* [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
@ 2019-02-28 19:22 H.J. Lu
2019-03-03 17:27 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: H.J. Lu @ 2019-02-28 19:22 UTC (permalink / raw)
To: gcc-patches; +Cc: Uros Bizjak
32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
when 32-bit indices are used as addresses, like in
vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
for x32 if there is no base register nor symbol.
This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
-Ofast -funroll-loops -march=haswell
gcc/
PR target/89523
* config/i386/i386.c (ix86_print_operand): Also handle '_' to
add addr32 prefix if required.
(ix86_print_operand_punct_valid_p): Allow '_'.
* config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
"%_".
(*avx512pf_gatherpf<mode>df_mask): Likewise.
(*avx512pf_scatterpf<mode>sf_mask): Likewise.
(*avx512pf_scatterpf<mode>df_mask): Likewise.
(*avx2_gathersi<mode>): Likewise.
(*avx2_gathersi<mode>_2): Likewise.
(*avx2_gatherdi<mode>): Likewise.
(*avx2_gatherdi<mode>_2): Likewise.
(*avx2_gatherdi<mode>_3): Likewise.
(*avx2_gatherdi<mode>_4): Likewise.
(*avx512f_gathersi<mode>): Likewise.
(*avx512f_gathersi<mode>_2): Likewise.
(*avx512f_gatherdi<mode>): Likewise.
(*avx512f_gatherdi<mode>_2): Likewise.
(*avx512f_scattersi<mode>): Likewise.
(*avx512f_scatterdi<mode>): Likewise.
gcc/testsuite/
PR target/89523
* gcc.target/i386/pr89523-1.c: New test.
* gcc.target/i386/pr89523-2.c: Likewise.
* gcc.target/i386/pr89523-3.c: Likewise.
* gcc.target/i386/pr89523-4.c: Likewise.
* gcc.target/i386/pr89523-5.c: Likewise.
* gcc.target/i386/pr89523-6.c: Likewise.
* gcc.target/i386/pr89523-7.c: Likewise.
* gcc.target/i386/pr89523-8.c: Likewise.
* gcc.target/i386/pr89523-9.c: Likewise.
xxx
---
gcc/config/i386/i386.c | 39 ++++++++++++++++++-
gcc/config/i386/sse.md | 46 +++++++++++------------
gcc/testsuite/gcc.target/i386/pr89523-1.c | 24 ++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-2.c | 17 +++++++++
gcc/testsuite/gcc.target/i386/pr89523-3.c | 17 +++++++++
gcc/testsuite/gcc.target/i386/pr89523-4.c | 16 ++++++++
gcc/testsuite/gcc.target/i386/pr89523-5.c | 18 +++++++++
gcc/testsuite/gcc.target/i386/pr89523-6.c | 17 +++++++++
gcc/testsuite/gcc.target/i386/pr89523-7.c | 19 ++++++++++
gcc/testsuite/gcc.target/i386/pr89523-8.c | 19 ++++++++++
gcc/testsuite/gcc.target/i386/pr89523-9.c | 16 ++++++++
11 files changed, 224 insertions(+), 24 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index b8357a7db5d..336696136de 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -17805,6 +17805,7 @@ print_reg (rtx x, int code, FILE *file)
~ -- print "i" if TARGET_AVX2, "f" otherwise.
^ -- print addr32 prefix if TARGET_64BIT and Pmode != word_mode
! -- print NOTRACK prefix for jxx/call/ret instructions if required.
+ _ -- print addr32 prefix if required.
*/
void
@@ -18356,6 +18357,42 @@ ix86_print_operand (FILE *file, rtx x, int code)
fputs ("addr32 ", file);
return;
+ case '_':
+ if (TARGET_X32)
+ {
+ subrtx_var_iterator::array_type array;
+ FOR_EACH_SUBRTX_VAR (iter, array,
+ PATTERN (current_output_insn), ALL)
+ {
+ rtx addr = *iter;
+ if (!MEM_P (addr))
+ continue;
+ addr = XEXP (addr, 0);
+ if (GET_CODE (addr) == UNSPEC
+ && XINT (addr, 1) == UNSPEC_VSIBADDR)
+ {
+ /* NB: 32-bit indices in VSIB address are
+ sign-extended to 64 bits. In x32, if 32-bit
+ address 0xf7fa3010 is sign-extended to
+ 0xfffffffff7fa3010 which is invalid address.
+ Add addr32 prefix if there is no base register
+ nor symbol. */
+ bool ok;
+ struct ix86_address parts;
+ ok = ix86_decompose_address (XVECEXP (addr, 0, 0),
+ &parts);
+ gcc_assert (ok && parts.index == NULL_RTX);
+ if (parts.base == NULL_RTX
+ && (parts.disp == NULL_RTX
+ || !symbolic_operand (parts.disp,
+ GET_MODE (parts.disp))))
+ fputs ("addr32 ", file);
+ break;
+ }
+ }
+ }
+ return;
+
case '!':
if (ix86_notrack_prefixed_insn_p (current_output_insn))
fputs ("notrack ", file);
@@ -18507,7 +18544,7 @@ static bool
ix86_print_operand_punct_valid_p (unsigned char code)
{
return (code == '*' || code == '+' || code == '&' || code == ';'
- || code == '~' || code == '^' || code == '!');
+ || code == '~' || code == '^' || code == '!' || code == '_');
}
\f
/* Print a memory operand whose address is ADDR. */
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index ac299495b2c..13692e47123 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -17401,9 +17401,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17448,9 +17448,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17496,10 +17496,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17545,10 +17545,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%_vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -20292,7 +20292,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
+ "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20312,7 +20312,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
+ "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20353,7 +20353,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
+ "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20375,8 +20375,8 @@
"TARGET_AVX2"
{
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
+ return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
+ return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
@@ -20400,7 +20400,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
+ "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20423,7 +20423,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
+ "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20463,7 +20463,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
+ "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20484,7 +20484,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
+ "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20525,7 +20525,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
+ "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20550,11 +20550,11 @@
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
{
if (<MODE_SIZE> != 64)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
+ return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
else
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
+ return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
}
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
+ return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
@@ -20593,7 +20593,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%_v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20631,7 +20631,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%_v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1.c b/gcc/testsuite/gcc.target/i386/pr89523-1.c
new file mode 100644
index 00000000000..f7ed24d1592
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-1.c
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -Ofast -funroll-loops -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+void foo (void);
+
+extern float *ncost;
+
+float
+bar (int type, int num)
+{
+ int i;
+ float cost;
+
+ cost = 0;
+ for (i = 0; i < num; i++)
+ if (type)
+ cost += ncost[i];
+ else
+ foo ();
+ return (cost);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
new file mode 100644
index 00000000000..7423f579b5e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+#include <immintrin.h>
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
new file mode 100644
index 00000000000..606f9aac659
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+#include <immintrin.h>
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i64gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
new file mode 100644
index 00000000000..155b818191f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
@@ -0,0 +1,16 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+#include <immintrin.h>
+
+__m128d x;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd (NULL, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
new file mode 100644
index 00000000000..11210ff8f78
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
@@ -0,0 +1,18 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+#include <immintrin.h>
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+void *base;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, base, 8, _MM_HINT_T0);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, _MM_HINT_T0);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
new file mode 100644
index 00000000000..0254ad435e3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+#include <immintrin.h>
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, NULL, 8, _MM_HINT_T0);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, NULL, 8, _MM_HINT_T0);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
new file mode 100644
index 00000000000..1c357bc8505
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
@@ -0,0 +1,19 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+#include <immintrin.h>
+
+volatile __m512d src;
+volatile __m256i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
new file mode 100644
index 00000000000..37b0a0bebb7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
@@ -0,0 +1,19 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+#include <immintrin.h>
+
+volatile __m512d src;
+volatile __m512i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i64scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
new file mode 100644
index 00000000000..a878f1e9efb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
@@ -0,0 +1,16 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler-not "\tvscatter" } } */
+/* { dg-final { scan-assembler "addr32 vscatter" } } */
+
+#include <immintrin.h>
+
+volatile __m512d src;
+volatile __m256i idx;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd (NULL, idx, src, 8);
+}
--
2.20.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-02-28 19:22 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions H.J. Lu
@ 2019-03-03 17:27 ` Uros Bizjak
2019-03-03 21:18 ` H.J. Lu
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2019-03-03 17:27 UTC (permalink / raw)
To: H.J. Lu; +Cc: gcc-patches
On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> when 32-bit indices are used as addresses, like in
>
> vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
>
> 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> for x32 if there is no base register nor symbol.
>
> This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
>
> -Ofast -funroll-loops -march=haswell
1. Testcases 2 to 9 fail on fedora-29 with:
In file included from /usr/include/features.h:452,
from /usr/include/bits/libc-header-start.h:33,
from /usr/include/stdlib.h:25,
from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
from
/home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
/usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
file or directory
2. Does the patch work with -maddress-mode={short,long}?
3. The implementation is wrong. You should use operand substitution
with VSIB address as operand, not substitution without operand.
4. The PR is not a regression.
Uros.
>
> gcc/
>
> PR target/89523
> * config/i386/i386.c (ix86_print_operand): Also handle '_' to
> add addr32 prefix if required.
> (ix86_print_operand_punct_valid_p): Allow '_'.
> * config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
> "%_".
> (*avx512pf_gatherpf<mode>df_mask): Likewise.
> (*avx512pf_scatterpf<mode>sf_mask): Likewise.
> (*avx512pf_scatterpf<mode>df_mask): Likewise.
> (*avx2_gathersi<mode>): Likewise.
> (*avx2_gathersi<mode>_2): Likewise.
> (*avx2_gatherdi<mode>): Likewise.
> (*avx2_gatherdi<mode>_2): Likewise.
> (*avx2_gatherdi<mode>_3): Likewise.
> (*avx2_gatherdi<mode>_4): Likewise.
> (*avx512f_gathersi<mode>): Likewise.
> (*avx512f_gathersi<mode>_2): Likewise.
