" "v")))]
+ "TARGET_AVX512F"
+ "vpmulld\t{%1<avx512bcst>, %2, %0<mask_operand3>|%0<mask_operand3>, %2, %1<avx512bcst>}"
+ [(set_attr "type" "sseimul")
+ (set_attr "prefix" "evex")
+ (set_attr "mode" "<sseinsnmode>")])
+
(define_expand "mul<mode>3"
[(set (match_operand:VI8_AVX2_AVX512F 0 "register_operand")
(mult:VI8_AVX2_AVX512F
diff --git a/gcc/testsuite/gcc.target/i386/avx2-broadcast-pr87767-1.c b/gcc/testsuite/gcc.target/i386/avx2-broadcast-pr87767-1.c
new file mode 100644
index 00000000000..800ef1f957e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx2-broadcast-pr87767-1.c
@@ -0,0 +1,40 @@
+/* PR target/87767 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx2" } */
+/* { dg-final { scan-assembler-not "\\\{1to\[248\]\\\}" } } */
+/* { dg-final { scan-assembler-not "\\\{1to16\\\}" } } */
+
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef int v8si __attribute__ ((vector_size (32)));
+typedef long long v2di __attribute__ ((vector_size (16)));
+typedef long long v4di __attribute__ ((vector_size (32)));
+typedef float v4sf __attribute__ ((vector_size (16)));
+typedef float v8sf __attribute__ ((vector_size (32)));
+typedef double v2df __attribute__ ((vector_size (16)));
+typedef double v4df __attribute__ ((vector_size (32)));
+
+#define FOO(VTYPE, OP_NAME, OP) \
+VTYPE \
+ __attribute__ ((noipa)) \
+foo_##OP_NAME##_##VTYPE (VTYPE a) \
+{ \
+ return a OP 101; \
+} \
+
+FOO (v4si, add, +);
+FOO (v8si, add, +);
+FOO (v2di, add, +);
+FOO (v4di, add, +);
+FOO (v4sf, add, +);
+FOO (v8sf, add, +);
+FOO (v2df, add, +);
+FOO (v4df, add, +);
+
+FOO (v4si, mul, *);
+FOO (v8si, mul, *);
+FOO (v2di, mul, *);
+FOO (v4di, mul, *);
+FOO (v4sf, mul, *);
+FOO (v8sf, mul, *);
+FOO (v2df, mul, *);
+FOO (v4df, mul, *);
diff --git a/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-1.c b/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-1.c
new file mode 100644
index 00000000000..21249bc0cf9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-1.c
@@ -0,0 +1,66 @@
+/* PR target/87767 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512f -mavx512vl -mavx512dq" } */
+/* { dg-final { scan-assembler-times "vpadd\[^\n\]*\\\{1to2\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpadd\[^\n\]*\\\{1to4\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpadd\[^\n\]*\\\{1to8\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpadd\[^\n\]*\\\{1to16\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmul\[^\n\]*\\\{1to2\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vpmul\[^\n\]*\\\{1to4\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpmul\[^\n\]*\\\{1to8\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vpmul\[^\n\]*\\\{1to16\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vadd\[^\n\]*\\\{1to2\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vadd\[^\n\]*\\\{1to4\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vadd\[^\n\]*\\\{1to8\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vadd\[^\n\]*\\\{1to16\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmul\[^\n\]*\\\{1to2\\\}" 1 } } */
+/* { dg-final { scan-assembler-times "vmul\[^\n\]*\\\{1to4\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vmul\[^\n\]*\\\{1to8\\\}" 2 } } */
+/* { dg-final { scan-assembler-times "vmul\[^\n\]*\\\{1to16\\\}" 1 } } */
+
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef int v8si __attribute__ ((vector_size (32)));
