Support commutative alternative for AVX512 vpcmpeq{b,w,d,q}
authorliuhongt <hongtao.liu@intel.com>
Thu, 6 Jan 2022 07:33:20 +0000 (15:33 +0800)
committerliuhongt <hongtao.liu@intel.com>
Fri, 7 Jan 2022 03:13:00 +0000 (11:13 +0800)
gcc/ChangeLog:

* config/i386/sse.md
(*<avx512>_eq<mode>3<mask_scalar_merge_name>_1): Extend to
UNSPEC_PCMP_UNSIGNED.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr103774.c: New test.
* gcc.target/i386/avx512bw-vpcmpequb-1.c: Adjust scan assembler
from vpcmpub to (?:vpcmpub|vpcmpeqb).
* gcc.target/i386/avx512bw-vpcmpequw-1.c: Ditto.
* gcc.target/i386/avx512bw-vpcmpub-1.c: Ditto.
* gcc.target/i386/avx512bw-vpcmpuw-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpequd-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpequq-1.c: Ditto.
* gcc.target/i386/avx512f-vpcmpud-1.c: Ditto.
* gcc.target/i386/avx512vl-vpcmpequd-1.c: Ditto.
* gcc.target/i386/avx512vl-vpcmpequq-1.c: Ditto.
* gcc.target/i386/avx512vl-vpcmpuq-1.c: Ditto.

12 files changed:
gcc/config/i386/sse.md
gcc/testsuite/gcc.target/i386/avx512bw-vpcmpequb-1.c
gcc/testsuite/gcc.target/i386/avx512bw-vpcmpequw-1.c
gcc/testsuite/gcc.target/i386/avx512bw-vpcmpub-1.c
gcc/testsuite/gcc.target/i386/avx512bw-vpcmpuw-1.c
gcc/testsuite/gcc.target/i386/avx512f-vpcmpequd-1.c
gcc/testsuite/gcc.target/i386/avx512f-vpcmpequq-1.c
gcc/testsuite/gcc.target/i386/avx512f-vpcmpud-1.c
gcc/testsuite/gcc.target/i386/avx512vl-vpcmpequd-1.c
gcc/testsuite/gcc.target/i386/avx512vl-vpcmpequq-1.c
gcc/testsuite/gcc.target/i386/avx512vl-vpcmpuq-1.c
gcc/testsuite/gcc.target/i386/pr103774.c [new file with mode: 0644]

index 6c1e94f..d8f3035 100644 (file)
           UNSPEC_PCMP_ITER))]
   "operands[4] = GEN_INT (INTVAL (operands[3]) ^ 4);")
 
+(define_insn "*<avx512>_eq<mode>3<mask_scalar_merge_name>_1"
+  [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k,k")
+       (unspec:<avx512fmaskmode>
+         [(match_operand:VI12_AVX512VL 1 "nonimm_or_0_operand" "%v,v")
+          (match_operand:VI12_AVX512VL 2 "nonimm_or_0_operand" "vm,C")
+          (const_int 0)]
+         UNSPEC_PCMP_ITER))]
+  "TARGET_AVX512BW && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
+  "@
+   vpcmpeq<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}
+   vptestnm<ssemodesuffix>\t{%1, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %1}"
+  [(set_attr "type" "ssecmp")
+   (set_attr "prefix_extra" "1")
+   (set_attr "prefix" "evex")
+   (set_attr "mode" "<sseinsnmode>")])
+
 (define_insn "<avx512>_ucmp<mode>3<mask_scalar_merge_name>"
   [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k")
        (unspec:<avx512fmaskmode>
    (set_attr "prefix" "evex")
    (set_attr "mode" "<VI12_AVX512VL:sseinsnmode>")])
 
+(define_insn "*<avx512>_eq<mode>3<mask_scalar_merge_name>_1"
+  [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k,k")
+       (unspec:<avx512fmaskmode>
+         [(match_operand:VI48_AVX512VL 1 "nonimm_or_0_operand" "%v,v")
+          (match_operand:VI48_AVX512VL 2 "nonimm_or_0_operand" "vm,C")
+          (const_int 0)]
+         UNSPEC_PCMP_ITER))]
+  "TARGET_AVX512F && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
+  "@
+   vpcmpeq<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}
+   vptestnm<ssemodesuffix>\t{%1, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %1}"
+  [(set_attr "type" "ssecmp")
+   (set_attr "prefix_extra" "1")
+   (set_attr "prefix" "evex")
+   (set_attr "mode" "<sseinsnmode>")])
+
 (define_insn "<avx512>_ucmp<mode>3<mask_scalar_merge_name>"
   [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k")
        (unspec:<avx512fmaskmode>
   "TARGET_AVX512F"
   "ix86_fixup_binary_operands_no_copy (EQ, <MODE>mode, operands);")
 