> (*avx512f_gatherdi<mode>): Likewise.
> (*avx512f_gatherdi<mode>_2): Likewise.
> (*avx512f_scattersi<mode>): Likewise.
> (*avx512f_scatterdi<mode>): Likewise.
>
> gcc/testsuite/
>
> PR target/89523
> * gcc.target/i386/pr89523-1.c: New test.
> * gcc.target/i386/pr89523-2.c: Likewise.
> * gcc.target/i386/pr89523-3.c: Likewise.
> * gcc.target/i386/pr89523-4.c: Likewise.
> * gcc.target/i386/pr89523-5.c: Likewise.
> * gcc.target/i386/pr89523-6.c: Likewise.
> * gcc.target/i386/pr89523-7.c: Likewise.
> * gcc.target/i386/pr89523-8.c: Likewise.
> * gcc.target/i386/pr89523-9.c: Likewise.
>
> xxx
> ---
> gcc/config/i386/i386.c | 39 ++++++++++++++++++-
> gcc/config/i386/sse.md | 46 +++++++++++------------
> gcc/testsuite/gcc.target/i386/pr89523-1.c | 24 ++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-2.c | 17 +++++++++
> gcc/testsuite/gcc.target/i386/pr89523-3.c | 17 +++++++++
> gcc/testsuite/gcc.target/i386/pr89523-4.c | 16 ++++++++
> gcc/testsuite/gcc.target/i386/pr89523-5.c | 18 +++++++++
> gcc/testsuite/gcc.target/i386/pr89523-6.c | 17 +++++++++
> gcc/testsuite/gcc.target/i386/pr89523-7.c | 19 ++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-8.c | 19 ++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-9.c | 16 ++++++++
> 11 files changed, 224 insertions(+), 24 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index b8357a7db5d..336696136de 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -17805,6 +17805,7 @@ print_reg (rtx x, int code, FILE *file)
> ~ -- print "i" if TARGET_AVX2, "f" otherwise.
> ^ -- print addr32 prefix if TARGET_64BIT and Pmode != word_mode
> ! -- print NOTRACK prefix for jxx/call/ret instructions if required.
> + _ -- print addr32 prefix if required.
> */
>
> void
> @@ -18356,6 +18357,42 @@ ix86_print_operand (FILE *file, rtx x, int code)
> fputs ("addr32 ", file);
> return;
>
> + case '_':
> + if (TARGET_X32)
> + {
> + subrtx_var_iterator::array_type array;
> + FOR_EACH_SUBRTX_VAR (iter, array,
> + PATTERN (current_output_insn), ALL)
> + {
> + rtx addr = *iter;
> + if (!MEM_P (addr))
> + continue;
> + addr = XEXP (addr, 0);
> + if (GET_CODE (addr) == UNSPEC
> + && XINT (addr, 1) == UNSPEC_VSIBADDR)
> + {
> + /* NB: 32-bit indices in VSIB address are
> + sign-extended to 64 bits. In x32, if 32-bit
> + address 0xf7fa3010 is sign-extended to
> + 0xfffffffff7fa3010 which is invalid address.
> + Add addr32 prefix if there is no base register
> + nor symbol. */
> + bool ok;
> + struct ix86_address parts;
> + ok = ix86_decompose_address (XVECEXP (addr, 0, 0),
> + &parts);
> + gcc_assert (ok && parts.index == NULL_RTX);
> + if (parts.base == NULL_RTX
> + && (parts.disp == NULL_RTX
> + || !symbolic_operand (parts.disp,
> + GET_MODE (parts.disp))))
> + fputs ("addr32 ", file);
> + break;
> + }
> + }
> + }
> + return;
> +
> case '!':
> if (ix86_notrack_prefixed_insn_p (current_output_insn))
> fputs ("notrack ", file);
> @@ -18507,7 +18544,7 @@ static bool
> ix86_print_operand_punct_valid_p (unsigned char code)
> {
> return (code == '*' || code == '+' || code == '&' || code == ';'
> - || code == '~' || code == '^' || code == '!');
> + || code == '~' || code == '^' || code == '!' || code == '_');
> }
>
> /* Print a memory operand whose address is ADDR. */
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index ac299495b2c..13692e47123 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -17401,9 +17401,9 @@
> case 3:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> - return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17448,9 +17448,9 @@
> case 3:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> - return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17496,10 +17496,10 @@
> case 7:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> case 6:
> - return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17545,10 +17545,10 @@
> case 7:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> case 6:
> - return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%_vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -20292,7 +20292,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20312,7 +20312,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20353,7 +20353,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20375,8 +20375,8 @@
> "TARGET_AVX2"
> {
> if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> }
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> @@ -20400,7 +20400,7 @@
> (const_int 2) (const_int 3)])))
> (clobber (match_scratch:VI4F_256 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20423,7 +20423,7 @@
> (const_int 2) (const_int 3)])))
> (clobber (match_scratch:VI4F_256 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20463,7 +20463,7 @@
> "TARGET_AVX512F"
> ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20484,7 +20484,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20525,7 +20525,7 @@
> "TARGET_AVX512F"
> ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20550,11 +20550,11 @@
> if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> {
> if (<MODE_SIZE> != 64)
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> else
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> }
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> }
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> @@ -20593,7 +20593,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> + "%_v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20631,7 +20631,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> + "%_v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1.c b/gcc/testsuite/gcc.target/i386/pr89523-1.c
> new file mode 100644
> index 00000000000..f7ed24d1592
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-1.c
> @@ -0,0 +1,24 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -Ofast -funroll-loops -march=haswell" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +void foo (void);
> +
> +extern float *ncost;
> +
> +float
> +bar (int type, int num)
> +{
> + int i;
> + float cost;
> +
> + cost = 0;
> + for (i = 0; i < num; i++)
> + if (type)
> + cost += ncost[i];
> + else
> + foo ();
> + return (cost);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> new file mode 100644
> index 00000000000..7423f579b5e
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +#include <immintrin.h>
> +
> +__m128d x;
> +double *base;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i32gather_pd (base, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> new file mode 100644
> index 00000000000..606f9aac659
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +#include <immintrin.h>
> +
> +__m128d x;
> +double *base;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i64gather_pd (base, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> new file mode 100644
> index 00000000000..155b818191f
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> @@ -0,0 +1,16 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +#include <immintrin.h>
> +
> +__m128d x;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i32gather_pd (NULL, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> new file mode 100644
> index 00000000000..11210ff8f78
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> @@ -0,0 +1,18 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +#include <immintrin.h>
> +
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +void *base;
> +
> +void extern
> +avx512pf_test (void)
> +{
> + _mm512_prefetch_i32gather_pd (idx, base, 8, _MM_HINT_T0);
> + _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, _MM_HINT_T0);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> new file mode 100644
> index 00000000000..0254ad435e3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +#include <immintrin.h>
> +
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +
> +void extern
> +avx512pf_test (void)
> +{
> + _mm512_prefetch_i32gather_pd (idx, NULL, 8, _MM_HINT_T0);
> + _mm512_mask_prefetch_i32gather_pd (idx, m8, NULL, 8, _MM_HINT_T0);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> new file mode 100644
> index 00000000000..1c357bc8505
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> @@ -0,0 +1,19 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler "\tvscatter" } } */
> +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> +
> +#include <immintrin.h>
> +
> +volatile __m512d src;
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +double *addr;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i32scatter_pd (addr, idx, src, 8);
> + _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> new file mode 100644
> index 00000000000..37b0a0bebb7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> @@ -0,0 +1,19 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler "\tvscatter" } } */
> +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> +
> +#include <immintrin.h>
> +
> +volatile __m512d src;
> +volatile __m512i idx;
> +volatile __mmask8 m8;
> +double *addr;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i64scatter_pd (addr, idx, src, 8);
> + _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> new file mode 100644
> index 00000000000..a878f1e9efb
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> @@ -0,0 +1,16 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler-not "\tvscatter" } } */
> +/* { dg-final { scan-assembler "addr32 vscatter" } } */
> +
> +#include <immintrin.h>
> +
> +volatile __m512d src;
> +volatile __m256i idx;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i32scatter_pd (NULL, idx, src, 8);
> +}
> --
> 2.20.1
>
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-03 17:27 ` Uros Bizjak
@ 2019-03-03 21:18 ` H.J. Lu
2019-03-03 21:34 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: H.J. Lu @ 2019-03-03 21:18 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
On Sun, Mar 3, 2019 at 9:27 AM Uros Bizjak <ubizjak@gmail.com> wrote:
>
> On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > when 32-bit indices are used as addresses, like in
> >
> > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> >
> > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > for x32 if there is no base register nor symbol.
> >
> > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> >
> > -Ofast -funroll-loops -march=haswell
>
> 1. Testcases 2 to 9 fail on fedora-29 with:
>
> In file included from /usr/include/features.h:452,
> from /usr/include/bits/libc-header-start.h:33,
> from /usr/include/stdlib.h:25,
> from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
> from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
> from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
> from
> /home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
> /usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
> file or directory
I will update tests to remove "#include immintrin.h"
> 2. Does the patch work with -maddress-mode={short,long}?
Yes.
> 3. The implementation is wrong. You should use operand substitution
> with VSIB address as operand, not substitution without operand.
How can I add an addr32 prefix with operand substitution? This is
very similar to "%^". My updated patch will use "%^".
> 4. The PR is not a regression.
Correct.
H.J.
> Uros.
>
> >
> > gcc/
> >
> > PR target/89523
> > * config/i386/i386.c (ix86_print_operand): Also handle '_' to
> > add addr32 prefix if required.
> > (ix86_print_operand_punct_valid_p): Allow '_'.
> > * config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
> > "%_".
> > (*avx512pf_gatherpf<mode>df_mask): Likewise.
> > (*avx512pf_scatterpf<mode>sf_mask): Likewise.
> > (*avx512pf_scatterpf<mode>df_mask): Likewise.