+typedef int v16si __attribute__ ((vector_size (64)));
+typedef long long v2di __attribute__ ((vector_size (16)));
+typedef long long v4di __attribute__ ((vector_size (32)));
+typedef long long v8di __attribute__ ((vector_size (64)));
+typedef float v4sf __attribute__ ((vector_size (16)));
+typedef float v8sf __attribute__ ((vector_size (32)));
+typedef float v16sf __attribute__ ((vector_size (64)));
+typedef double v2df __attribute__ ((vector_size (16)));
+typedef double v4df __attribute__ ((vector_size (32)));
+typedef double v8df __attribute__ ((vector_size (64)));
+
+#define FOO(VTYPE, OP_NAME, OP) \
+VTYPE \
+ __attribute__ ((noipa)) \
+foo_##OP_NAME##_##VTYPE (VTYPE a) \
+{ \
+ return a OP 101; \
+} \
+
+FOO (v4si, add, +);
+FOO (v8si, add, +);
+FOO (v16si, add, +);
+FOO (v2di, add, +);
+FOO (v4di, add, +);
+FOO (v8di, add, +);
+FOO (v4sf, add, +);
+FOO (v8sf, add, +);
+FOO (v16sf, add, +);
+FOO (v2df, add, +);
+FOO (v4df, add, +);
+FOO (v8df, add, +);
+
+FOO (v4si, mul, *);
+FOO (v8si, mul, *);
+FOO (v16si, mul, *);
+FOO (v2di, mul, *);
+FOO (v4di, mul, *);
+FOO (v8di, mul, *);
+FOO (v4sf, mul, *);
+FOO (v8sf, mul, *);
+FOO (v16sf, mul, *);
+FOO (v2df, mul, *);
+FOO (v4df, mul, *);
+FOO (v8df, mul, *);
diff --git a/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-2.c b/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-2.c
new file mode 100644
index 00000000000..938346743c2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512f-broadcast-pr87767-2.c
@@ -0,0 +1,54 @@
+/* PR target/87767 */
+/* { dg-do run } */
+/* { dg-options "-O2 -mavx512f -mavx512vl -mavx512dq" } */
+
+#include<stdlib.h>
+#include<stdio.h>
+#include "avx512f-broadcast-pr87767-1.c"
+
+#define TEST(VTYPE, TYPE, N, OP_NAME, OP) \
+ do \
+ { \
+ TYPE exp[N], src[N]; \
+ VTYPE res; \
+ for (int i = 0; i < N; i++) \
+ src[i] = i * i * 107; \
+ res = foo_##OP_NAME##_##VTYPE (*(VTYPE*)&src[0]); \
+ for (int i = 0; i < N; i ++) \
+ exp[i] = src[i] OP 101; \
+ for (int j = 0; j < N; j++) \
+ { \
+ if (res[j] != exp[j]) \
+ abort(); \
+ } \
+ } \
+ while (0)
+
+int main()
+{
+ TEST (v4si, int, 4, add, +);
+ TEST (v8si, int, 8, add, +);
+ TEST (v16si, int, 16, add, +);
+ TEST (v2di, long long, 2, add, +);
+ TEST (v4di, long long, 4, add, +);
+ TEST (v8di, long long, 8, add, +);
+ TEST (v4sf, float, 4, add, +);
+ TEST (v8sf, float, 8, add, +);
+ TEST (v16sf, float, 16, add, +);
+ TEST (v2df, double, 2, add, +);
+ TEST (v4df, double, 4, add, +);
+ TEST (v8df, double, 8, add, +);
+
+ TEST (v4si, int, 4, mul, *);
+ TEST (v8si, int, 8, mul, *);
+ TEST (v16si, int, 16, mul, *);
+ TEST (v2di, long long, 2, mul, *);
+ TEST (v4di, long long, 4, mul, *);
+ TEST (v8di, long long, 8, mul, *);
+ TEST (v4sf, float, 4, mul, *);
+ TEST (v8sf, float, 8, mul, *);
+ TEST (v16sf, float, 16, mul, *);
+ TEST (v2df, double, 2, mul, *);
+ TEST (v4df, double, 4, mul, *);
+ TEST (v8df, double, 8, mul, *);
+}
diff --git a/gcc/testsuite/gcc.target/i386/avx512vl-broadcast-pr87767-1.c b/gcc/testsuite/gcc.target/i386/avx512vl-broadcast-pr87767-1.c
new file mode 100644
index 00000000000..ec159a68158
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/avx512vl-broadcast-pr87767-1.c
@@ -0,0 +1,40 @@
+/* PR target/87767 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mavx512f" } */
+/* { dg-final { scan-assembler-not "\\\{1to\[248\]\\\}" } } */