-(define_insn "*<avx512>_eq<mode>3<mask_scalar_merge_name>_1"
-  [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k,k")
-       (unspec:<avx512fmaskmode>
-         [(match_operand:VI12_AVX512VL 1 "nonimm_or_0_operand" "%v,v")
-          (match_operand:VI12_AVX512VL 2 "nonimm_or_0_operand" "vm,C")
-          (const_int 0)]
-         UNSPEC_PCMP))]
-  "TARGET_AVX512BW && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
-  "@
-   vpcmpeq<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}
-   vptestnm<ssemodesuffix>\t{%1, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %1}"
-  [(set_attr "type" "ssecmp")
-   (set_attr "prefix_extra" "1")
-   (set_attr "prefix" "evex")
-   (set_attr "mode" "<sseinsnmode>")])
-
-(define_insn "*<avx512>_eq<mode>3<mask_scalar_merge_name>_1"
-  [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=k,k")
-       (unspec:<avx512fmaskmode>
-         [(match_operand:VI48_AVX512VL 1 "nonimm_or_0_operand" "%v,v")
-          (match_operand:VI48_AVX512VL 2 "nonimm_or_0_operand" "vm,C")
-          (const_int 0)]
-         UNSPEC_PCMP))]
-  "TARGET_AVX512F && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
-  "@
-   vpcmpeq<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}
-   vptestnm<ssemodesuffix>\t{%1, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %1}"
-  [(set_attr "type" "ssecmp")
-   (set_attr "prefix_extra" "1")
-   (set_attr "prefix" "evex")
-   (set_attr "mode" "<sseinsnmode>")])
-
 (define_insn "*sse4_1_eqv2di3"
   [(set (match_operand:V2DI 0 "register_operand" "=Yr,*x,x")
        (eq:V2DI
index 31a0afc..9a85753 100644 (file)
@@ -1,11 +1,11 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512bw -mavx512vl -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index da5c401..a6c7776 100644 (file)
@@ -1,11 +1,11 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512bw -mavx512vl -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index 120fe8f..6d8a442 100644 (file)
@@ -1,6 +1,6 @@
 /* { dg-do compile } */
 /* { dg-options "-O2 -mavx512bw -mavx512vl" } */
-/* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpub|vpcmpeqb)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpub\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
index f17c73b..c0d84de 100644 (file)
@@ -5,7 +5,7 @@
 /* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuw\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuw|vpcmpeqw)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index c02f2b1..3c567c6 100644 (file)
@@ -1,7 +1,7 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[0-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[0-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index 1a443d6..f3f35ec 100644 (file)
@@ -1,7 +1,7 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512f -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[0-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[0-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index 331f95c..be0a8fe 100644 (file)
@@ -1,6 +1,6 @@
 /* { dg-do compile } */
 /* { dg-options "-O2 -mavx512f" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
index 7c87035..a0bee11 100644 (file)
@@ -1,9 +1,9 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512vl -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpud\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpud|vpcmpeqd)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index 4d672a5..c925d32 100644 (file)
@@ -1,9 +1,9 @@
 /* { dg-do compile } */
 /* { dg-options "-mavx512vl -O2" } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
 
index a5491c9..f93fca2 100644 (file)
@@ -2,7 +2,7 @@
 /* { dg-options "-O2 -mavx512vl" } */
 /* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
-/* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
+/* { dg-final { scan-assembler-times "(?:vpcmpuq|vpcmpeqq)\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\](?:\n|\[ \\t\]+#)" 1 } } */
 /* { dg-final { scan-assembler-times "vpcmpuq\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n^k\]*%k\[0-7\]\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
 
 #include <immintrin.h>
diff --git a/gcc/testsuite/gcc.target/i386/pr103774.c b/gcc/testsuite/gcc.target/i386/pr103774.c
new file mode 100644 (file)
index 0000000..0dafb1b
--- /dev/null
@@ -0,0 +1,25 @@
+/* { dg-do compile } */
+/* { dg-options "-march=cannonlake -O2" } */
+/* { dg-final { scan-assembler-not {(?n)vmovdq} } } */
+/* There should be no load + vpcmp, just vpcmp with memory operand.  */
+
+#include<immintrin.h>
+__mmask16
+foo (__m256i* n, __m256i mch256)
+{
+  __m256i data1 = _mm256_loadu_si256(n);
+  __m256i data2 = _mm256_loadu_si256(n+1);
+  __mmask16 mask1 = _mm256_cmpeq_epu16_mask(data1, mch256);
+  __mmask16 mask2 = _mm256_cmpeq_epu16_mask(data2, mch256);
+  return mask1 + mask2;
+}
+
+__mmask16
+foo1 (__m256i* n, __m256i mch256)
+{
+  __m256i data1 = _mm256_loadu_si256(n);
+  __m256i data2 = _mm256_loadu_si256(n+1);
+  __mmask16 mask1 = _mm256_cmpeq_epu16_mask(mch256, data1);
+  __mmask16 mask2 = _mm256_cmpeq_epu16_mask(mch256, data2);
+  return mask1 + mask2;
+}