> > (*avx2_gathersi<mode>): Likewise.
> > (*avx2_gathersi<mode>_2): Likewise.
> > (*avx2_gatherdi<mode>): Likewise.
> > (*avx2_gatherdi<mode>_2): Likewise.
> > (*avx2_gatherdi<mode>_3): Likewise.
> > (*avx2_gatherdi<mode>_4): Likewise.
> > (*avx512f_gathersi<mode>): Likewise.
> > (*avx512f_gathersi<mode>_2): Likewise.
> > (*avx512f_gatherdi<mode>): Likewise.
> > (*avx512f_gatherdi<mode>_2): Likewise.
> > (*avx512f_scattersi<mode>): Likewise.
> > (*avx512f_scatterdi<mode>): Likewise.
> >
> > gcc/testsuite/
> >
> > PR target/89523
> > * gcc.target/i386/pr89523-1.c: New test.
> > * gcc.target/i386/pr89523-2.c: Likewise.
> > * gcc.target/i386/pr89523-3.c: Likewise.
> > * gcc.target/i386/pr89523-4.c: Likewise.
> > * gcc.target/i386/pr89523-5.c: Likewise.
> > * gcc.target/i386/pr89523-6.c: Likewise.
> > * gcc.target/i386/pr89523-7.c: Likewise.
> > * gcc.target/i386/pr89523-8.c: Likewise.
> > * gcc.target/i386/pr89523-9.c: Likewise.
> >
> > xxx
> > ---
> > gcc/config/i386/i386.c | 39 ++++++++++++++++++-
> > gcc/config/i386/sse.md | 46 +++++++++++------------
> > gcc/testsuite/gcc.target/i386/pr89523-1.c | 24 ++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-2.c | 17 +++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-3.c | 17 +++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-4.c | 16 ++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-5.c | 18 +++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-6.c | 17 +++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-7.c | 19 ++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-8.c | 19 ++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-9.c | 16 ++++++++
> > 11 files changed, 224 insertions(+), 24 deletions(-)
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
> >
> > diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> > index b8357a7db5d..336696136de 100644
> > --- a/gcc/config/i386/i386.c
> > +++ b/gcc/config/i386/i386.c
> > @@ -17805,6 +17805,7 @@ print_reg (rtx x, int code, FILE *file)
> > ~ -- print "i" if TARGET_AVX2, "f" otherwise.
> > ^ -- print addr32 prefix if TARGET_64BIT and Pmode != word_mode
> > ! -- print NOTRACK prefix for jxx/call/ret instructions if required.
> > + _ -- print addr32 prefix if required.
> > */
> >
> > void
> > @@ -18356,6 +18357,42 @@ ix86_print_operand (FILE *file, rtx x, int code)
> > fputs ("addr32 ", file);
> > return;
> >
> > + case '_':
> > + if (TARGET_X32)
> > + {
> > + subrtx_var_iterator::array_type array;
> > + FOR_EACH_SUBRTX_VAR (iter, array,
> > + PATTERN (current_output_insn), ALL)
> > + {
> > + rtx addr = *iter;
> > + if (!MEM_P (addr))
> > + continue;
> > + addr = XEXP (addr, 0);
> > + if (GET_CODE (addr) == UNSPEC
> > + && XINT (addr, 1) == UNSPEC_VSIBADDR)
> > + {
> > + /* NB: 32-bit indices in VSIB address are
> > + sign-extended to 64 bits. In x32, if 32-bit
> > + address 0xf7fa3010 is sign-extended to
> > + 0xfffffffff7fa3010 which is invalid address.
> > + Add addr32 prefix if there is no base register
> > + nor symbol. */
> > + bool ok;
> > + struct ix86_address parts;
> > + ok = ix86_decompose_address (XVECEXP (addr, 0, 0),
> > + &parts);
> > + gcc_assert (ok && parts.index == NULL_RTX);
> > + if (parts.base == NULL_RTX
> > + && (parts.disp == NULL_RTX
> > + || !symbolic_operand (parts.disp,
> > + GET_MODE (parts.disp))))
> > + fputs ("addr32 ", file);
> > + break;
> > + }
> > + }
> > + }
> > + return;
> > +
> > case '!':
> > if (ix86_notrack_prefixed_insn_p (current_output_insn))
> > fputs ("notrack ", file);
> > @@ -18507,7 +18544,7 @@ static bool
> > ix86_print_operand_punct_valid_p (unsigned char code)
> > {
> > return (code == '*' || code == '+' || code == '&' || code == ';'
> > - || code == '~' || code == '^' || code == '!');
> > + || code == '~' || code == '^' || code == '!' || code == '_');
> > }
> >
> > /* Print a memory operand whose address is ADDR. */
> > diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> > index ac299495b2c..13692e47123 100644
> > --- a/gcc/config/i386/sse.md
> > +++ b/gcc/config/i386/sse.md
> > @@ -17401,9 +17401,9 @@
> > case 3:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > - return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17448,9 +17448,9 @@
> > case 3:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > - return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17496,10 +17496,10 @@
> > case 7:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > case 6:
> > - return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17545,10 +17545,10 @@
> > case 7:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > case 6:
> > - return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%_vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -20292,7 +20292,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> > + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20312,7 +20312,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> > + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20353,7 +20353,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> > + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20375,8 +20375,8 @@
> > "TARGET_AVX2"
> > {
> > if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> > + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> > + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> > }
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > @@ -20400,7 +20400,7 @@
> > (const_int 2) (const_int 3)])))
> > (clobber (match_scratch:VI4F_256 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> > + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20423,7 +20423,7 @@
> > (const_int 2) (const_int 3)])))
> > (clobber (match_scratch:VI4F_256 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> > + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20463,7 +20463,7 @@
> > "TARGET_AVX512F"
> > ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> > + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20484,7 +20484,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> > + "%_v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20525,7 +20525,7 @@
> > "TARGET_AVX512F"
> > ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> > + "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20550,11 +20550,11 @@
> > if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> > {
> > if (<MODE_SIZE> != 64)
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> > + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> > else
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> > + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> > }
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> > + return "%_v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> > }
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > @@ -20593,7 +20593,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > + "%_v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20631,7 +20631,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > + "%_v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1.c b/gcc/testsuite/gcc.target/i386/pr89523-1.c
> > new file mode 100644
> > index 00000000000..f7ed24d1592
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-1.c
> > @@ -0,0 +1,24 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -Ofast -funroll-loops -march=haswell" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +void foo (void);
> > +
> > +extern float *ncost;
> > +
> > +float
> > +bar (int type, int num)
> > +{
> > + int i;
> > + float cost;
> > +
> > + cost = 0;
> > + for (i = 0; i < num; i++)
> > + if (type)
> > + cost += ncost[i];
> > + else
> > + foo ();
> > + return (cost);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> > new file mode 100644
> > index 00000000000..7423f579b5e
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> > @@ -0,0 +1,17 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +__m128d x;
> > +double *base;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i32gather_pd (base, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> > new file mode 100644
> > index 00000000000..606f9aac659
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> > @@ -0,0 +1,17 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +__m128d x;
> > +double *base;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i64gather_pd (base, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> > new file mode 100644
> > index 00000000000..155b818191f
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> > @@ -0,0 +1,16 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +__m128d x;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i32gather_pd (NULL, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> > new file mode 100644
> > index 00000000000..11210ff8f78
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> > @@ -0,0 +1,18 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +void *base;
> > +
> > +void extern
> > +avx512pf_test (void)
> > +{
> > + _mm512_prefetch_i32gather_pd (idx, base, 8, _MM_HINT_T0);
> > + _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, _MM_HINT_T0);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> > new file mode 100644
> > index 00000000000..0254ad435e3
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> > @@ -0,0 +1,17 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +
> > +void extern
> > +avx512pf_test (void)
> > +{
> > + _mm512_prefetch_i32gather_pd (idx, NULL, 8, _MM_HINT_T0);
> > + _mm512_mask_prefetch_i32gather_pd (idx, m8, NULL, 8, _MM_HINT_T0);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> > new file mode 100644
> > index 00000000000..1c357bc8505
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> > @@ -0,0 +1,19 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler "\tvscatter" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +volatile __m512d src;
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +double *addr;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i32scatter_pd (addr, idx, src, 8);
> > + _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> > new file mode 100644
> > index 00000000000..37b0a0bebb7
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> > @@ -0,0 +1,19 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler "\tvscatter" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +volatile __m512d src;
> > +volatile __m512i idx;
> > +volatile __mmask8 m8;
> > +double *addr;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i64scatter_pd (addr, idx, src, 8);
> > + _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> > new file mode 100644
> > index 00000000000..a878f1e9efb
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> > @@ -0,0 +1,16 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler-not "\tvscatter" } } */
> > +/* { dg-final { scan-assembler "addr32 vscatter" } } */
> > +
> > +#include <immintrin.h>
> > +
> > +volatile __m512d src;
> > +volatile __m256i idx;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i32scatter_pd (NULL, idx, src, 8);
> > +}
> > --
> > 2.20.1
> >
--
H.J.
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-03 21:18 ` H.J. Lu
@ 2019-03-03 21:34 ` Uros Bizjak
2019-03-04 13:54 ` V2 " H.J. Lu
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2019-03-03 21:34 UTC (permalink / raw)
To: H.J. Lu; +Cc: gcc-patches
On Sun, Mar 3, 2019 at 10:18 PM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Sun, Mar 3, 2019 at 9:27 AM Uros Bizjak <ubizjak@gmail.com> wrote:
> >
> > On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > >
> > > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > > when 32-bit indices are used as addresses, like in
> > >
> > > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> > >
> > > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > > for x32 if there is no base register nor symbol.
> > >
> > > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> > >
> > > -Ofast -funroll-loops -march=haswell
> >
> > 1. Testcases 2 to 9 fail on fedora-29 with:
> >
> > In file included from /usr/include/features.h:452,
> > from /usr/include/bits/libc-header-start.h:33,
> > from /usr/include/stdlib.h:25,
> > from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
> > from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
> > from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
> > from
> > /home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
> > /usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
> > file or directory
>
> I will update tests to remove "#include immintrin.h"
>
> > 2. Does the patch work with -maddress-mode={short,long}?