+/* { dg-final { scan-assembler-not "\\\{1to16\\\}" } } */
+
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef int v8si __attribute__ ((vector_size (32)));
+typedef long long v2di __attribute__ ((vector_size (16)));
+typedef long long v4di __attribute__ ((vector_size (32)));
+typedef float v4sf __attribute__ ((vector_size (16)));
+typedef float v8sf __attribute__ ((vector_size (32)));
+typedef double v2df __attribute__ ((vector_size (16)));
+typedef double v4df __attribute__ ((vector_size (32)));
+
+#define FOO(VTYPE, OP_NAME, OP) \
+VTYPE \
+ __attribute__ ((noipa)) \
+foo_##OP_NAME##_##VTYPE (VTYPE a) \
+{ \
+ return a OP 101; \
+} \
+
+FOO (v4si, add, +);
+FOO (v8si, add, +);
+FOO (v2di, add, +);
+FOO (v4di, add, +);
+FOO (v4sf, add, +);
+FOO (v8sf, add, +);
+FOO (v2df, add, +);
+FOO (v4df, add, +);
+
+FOO (v4si, mul, *);
+FOO (v8si, mul, *);
+FOO (v2di, mul, *);
+FOO (v4di, mul, *);
+FOO (v4sf, mul, *);
+FOO (v8sf, mul, *);
+FOO (v2df, mul, *);
+FOO (v4df, mul, *);
diff --git a/gcc/testsuite/gcc.target/i386/pr92865-1.c b/gcc/testsuite/gcc.target/i386/pr92865-1.c
index 49b5778a067..a37487d9af7 100644
--- a/gcc/testsuite/gcc.target/i386/pr92865-1.c
+++ b/gcc/testsuite/gcc.target/i386/pr92865-1.c
@@ -3,10 +3,11 @@
/* { dg-options "-Ofast -mavx512f -mavx512bw -mxop" } */
/* { dg-final { scan-assembler-times "vpcmp\[bwdq\]\[\t ]" 4 } } */
/* { dg-final { scan-assembler-times "vpcmpu\[bwdq\]\[\t ]" 4 } } */
-/* { dg-final { scan-assembler-times "vmovdq\[au\]8\[\t ]" 4 } } */
-/* { dg-final { scan-assembler-times "vmovdq\[au\]16\[\t ]" 4 } } *
-/* { dg-final { scan-assembler-times "vmovdq\[au\]32\[\t ]" 4 } } */
-/* { dg-final { scan-assembler-times "vmovdq\[au\]64\[\t ]" 4 } } */
+/* { dg-final { scan-assembler-times "vmovdq\[au\]8\[\t ]" 2 } } */
+/* { dg-final { scan-assembler-times "vmovdq\[au\]16\[\t ]" 2 } } *
+/* { dg-final { scan-assembler-times "vmovdq\[au\]32\[\t ]" 2 } } */
+/* { dg-final { scan-assembler-times "vmovdq\[au\]64\[\t ]" 2 } } */
+/* { dg-final { scan-assembler-times "vpbroadcast\[bwqd\]\[\t ]" 16 } } */
extern char arraysb[64];
extern short arraysw[32];
--
2.18.1
next reply other threads:[~2020-07-09 8:33 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-09 8:33 Hongtao Liu [this message]
2020-07-10 9:24 ` Hongtao Liu
2020-07-17 7:24 ` Hongtao Liu
2020-07-23 8:39 ` Jan Hubicka
2020-07-23 13:53 ` Hongtao Liu
2020-07-24 2:37 ` Hongtao Liu
2020-08-04 6:05 ` Hongtao Liu
2020-08-26 21:23 ` Jeff Law
2020-08-27 11:09 ` Jan Hubicka
2020-08-27 12:24 ` Jakub Jelinek
2020-08-27 13:07 ` Richard Biener
2020-08-27 13:20 ` Jakub Jelinek
2020-08-28 6:47 ` Richard Biener
2020-08-28 8:52 ` Jakub Jelinek
2020-08-28 10:36 ` Richard Biener
2020-08-28 10:47 ` Jakub Jelinek
2020-08-28 11:06 ` Richard Biener
2020-08-28 11:26 ` Jakub Jelinek
2020-08-28 14:53 ` Jakub Jelinek
2020-08-28 16:07 ` Richard Sandiford
2020-08-28 16:25 ` Jakub Jelinek
2020-08-30 9:24 ` Jakub Jelinek
2020-08-31 8:18 ` Richard Biener
2020-08-28 17:18 ` Hongtao Liu
2020-09-01 9:55 ` Hongtao Liu
2020-09-01 10:11 ` Jakub Jelinek
2020-09-02 1:57 ` Hongtao Liu
2020-09-02 9:58 ` Jakub Jelinek
2020-09-03 2:11 ` Hongtao Liu
2020-09-03 7:27 ` Jakub Jelinek
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAMZc-bxdgUxqd=ZRyQ0ciPpLr=Deg=U8vhERov0f+B2Rgm17yA@mail.gmail.com' \
--to=crazylht@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).