>
> Yes.
>
> > 3. The implementation is wrong. You should use operand substitution
> > with VSIB address as operand, not substitution without operand.
>
> How can I add an addr32 prefix with operand substitution? This is
> very similar to "%^". My updated patch will use "%^".
Yes, using %^ is what I think would be the optimal solution. Other
than that, in your proposed patch, operand-less %_ scans the entire
current_output_insn to dig to the UNSPEC_VSIBADDR. You can just use
operand substitution, and do e.g. "%X2vgatherpf0..." where 'X'
processes operand 2 (vsib_address_operand) and conditionally outputs
addr32.
BTW: In a new version of the patch, please specify what is changed
from the previous version. Otherwise, review of a new version is more
or less a guesswork what changed.
Uros.
^ permalink raw reply [flat|nested] 8+ messages in thread
* V2 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-03 21:34 ` Uros Bizjak
@ 2019-03-04 13:54 ` H.J. Lu
2019-03-04 14:09 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: H.J. Lu @ 2019-03-04 13:54 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
On Sun, Mar 03, 2019 at 10:34:29PM +0100, Uros Bizjak wrote:
> On Sun, Mar 3, 2019 at 10:18 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Sun, Mar 3, 2019 at 9:27 AM Uros Bizjak <ubizjak@gmail.com> wrote:
> > >
> > > On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > > >
> > > > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > > > when 32-bit indices are used as addresses, like in
> > > >
> > > > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> > > >
> > > > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > > > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > > > for x32 if there is no base register nor symbol.
> > > >
> > > > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> > > >
> > > > -Ofast -funroll-loops -march=haswell
> > >
> > > 1. Testcases 2 to 9 fail on fedora-29 with:
> > >
> > > In file included from /usr/include/features.h:452,
> > > from /usr/include/bits/libc-header-start.h:33,
> > > from /usr/include/stdlib.h:25,
> > > from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
> > > from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
> > > from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
> > > from
> > > /home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
> > > /usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
> > > file or directory
> >
> > I will update tests to remove "#include immintrin.h"
> >
> > > 2. Does the patch work with -maddress-mode={short,long}?
> >
> > Yes.
> >
> > > 3. The implementation is wrong. You should use operand substitution
> > > with VSIB address as operand, not substitution without operand.
> >
> > How can I add an addr32 prefix with operand substitution? This is
> > very similar to "%^". My updated patch will use "%^".
>
> Yes, using %^ is what I think would be the optimal solution. Other
> than that, in your proposed patch, operand-less %_ scans the entire
> current_output_insn to dig to the UNSPEC_VSIBADDR. You can just use
> operand substitution, and do e.g. "%X2vgatherpf0..." where 'X'
> processes operand 2 (vsib_address_operand) and conditionally outputs
> addr32.
>
> BTW: In a new version of the patch, please specify what is changed
> from the previous version. Otherwise, review of a new version is more
> or less a guesswork what changed.
>
Here is the updated patch. The change is
return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
instead of
return "%^vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
We can't use the %X5 since %X5 is used on operands.
I also added a test for -maddress-mode=long.
H.J.
---
32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
when 32-bit indices are used as addresses, like in
vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
for x32 if there is no base register nor symbol.
This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
-Ofast -funroll-loops -march=haswell
gcc/
PR target/89523
* config/i386/i386.c (ix86_print_operand): Handle UNSPEC_VSIBADDR
instructions for '%P' to add addr32 prefix if required.
* config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
"%P5" to opcode.
(*avx512pf_gatherpf<mode>df_mask): Likewise.
(*avx512pf_scatterpf<mode>sf_mask): Likewise.
(*avx512pf_scatterpf<mode>df_mask): Likewise.
(*avx2_gathersi<mode>): Prepend "%P7" to opcode.
(*avx2_gathersi<mode>_2): Prepend "%P6" to opcode.
(*avx2_gatherdi<mode>): Prepend "%P7" to opcode.
(*avx2_gatherdi<mode>_2): Prepend "%P6" to opcode.
(*avx2_gatherdi<mode>_3): Prepend "%P7" to opcode.
(*avx2_gatherdi<mode>_4): Prepend "%P6" to opcode.`
(*avx512f_gathersi<mode>): Prepend "%P5" to opcode.
(*avx512f_gathersi<mode>_2): Prepend "%P6" to opcode.
(*avx512f_gatherdi<mode>): Prepend "%P5" to opcode.
(*avx512f_gatherdi<mode>_2): Likewise.
(*avx512f_scattersi<mode>): Likewise.
(*avx512f_scatterdi<mode>): Likewise.
gcc/testsuite/
PR target/89523
* gcc.target/i386/pr89523-1a.c: New test.
* gcc.target/i386/pr89523-1b.c: Likewise.
* gcc.target/i386/pr89523-2.c: Likewise.
* gcc.target/i386/pr89523-3.c: Likewise.
* gcc.target/i386/pr89523-4.c: Likewise.
* gcc.target/i386/pr89523-5.c: Likewise.
* gcc.target/i386/pr89523-6.c: Likewise.
* gcc.target/i386/pr89523-7.c: Likewise.
* gcc.target/i386/pr89523-8.c: Likewise.
* gcc.target/i386/pr89523-9.c: Likewise.
---
gcc/config/i386/i386.c | 35 +++++++++++++++-
gcc/config/i386/sse.md | 46 +++++++++++-----------
gcc/testsuite/gcc.target/i386/pr89523-1a.c | 24 +++++++++++
gcc/testsuite/gcc.target/i386/pr89523-1b.c | 7 ++++
gcc/testsuite/gcc.target/i386/pr89523-2.c | 37 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-3.c | 36 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-4.c | 36 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-5.c | 39 ++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-6.c | 38 ++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-7.c | 42 ++++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-8.c | 41 +++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-9.c | 30 ++++++++++++++
12 files changed, 386 insertions(+), 25 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1a.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1b.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index c8f9957163b..ae9befb638d 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -17793,7 +17793,8 @@ print_reg (rtx x, int code, FILE *file)
y -- print "st(0)" instead of "st" as a register.
d -- print duplicated register operand for AVX instruction.
D -- print condition for SSE cmp instruction.
- P -- if PIC, print an @PLT suffix.
+ P -- if PIC, print an @PLT suffix or print addr32 prefix for
+ TARGET_X32 with UNSPEC_VSIBADDR operand.
p -- print raw symbol name.
X -- don't print any sort of PIC '@' suffix for a symbol.
& -- print some in-use local-dynamic symbol name.
@@ -18010,6 +18011,37 @@ ix86_print_operand (FILE *file, rtx x, int code)
output_operand_lossage ("invalid operand size for operand code 'Z'");
return;
+ case 'P':
+ if (MEM_P (x))
+ {
+ x = XEXP (x, 0);
+ if (GET_CODE (x) == UNSPEC
+ || XINT (x, 1) == UNSPEC_VSIBADDR)
+ {
+ if (TARGET_X32)
+ {
+ /* NB: 32-bit indices in VSIB address are
+ sign-extended to 64 bits. In x32, if 32-bit
+ address 0xf7fa3010 is sign-extended to
+ 0xfffffffff7fa3010 which is invalid address.
+ Add addr32 prefix if there is no base register
+ nor symbol. */
+ bool ok;
+ struct ix86_address parts;
+ ok = ix86_decompose_address (XVECEXP (x, 0, 0),
+ &parts);
+ gcc_assert (ok && parts.index == NULL_RTX);
+ if (parts.base == NULL_RTX
+ && (parts.disp == NULL_RTX
+ || !symbolic_operand (parts.disp,
+ GET_MODE (parts.disp))))
+ fputs ("addr32 ", file);
+ }
+ return;
+ }
+ }
+ break;
+
case 'd':
case 'b':
case 'w':
@@ -18021,7 +18053,6 @@ ix86_print_operand (FILE *file, rtx x, int code)
case 'y':
case 'x':
case 'X':
- case 'P':
case 'p':
case 'V':
break;
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index ac299495b2c..ac500f9cc63 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -17401,9 +17401,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17448,9 +17448,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17496,10 +17496,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17545,10 +17545,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%P5vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -20292,7 +20292,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
+ "%P7v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20312,7 +20312,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
+ "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20353,7 +20353,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
+ "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20375,8 +20375,8 @@
"TARGET_AVX2"
{
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
+ return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
+ return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
@@ -20400,7 +20400,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
+ "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20423,7 +20423,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
+ "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20463,7 +20463,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
+ "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20484,7 +20484,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
+ "%P5v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20525,7 +20525,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
+ "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20550,11 +20550,11 @@
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
{
if (<MODE_SIZE> != 64)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
+ return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
else
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
+ return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
}
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
+ return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
@@ -20593,7 +20593,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%P5v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20631,7 +20631,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%P5v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1a.c b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
new file mode 100644
index 00000000000..0d0edab0363
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-maddress-mode=short -mx32 -Ofast -funroll-loops -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+void foo (void);
+
+extern float *ncost;
+
+float
+bar (int type, int num)
+{
+ int i;
+ float cost;
+
+ cost = 0;
+ for (i = 0; i < num; i++)
+ if (type)
+ cost += ncost[i];
+ else
+ foo ();
+ return (cost);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1b.c b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
new file mode 100644
index 00000000000..6a5c1d43625
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
@@ -0,0 +1,7 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-maddress-mode=long -mx32 -Ofast -funroll-loops -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+#include "pr89523-1a.c"
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
new file mode 100644
index 00000000000..2ffbffe5e40
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
@@ -0,0 +1,37 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+ __v2df x = x;
+
+ return (__m128d) __builtin_ia32_gathersiv2df (x,
+ __base,
+ (__v4si)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
new file mode 100644
index 00000000000..fc3631b694b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
@@ -0,0 +1,36 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i64gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+
+ return (__m128d) __builtin_ia32_gatherdiv2df (__zero,
+ __base,
+ (__v2di)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i64gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
new file mode 100644
index 00000000000..3436e5dcae3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
@@ -0,0 +1,36 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+ __v2df x = x;
+
+ return (__m128d) __builtin_ia32_gathersiv2df (x,
+ __base,
+ (__v4si)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd ((void *) 0, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
new file mode 100644
index 00000000000..6a769c7a249
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
@@ -0,0 +1,39 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
+ int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
+ __scale, __hint);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
+ void const *__addr, int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
+ __hint);
+}
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+void *base;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, base, 8, 3);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, 3);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
new file mode 100644
index 00000000000..82f795e085c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
@@ -0,0 +1,38 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
+ int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
+ __scale, __hint);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
+ void const *__addr, int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
+ __hint);
+}
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, (void *) 0, 8, 3);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, (void *) 0, 8, 3);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
new file mode 100644
index 00000000000..030b00d268a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
@@ -0,0 +1,42 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
+ (__v8si) __index, (__v8df) __v1, __scale);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_i32scatter_pd (void *__addr, __mmask8 __mask,
+ __m256i __index, __m512d __v1, int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, __mask, (__v8si) __index,
+ (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m256i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
new file mode 100644
index 00000000000..465c985c2b7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
@@ -0,0 +1,41 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+typedef long long __v8di __attribute__ ((__vector_size__ (64)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i64scatter_pd (void *__addr, __m512i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scatterdiv8df (__addr, (__mmask8) 0xFF,
+ (__v8di) __index, (__v8df) __v1, __scale);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_i64scatter_pd (void *__addr, __mmask8 __mask,
+ __m512i __index, __m512d __v1, int __scale)
+{
+ __builtin_ia32_scatterdiv8df (__addr, __mask, (__v8di) __index,
+ (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m512i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i64scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
new file mode 100644
index 00000000000..e9323126bd6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
@@ -0,0 +1,30 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler-not "\tvscatter" } } */
+/* { dg-final { scan-assembler "addr32 vscatter" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
+ (__v8si) __index, (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m256i idx;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd ((void *) 0, idx, src, 8);
+}
--
2.20.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: V2 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-04 13:54 ` V2 " H.J. Lu
@ 2019-03-04 14:09 ` Uros Bizjak
2019-03-14 8:34 ` H.J. Lu
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2019-03-04 14:09 UTC (permalink / raw)
To: H.J. Lu; +Cc: gcc-patches
On Mon, Mar 4, 2019 at 2:54 PM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Sun, Mar 03, 2019 at 10:34:29PM +0100, Uros Bizjak wrote:
> > On Sun, Mar 3, 2019 at 10:18 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > >
> > > On Sun, Mar 3, 2019 at 9:27 AM Uros Bizjak <ubizjak@gmail.com> wrote:
> > > >
> > > > On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > > > >
> > > > > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > > > > when 32-bit indices are used as addresses, like in
> > > > >
> > > > > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> > > > >
> > > > > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > > > > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > > > > for x32 if there is no base register nor symbol.
> > > > >
> > > > > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> > > > >
> > > > > -Ofast -funroll-loops -march=haswell
> > > >
> > > > 1. Testcases 2 to 9 fail on fedora-29 with:
> > > >
> > > > In file included from /usr/include/features.h:452,
> > > > from /usr/include/bits/libc-header-start.h:33,
> > > > from /usr/include/stdlib.h:25,
> > > > from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
> > > > from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
> > > > from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
> > > > from
> > > > /home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
> > > > /usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
> > > > file or directory
> > >
> > > I will update tests to remove "#include immintrin.h"
> > >
> > > > 2. Does the patch work with -maddress-mode={short,long}?
> > >
> > > Yes.
> > >
> > > > 3. The implementation is wrong. You should use operand substitution
> > > > with VSIB address as operand, not substitution without operand.
> > >
> > > How can I add an addr32 prefix with operand substitution? This is
> > > very similar to "%^". My updated patch will use "%^".
> >
> > Yes, using %^ is what I think would be the optimal solution. Other
> > than that, in your proposed patch, operand-less %_ scans the entire
> > current_output_insn to dig to the UNSPEC_VSIBADDR. You can just use
> > operand substitution, and do e.g. "%X2vgatherpf0..." where 'X'
> > processes operand 2 (vsib_address_operand) and conditionally outputs
> > addr32.
> >
> > BTW: In a new version of the patch, please specify what is changed
> > from the previous version. Otherwise, review of a new version is more
> > or less a guesswork what changed.
> >
>
> Here is the updated patch. The change is
>
> return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
>
> instead of
>
> return "%^vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
Did I miss some version of the patch that introduced %^? You used %_
in your previous patch. Did your try with %^?
> We can't use the %X5 since %X5 is used on operands.
So, please introduce some other modifier ("X" was not to be taken
literally, but *some* letter). Why are you overloading 'P'?
I don't know why are you using operand 5 here, you can use operand 2 directly.
Uros.
> I also added a test for -maddress-mode=long.
>
>
> H.J.
> ---
> 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> when 32-bit indices are used as addresses, like in
>
> vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
>
> 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> for x32 if there is no base register nor symbol.
>
> This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
>
> -Ofast -funroll-loops -march=haswell
>
> gcc/
>
> PR target/89523
> * config/i386/i386.c (ix86_print_operand): Handle UNSPEC_VSIBADDR
> instructions for '%P' to add addr32 prefix if required.
> * config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
> "%P5" to opcode.
> (*avx512pf_gatherpf<mode>df_mask): Likewise.
> (*avx512pf_scatterpf<mode>sf_mask): Likewise.
> (*avx512pf_scatterpf<mode>df_mask): Likewise.
> (*avx2_gathersi<mode>): Prepend "%P7" to opcode.
> (*avx2_gathersi<mode>_2): Prepend "%P6" to opcode.
> (*avx2_gatherdi<mode>): Prepend "%P7" to opcode.
> (*avx2_gatherdi<mode>_2): Prepend "%P6" to opcode.
> (*avx2_gatherdi<mode>_3): Prepend "%P7" to opcode.
> (*avx2_gatherdi<mode>_4): Prepend "%P6" to opcode.`
> (*avx512f_gathersi<mode>): Prepend "%P5" to opcode.
> (*avx512f_gathersi<mode>_2): Prepend "%P6" to opcode.
> (*avx512f_gatherdi<mode>): Prepend "%P5" to opcode.
> (*avx512f_gatherdi<mode>_2): Likewise.
> (*avx512f_scattersi<mode>): Likewise.
> (*avx512f_scatterdi<mode>): Likewise.
>
> gcc/testsuite/
>
> PR target/89523
> * gcc.target/i386/pr89523-1a.c: New test.
> * gcc.target/i386/pr89523-1b.c: Likewise.
> * gcc.target/i386/pr89523-2.c: Likewise.
> * gcc.target/i386/pr89523-3.c: Likewise.
> * gcc.target/i386/pr89523-4.c: Likewise.
> * gcc.target/i386/pr89523-5.c: Likewise.
> * gcc.target/i386/pr89523-6.c: Likewise.
> * gcc.target/i386/pr89523-7.c: Likewise.
> * gcc.target/i386/pr89523-8.c: Likewise.
> * gcc.target/i386/pr89523-9.c: Likewise.
> ---
> gcc/config/i386/i386.c | 35 +++++++++++++++-
> gcc/config/i386/sse.md | 46 +++++++++++-----------
> gcc/testsuite/gcc.target/i386/pr89523-1a.c | 24 +++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-1b.c | 7 ++++
> gcc/testsuite/gcc.target/i386/pr89523-2.c | 37 +++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-3.c | 36 +++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-4.c | 36 +++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-5.c | 39 ++++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-6.c | 38 ++++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-7.c | 42 ++++++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-8.c | 41 +++++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr89523-9.c | 30 ++++++++++++++
> 12 files changed, 386 insertions(+), 25 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1a.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1b.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index c8f9957163b..ae9befb638d 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -17793,7 +17793,8 @@ print_reg (rtx x, int code, FILE *file)
> y -- print "st(0)" instead of "st" as a register.
> d -- print duplicated register operand for AVX instruction.
> D -- print condition for SSE cmp instruction.
> - P -- if PIC, print an @PLT suffix.
> + P -- if PIC, print an @PLT suffix or print addr32 prefix for
> + TARGET_X32 with UNSPEC_VSIBADDR operand.
> p -- print raw symbol name.
> X -- don't print any sort of PIC '@' suffix for a symbol.
> & -- print some in-use local-dynamic symbol name.
> @@ -18010,6 +18011,37 @@ ix86_print_operand (FILE *file, rtx x, int code)
> output_operand_lossage ("invalid operand size for operand code 'Z'");
> return;
>
> + case 'P':
> + if (MEM_P (x))
> + {
> + x = XEXP (x, 0);
> + if (GET_CODE (x) == UNSPEC
> + || XINT (x, 1) == UNSPEC_VSIBADDR)
> + {
> + if (TARGET_X32)
> + {
> + /* NB: 32-bit indices in VSIB address are
> + sign-extended to 64 bits. In x32, if 32-bit
> + address 0xf7fa3010 is sign-extended to
> + 0xfffffffff7fa3010 which is invalid address.
> + Add addr32 prefix if there is no base register
> + nor symbol. */
> + bool ok;
> + struct ix86_address parts;
> + ok = ix86_decompose_address (XVECEXP (x, 0, 0),
> + &parts);
> + gcc_assert (ok && parts.index == NULL_RTX);
> + if (parts.base == NULL_RTX
> + && (parts.disp == NULL_RTX
> + || !symbolic_operand (parts.disp,
> + GET_MODE (parts.disp))))
> + fputs ("addr32 ", file);
> + }
> + return;
> + }
> + }
> + break;
> +
> case 'd':
> case 'b':
> case 'w':
> @@ -18021,7 +18053,6 @@ ix86_print_operand (FILE *file, rtx x, int code)
> case 'y':
> case 'x':
> case 'X':
> - case 'P':
> case 'p':
> case 'V':
> break;
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index ac299495b2c..ac500f9cc63 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -17401,9 +17401,9 @@
> case 3:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> - return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17448,9 +17448,9 @@
> case 3:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> - return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17496,10 +17496,10 @@
> case 7:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> case 6:
> - return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -17545,10 +17545,10 @@
> case 7:
> /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> gas changed what it requires incompatibly. */
> - return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> case 2:
> case 6:
> - return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> + return "%P5vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> default:
> gcc_unreachable ();
> }
> @@ -20292,7 +20292,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> + "%P7v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20312,7 +20312,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> + "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20353,7 +20353,7 @@
> UNSPEC_GATHER))
> (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> + "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20375,8 +20375,8 @@
> "TARGET_AVX2"
> {
> if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> + return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> + return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> }
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> @@ -20400,7 +20400,7 @@
> (const_int 2) (const_int 3)])))
> (clobber (match_scratch:VI4F_256 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> + "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20423,7 +20423,7 @@
> (const_int 2) (const_int 3)])))
> (clobber (match_scratch:VI4F_256 1 "=&x"))]
> "TARGET_AVX2"
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> + "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "vex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20463,7 +20463,7 @@
> "TARGET_AVX512F"
> ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> + "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20484,7 +20484,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> + "%P5v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20525,7 +20525,7 @@
> "TARGET_AVX512F"
> ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> + "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20550,11 +20550,11 @@
> if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> {
> if (<MODE_SIZE> != 64)
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> else
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> }
> - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> }
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> @@ -20593,7 +20593,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> + "%P5v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> @@ -20631,7 +20631,7 @@
> "TARGET_AVX512F"
> ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> ;; gas changed what it requires incompatibly.
> - "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> + "%P5v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> [(set_attr "type" "ssemov")
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1a.c b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
> new file mode 100644
> index 00000000000..0d0edab0363
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
> @@ -0,0 +1,24 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-maddress-mode=short -mx32 -Ofast -funroll-loops -march=haswell" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +void foo (void);
> +
> +extern float *ncost;
> +
> +float
> +bar (int type, int num)
> +{
> + int i;
> + float cost;
> +
> + cost = 0;
> + for (i = 0; i < num; i++)
> + if (type)
> + cost += ncost[i];
> + else
> + foo ();
> + return (cost);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1b.c b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
> new file mode 100644
> index 00000000000..6a5c1d43625
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
> @@ -0,0 +1,7 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-maddress-mode=long -mx32 -Ofast -funroll-loops -march=haswell" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +#include "pr89523-1a.c"
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> new file mode 100644
> index 00000000000..2ffbffe5e40
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> @@ -0,0 +1,37 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> +
> +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> +
> +extern __inline __m128d
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
> +{
> + __v2df __zero = { 0.0, 0.0 };
> + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> + __v2df x = x;
> +
> + return (__m128d) __builtin_ia32_gathersiv2df (x,
> + __base,
> + (__v4si)__index,
> + __mask,
> + __scale);
> +}
> +
> +__m128d x;
> +double *base;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i32gather_pd (base, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> new file mode 100644
> index 00000000000..fc3631b694b
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> @@ -0,0 +1,36 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> +
> +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> +
> +extern __inline __m128d
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm_i64gather_pd (double const *__base, __m128i __index, const int __scale)
> +{
> + __v2df __zero = { 0.0, 0.0 };
> + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> +
> + return (__m128d) __builtin_ia32_gatherdiv2df (__zero,
> + __base,
> + (__v2di)__index,
> + __mask,
> + __scale);
> +}
> +
> +__m128d x;
> +double *base;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i64gather_pd (base, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> new file mode 100644
> index 00000000000..3436e5dcae3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> @@ -0,0 +1,36 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -march=haswell" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> +
> +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> +
> +extern __inline __m128d
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
> +{
> + __v2df __zero = { 0.0, 0.0 };
> + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> + __v2df x = x;
> +
> + return (__m128d) __builtin_ia32_gathersiv2df (x,
> + __base,
> + (__v4si)__index,
> + __mask,
> + __scale);
> +}
> +
> +__m128d x;
> +__m128i idx;
> +
> +void extern
> +avx2_test (void)
> +{
> + x = _mm_i32gather_pd ((void *) 0, idx, 1);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> new file mode 100644
> index 00000000000..6a769c7a249
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> @@ -0,0 +1,39 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> +/* { dg-final { scan-assembler "\tvgather" } } */
> +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> +
> +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> + __may_alias__));
> +typedef unsigned char __mmask8;
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
> + int __scale, int __hint)
> +{
> + __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
> + __scale, __hint);
> +}
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
> + void const *__addr, int __scale, int __hint)
> +{
> + __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
> + __hint);
> +}
> +
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +void *base;
> +
> +void extern
> +avx512pf_test (void)
> +{
> + _mm512_prefetch_i32gather_pd (idx, base, 8, 3);
> + _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, 3);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> new file mode 100644
> index 00000000000..82f795e085c
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> @@ -0,0 +1,38 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> +/* { dg-final { scan-assembler-not "\tvgather" } } */
> +/* { dg-final { scan-assembler "addr32 vgather" } } */
> +
> +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> + __may_alias__));
> +typedef unsigned char __mmask8;
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
> + int __scale, int __hint)
> +{
> + __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
> + __scale, __hint);
> +}
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
> + void const *__addr, int __scale, int __hint)
> +{
> + __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
> + __hint);
> +}
> +
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +
> +void extern
> +avx512pf_test (void)
> +{
> + _mm512_prefetch_i32gather_pd (idx, (void *) 0, 8, 3);
> + _mm512_mask_prefetch_i32gather_pd (idx, m8, (void *) 0, 8, 3);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> new file mode 100644
> index 00000000000..030b00d268a
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> @@ -0,0 +1,42 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler "\tvscatter" } } */
> +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> +
> +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> + __may_alias__));
> +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> +typedef unsigned char __mmask8;
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
> + int __scale)
> +{
> + __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
> + (__v8si) __index, (__v8df) __v1, __scale);
> +}
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_mask_i32scatter_pd (void *__addr, __mmask8 __mask,
> + __m256i __index, __m512d __v1, int __scale)
> +{
> + __builtin_ia32_scattersiv8df (__addr, __mask, (__v8si) __index,
> + (__v8df) __v1, __scale);
> +}
> +
> +volatile __m512d src;
> +volatile __m256i idx;
> +volatile __mmask8 m8;
> +double *addr;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i32scatter_pd (addr, idx, src, 8);
> + _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> new file mode 100644
> index 00000000000..465c985c2b7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> @@ -0,0 +1,41 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler "\tvscatter" } } */
> +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> +
> +typedef long long __v8di __attribute__ ((__vector_size__ (64)));
> +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> +typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
> +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> +typedef unsigned char __mmask8;
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_i64scatter_pd (void *__addr, __m512i __index, __m512d __v1,
> + int __scale)
> +{
> + __builtin_ia32_scatterdiv8df (__addr, (__mmask8) 0xFF,
> + (__v8di) __index, (__v8df) __v1, __scale);
> +}
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_mask_i64scatter_pd (void *__addr, __mmask8 __mask,
> + __m512i __index, __m512d __v1, int __scale)
> +{
> + __builtin_ia32_scatterdiv8df (__addr, __mask, (__v8di) __index,
> + (__v8df) __v1, __scale);
> +}
> +
> +volatile __m512d src;
> +volatile __m512i idx;
> +volatile __mmask8 m8;
> +double *addr;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i64scatter_pd (addr, idx, src, 8);
> + _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> new file mode 100644
> index 00000000000..e9323126bd6
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> @@ -0,0 +1,30 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-require-effective-target maybe_x32 } */
> +/* { dg-options "-mx32 -O2 -mavx512f" } */
> +/* { dg-final { scan-assembler-not "\tvscatter" } } */
> +/* { dg-final { scan-assembler "addr32 vscatter" } } */
> +
> +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> + __may_alias__));
> +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> +typedef unsigned char __mmask8;
> +
> +extern __inline void
> +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> +_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
> + int __scale)
> +{
> + __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
> + (__v8si) __index, (__v8df) __v1, __scale);
> +}
> +
> +volatile __m512d src;
> +volatile __m256i idx;
> +
> +void extern
> +avx512f_test (void)
> +{
> + _mm512_i32scatter_pd ((void *) 0, idx, src, 8);
> +}
> --
> 2.20.1
>
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: V2 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-04 14:09 ` Uros Bizjak
@ 2019-03-14 8:34 ` H.J. Lu
2019-03-14 8:47 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: H.J. Lu @ 2019-03-14 8:34 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 34702 bytes --]
On Mon, Mar 4, 2019 at 10:09 PM Uros Bizjak <ubizjak@gmail.com> wrote:
>
> On Mon, Mar 4, 2019 at 2:54 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Sun, Mar 03, 2019 at 10:34:29PM +0100, Uros Bizjak wrote:
> > > On Sun, Mar 3, 2019 at 10:18 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > > >
> > > > On Sun, Mar 3, 2019 at 9:27 AM Uros Bizjak <ubizjak@gmail.com> wrote:
> > > > >
> > > > > On Thu, Feb 28, 2019 at 8:10 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> > > > > >
> > > > > > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > > > > > when 32-bit indices are used as addresses, like in
> > > > > >
> > > > > > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> > > > > >
> > > > > > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > > > > > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > > > > > for x32 if there is no base register nor symbol.
> > > > > >
> > > > > > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> > > > > >
> > > > > > -Ofast -funroll-loops -march=haswell
> > > > >
> > > > > 1. Testcases 2 to 9 fail on fedora-29 with:
> > > > >
> > > > > In file included from /usr/include/features.h:452,
> > > > > from /usr/include/bits/libc-header-start.h:33,
> > > > > from /usr/include/stdlib.h:25,
> > > > > from /ssd/uros/gcc-build-fast/gcc/include/mm_malloc.h:27,
> > > > > from /ssd/uros/gcc-build-fast/gcc/include/xmmintrin.h:34,
> > > > > from /ssd/uros/gcc-build-fast/gcc/include/immintrin.h:29,
> > > > > from
> > > > > /home/uros/gcc-svn/trunk/gcc/testsuite/gcc.target/i386/pr89523-2.c:7:
> > > > > /usr/include/gnu/stubs.h:13:11: fatal error: gnu/stubs-x32.h: No such
> > > > > file or directory
> > > >
> > > > I will update tests to remove "#include immintrin.h"
> > > >
> > > > > 2. Does the patch work with -maddress-mode={short,long}?
> > > >
> > > > Yes.
> > > >
> > > > > 3. The implementation is wrong. You should use operand substitution
> > > > > with VSIB address as operand, not substitution without operand.
> > > >
> > > > How can I add an addr32 prefix with operand substitution? This is
> > > > very similar to "%^". My updated patch will use "%^".
> > >
> > > Yes, using %^ is what I think would be the optimal solution. Other
> > > than that, in your proposed patch, operand-less %_ scans the entire
> > > current_output_insn to dig to the UNSPEC_VSIBADDR. You can just use
> > > operand substitution, and do e.g. "%X2vgatherpf0..." where 'X'
> > > processes operand 2 (vsib_address_operand) and conditionally outputs
> > > addr32.
> > >
> > > BTW: In a new version of the patch, please specify what is changed
> > > from the previous version. Otherwise, review of a new version is more
> > > or less a guesswork what changed.
> > >
> >
> > Here is the updated patch. The change is
> >
> > return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> >
> > instead of
> >
> > return "%^vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
>
> Did I miss some version of the patch that introduced %^? You used %_
> in your previous patch. Did your try with %^?
Yes. It is very similar to
https://gcc.gnu.org/ml/gcc-patches/2019-02/msg02109.html
with
+ case '^':
+ if (TARGET_X32)
+ {
+ subrtx_var_iterator::array_type array;
+ FOR_EACH_SUBRTX_VAR (iter, array,
+ PATTERN (current_output_insn), ALL)
+ {
+ rtx addr = *iter;
+ if (!MEM_P (addr))
+ continue;
+ addr = XEXP (addr, 0);
+ if (GET_CODE (addr) == UNSPEC
+ && XINT (addr, 1) == UNSPEC_VSIBADDR)
+ {
+ /* NB: 32-bit indices in VSIB address are
+ sign-extended to 64 bits. In x32, if 32-bit
+ address 0xf7fa3010 is sign-extended to
+ 0xfffffffff7fa3010 which is invalid address.
+ Add addr32 prefix if there is no base register
+ nor symbol. */
+ bool ok;
+ struct ix86_address parts;
+ ok = ix86_decompose_address (XVECEXP (addr, 0, 0),
+ &parts);
+ gcc_assert (ok && parts.index == NULL_RTX);
+ if (parts.base == NULL_RTX
+ && (parts.disp == NULL_RTX
+ || !symbolic_operand (parts.disp,
+ GET_MODE (parts.disp))))
+ fputs ("addr32 ", file);
+ break;
+ }
+ }
+ }
> > We can't use the %X5 since %X5 is used on operands.
>
> So, please introduce some other modifier ("X" was not to be taken
> literally, but *some* letter). Why are you overloading 'P'?
Here is the updated patch with the 'M' modifier.
H.J.
> I don't know why are you using operand 5 here, you can use operand 2 directly.
> Uros.
>
> > I also added a test for -maddress-mode=long.
> >
> >
> > H.J.
> > ---
> > 32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
> > when 32-bit indices are used as addresses, like in
> >
> > vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
> >
> > 32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
> > is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
> > for x32 if there is no base register nor symbol.
> >
> > This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
> >
> > -Ofast -funroll-loops -march=haswell
> >
> > gcc/
> >
> > PR target/89523
> > * config/i386/i386.c (ix86_print_operand): Handle UNSPEC_VSIBADDR
> > instructions for '%P' to add addr32 prefix if required.
> > * config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
> > "%P5" to opcode.
> > (*avx512pf_gatherpf<mode>df_mask): Likewise.
> > (*avx512pf_scatterpf<mode>sf_mask): Likewise.
> > (*avx512pf_scatterpf<mode>df_mask): Likewise.
> > (*avx2_gathersi<mode>): Prepend "%P7" to opcode.
> > (*avx2_gathersi<mode>_2): Prepend "%P6" to opcode.
> > (*avx2_gatherdi<mode>): Prepend "%P7" to opcode.
> > (*avx2_gatherdi<mode>_2): Prepend "%P6" to opcode.
> > (*avx2_gatherdi<mode>_3): Prepend "%P7" to opcode.
> > (*avx2_gatherdi<mode>_4): Prepend "%P6" to opcode.`
> > (*avx512f_gathersi<mode>): Prepend "%P5" to opcode.
> > (*avx512f_gathersi<mode>_2): Prepend "%P6" to opcode.
> > (*avx512f_gatherdi<mode>): Prepend "%P5" to opcode.
> > (*avx512f_gatherdi<mode>_2): Likewise.
> > (*avx512f_scattersi<mode>): Likewise.
> > (*avx512f_scatterdi<mode>): Likewise.
> >
> > gcc/testsuite/
> >
> > PR target/89523
> > * gcc.target/i386/pr89523-1a.c: New test.
> > * gcc.target/i386/pr89523-1b.c: Likewise.
> > * gcc.target/i386/pr89523-2.c: Likewise.
> > * gcc.target/i386/pr89523-3.c: Likewise.
> > * gcc.target/i386/pr89523-4.c: Likewise.
> > * gcc.target/i386/pr89523-5.c: Likewise.
> > * gcc.target/i386/pr89523-6.c: Likewise.
> > * gcc.target/i386/pr89523-7.c: Likewise.
> > * gcc.target/i386/pr89523-8.c: Likewise.
> > * gcc.target/i386/pr89523-9.c: Likewise.
> > ---
> > gcc/config/i386/i386.c | 35 +++++++++++++++-
> > gcc/config/i386/sse.md | 46 +++++++++++-----------
> > gcc/testsuite/gcc.target/i386/pr89523-1a.c | 24 +++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-1b.c | 7 ++++
> > gcc/testsuite/gcc.target/i386/pr89523-2.c | 37 +++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-3.c | 36 +++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-4.c | 36 +++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-5.c | 39 ++++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-6.c | 38 ++++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-7.c | 42 ++++++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-8.c | 41 +++++++++++++++++++
> > gcc/testsuite/gcc.target/i386/pr89523-9.c | 30 ++++++++++++++
> > 12 files changed, 386 insertions(+), 25 deletions(-)
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1a.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1b.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
> >
> > diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> > index c8f9957163b..ae9befb638d 100644
> > --- a/gcc/config/i386/i386.c
> > +++ b/gcc/config/i386/i386.c
> > @@ -17793,7 +17793,8 @@ print_reg (rtx x, int code, FILE *file)
> > y -- print "st(0)" instead of "st" as a register.
> > d -- print duplicated register operand for AVX instruction.
> > D -- print condition for SSE cmp instruction.
> > - P -- if PIC, print an @PLT suffix.
> > + P -- if PIC, print an @PLT suffix or print addr32 prefix for
> > + TARGET_X32 with UNSPEC_VSIBADDR operand.
> > p -- print raw symbol name.
> > X -- don't print any sort of PIC '@' suffix for a symbol.
> > & -- print some in-use local-dynamic symbol name.
> > @@ -18010,6 +18011,37 @@ ix86_print_operand (FILE *file, rtx x, int code)
> > output_operand_lossage ("invalid operand size for operand code 'Z'");
> > return;
> >
> > + case 'P':
> > + if (MEM_P (x))
> > + {
> > + x = XEXP (x, 0);
> > + if (GET_CODE (x) == UNSPEC
> > + || XINT (x, 1) == UNSPEC_VSIBADDR)
> > + {
> > + if (TARGET_X32)
> > + {
> > + /* NB: 32-bit indices in VSIB address are
> > + sign-extended to 64 bits. In x32, if 32-bit
> > + address 0xf7fa3010 is sign-extended to
> > + 0xfffffffff7fa3010 which is invalid address.
> > + Add addr32 prefix if there is no base register
> > + nor symbol. */
> > + bool ok;
> > + struct ix86_address parts;
> > + ok = ix86_decompose_address (XVECEXP (x, 0, 0),
> > + &parts);
> > + gcc_assert (ok && parts.index == NULL_RTX);
> > + if (parts.base == NULL_RTX
> > + && (parts.disp == NULL_RTX
> > + || !symbolic_operand (parts.disp,
> > + GET_MODE (parts.disp))))
> > + fputs ("addr32 ", file);
> > + }
> > + return;
> > + }
> > + }
> > + break;
> > +
> > case 'd':
> > case 'b':
> > case 'w':
> > @@ -18021,7 +18053,6 @@ ix86_print_operand (FILE *file, rtx x, int code)
> > case 'y':
> > case 'x':
> > case 'X':
> > - case 'P':
> > case 'p':
> > case 'V':
> > break;
> > diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> > index ac299495b2c..ac500f9cc63 100644
> > --- a/gcc/config/i386/sse.md
> > +++ b/gcc/config/i386/sse.md
> > @@ -17401,9 +17401,9 @@
> > case 3:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > - return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17448,9 +17448,9 @@
> > case 3:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > - return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17496,10 +17496,10 @@
> > case 7:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > case 6:
> > - return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -17545,10 +17545,10 @@
> > case 7:
> > /* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > gas changed what it requires incompatibly. */
> > - return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > case 2:
> > case 6:
> > - return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > + return "%P5vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
> > default:
> > gcc_unreachable ();
> > }
> > @@ -20292,7 +20292,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> > + "%P7v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20312,7 +20312,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> > + "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20353,7 +20353,7 @@
> > UNSPEC_GATHER))
> > (clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> > + "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20375,8 +20375,8 @@
> > "TARGET_AVX2"
> > {
> > if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> > + return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
> > + return "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
> > }
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > @@ -20400,7 +20400,7 @@
> > (const_int 2) (const_int 3)])))
> > (clobber (match_scratch:VI4F_256 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> > + "%P7v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20423,7 +20423,7 @@
> > (const_int 2) (const_int 3)])))
> > (clobber (match_scratch:VI4F_256 1 "=&x"))]
> > "TARGET_AVX2"
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> > + "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "vex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20463,7 +20463,7 @@
> > "TARGET_AVX512F"
> > ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> > + "%P6v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20484,7 +20484,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> > + "%P5v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20525,7 +20525,7 @@
> > "TARGET_AVX512F"
> > ;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> > + "%P6v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20550,11 +20550,11 @@
> > if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
> > {
> > if (<MODE_SIZE> != 64)
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> > + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
> > else
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> > + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
> > }
> > - return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> > + return "%P5v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
> > }
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > @@ -20593,7 +20593,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > + "%P5v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > @@ -20631,7 +20631,7 @@
> > "TARGET_AVX512F"
> > ;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
> > ;; gas changed what it requires incompatibly.
> > - "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > + "%P5v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "evex")
> > (set_attr "mode" "<sseinsnmode>")])
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1a.c b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
> > new file mode 100644
> > index 00000000000..0d0edab0363
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
> > @@ -0,0 +1,24 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-maddress-mode=short -mx32 -Ofast -funroll-loops -march=haswell" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +void foo (void);
> > +
> > +extern float *ncost;
> > +
> > +float
> > +bar (int type, int num)
> > +{
> > + int i;
> > + float cost;
> > +
> > + cost = 0;
> > + for (i = 0; i < num; i++)
> > + if (type)
> > + cost += ncost[i];
> > + else
> > + foo ();
> > + return (cost);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1b.c b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
> > new file mode 100644
> > index 00000000000..6a5c1d43625
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
> > @@ -0,0 +1,7 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-maddress-mode=long -mx32 -Ofast -funroll-loops -march=haswell" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +#include "pr89523-1a.c"
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> > new file mode 100644
> > index 00000000000..2ffbffe5e40
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
> > @@ -0,0 +1,37 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> > +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> > +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> > +
> > +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> > +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> > +
> > +extern __inline __m128d
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
> > +{
> > + __v2df __zero = { 0.0, 0.0 };
> > + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> > + __v2df x = x;
> > +
> > + return (__m128d) __builtin_ia32_gathersiv2df (x,
> > + __base,
> > + (__v4si)__index,
> > + __mask,
> > + __scale);
> > +}
> > +
> > +__m128d x;
> > +double *base;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i32gather_pd (base, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> > new file mode 100644
> > index 00000000000..fc3631b694b
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
> > @@ -0,0 +1,36 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> > +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> > +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> > +
> > +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> > +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> > +
> > +extern __inline __m128d
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm_i64gather_pd (double const *__base, __m128i __index, const int __scale)
> > +{
> > + __v2df __zero = { 0.0, 0.0 };
> > + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> > +
> > + return (__m128d) __builtin_ia32_gatherdiv2df (__zero,
> > + __base,
> > + (__v2di)__index,
> > + __mask,
> > + __scale);
> > +}
> > +
> > +__m128d x;
> > +double *base;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i64gather_pd (base, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> > new file mode 100644
> > index 00000000000..3436e5dcae3
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
> > @@ -0,0 +1,36 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -march=haswell" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +typedef double __v2df __attribute__ ((__vector_size__ (16)));
> > +typedef int __v4si __attribute__ ((__vector_size__ (16)));
> > +typedef long long __v2di __attribute__ ((__vector_size__ (16)));
> > +
> > +typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
> > +typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
> > +
> > +extern __inline __m128d
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
> > +{
> > + __v2df __zero = { 0.0, 0.0 };
> > + __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
> > + __v2df x = x;
> > +
> > + return (__m128d) __builtin_ia32_gathersiv2df (x,
> > + __base,
> > + (__v4si)__index,
> > + __mask,
> > + __scale);
> > +}
> > +
> > +__m128d x;
> > +__m128i idx;
> > +
> > +void extern
> > +avx2_test (void)
> > +{
> > + x = _mm_i32gather_pd ((void *) 0, idx, 1);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> > new file mode 100644
> > index 00000000000..6a769c7a249
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
> > @@ -0,0 +1,39 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> > +/* { dg-final { scan-assembler "\tvgather" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vgather" } } */
> > +
> > +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> > +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> > + __may_alias__));
> > +typedef unsigned char __mmask8;
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
> > + int __scale, int __hint)
> > +{
> > + __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
> > + __scale, __hint);
> > +}
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
> > + void const *__addr, int __scale, int __hint)
> > +{
> > + __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
> > + __hint);
> > +}
> > +
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +void *base;
> > +
> > +void extern
> > +avx512pf_test (void)
> > +{
> > + _mm512_prefetch_i32gather_pd (idx, base, 8, 3);
> > + _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, 3);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> > new file mode 100644
> > index 00000000000..82f795e085c
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
> > @@ -0,0 +1,38 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512pf" } */
> > +/* { dg-final { scan-assembler-not "\tvgather" } } */
> > +/* { dg-final { scan-assembler "addr32 vgather" } } */
> > +
> > +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> > +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> > + __may_alias__));
> > +typedef unsigned char __mmask8;
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
> > + int __scale, int __hint)
> > +{
> > + __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
> > + __scale, __hint);
> > +}
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
> > + void const *__addr, int __scale, int __hint)
> > +{
> > + __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
> > + __hint);
> > +}
> > +
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +
> > +void extern
> > +avx512pf_test (void)
> > +{
> > + _mm512_prefetch_i32gather_pd (idx, (void *) 0, 8, 3);
> > + _mm512_mask_prefetch_i32gather_pd (idx, m8, (void *) 0, 8, 3);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> > new file mode 100644
> > index 00000000000..030b00d268a
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
> > @@ -0,0 +1,42 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler "\tvscatter" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> > +
> > +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> > +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> > +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> > + __may_alias__));
> > +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> > +typedef unsigned char __mmask8;
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
> > + int __scale)
> > +{
> > + __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
> > + (__v8si) __index, (__v8df) __v1, __scale);
> > +}
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_mask_i32scatter_pd (void *__addr, __mmask8 __mask,
> > + __m256i __index, __m512d __v1, int __scale)
> > +{
> > + __builtin_ia32_scattersiv8df (__addr, __mask, (__v8si) __index,
> > + (__v8df) __v1, __scale);
> > +}
> > +
> > +volatile __m512d src;
> > +volatile __m256i idx;
> > +volatile __mmask8 m8;
> > +double *addr;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i32scatter_pd (addr, idx, src, 8);
> > + _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> > new file mode 100644
> > index 00000000000..465c985c2b7
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
> > @@ -0,0 +1,41 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler "\tvscatter" } } */
> > +/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
> > +
> > +typedef long long __v8di __attribute__ ((__vector_size__ (64)));
> > +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> > +typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
> > +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> > +typedef unsigned char __mmask8;
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_i64scatter_pd (void *__addr, __m512i __index, __m512d __v1,
> > + int __scale)
> > +{
> > + __builtin_ia32_scatterdiv8df (__addr, (__mmask8) 0xFF,
> > + (__v8di) __index, (__v8df) __v1, __scale);
> > +}
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_mask_i64scatter_pd (void *__addr, __mmask8 __mask,
> > + __m512i __index, __m512d __v1, int __scale)
> > +{
> > + __builtin_ia32_scatterdiv8df (__addr, __mask, (__v8di) __index,
> > + (__v8df) __v1, __scale);
> > +}
> > +
> > +volatile __m512d src;
> > +volatile __m512i idx;
> > +volatile __mmask8 m8;
> > +double *addr;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i64scatter_pd (addr, idx, src, 8);
> > + _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
> > +}
> > diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> > new file mode 100644
> > index 00000000000..e9323126bd6
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
> > @@ -0,0 +1,30 @@
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-require-effective-target maybe_x32 } */
> > +/* { dg-options "-mx32 -O2 -mavx512f" } */
> > +/* { dg-final { scan-assembler-not "\tvscatter" } } */
> > +/* { dg-final { scan-assembler "addr32 vscatter" } } */
> > +
> > +typedef int __v8si __attribute__ ((__vector_size__ (32)));
> > +typedef double __v8df __attribute__ ((__vector_size__ (64)));
> > +typedef long long __m256i __attribute__ ((__vector_size__ (32),
> > + __may_alias__));
> > +typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
> > +typedef unsigned char __mmask8;
> > +
> > +extern __inline void
> > +__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
> > +_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
> > + int __scale)
> > +{
> > + __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
> > + (__v8si) __index, (__v8df) __v1, __scale);
> > +}
> > +
> > +volatile __m512d src;
> > +volatile __m256i idx;
> > +
> > +void extern
> > +avx512f_test (void)
> > +{
> > + _mm512_i32scatter_pd ((void *) 0, idx, src, 8);
> > +}
> > --
> > 2.20.1
> >
--
H.J.
[-- Attachment #2: 0001-x32-Add-addr32-prefix-to-VSIB-address.patch --]
[-- Type: text/x-patch, Size: 25088 bytes --]
From 5533e852483b8fbebab6da2cff7c36b8b51e227e Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Wed, 27 Feb 2019 15:46:51 -0800
Subject: [PATCH] x32: Add addr32 prefix to VSIB address
32-bit indices in VSIB address are sign-extended to 64 bits. In x32,
when 32-bit indices are used as addresses, like in
vgatherdps %ymm7, 0(,%ymm9,1), %ymm6
32-bit indices, 0xf7fa3010, is sign-extended to 0xfffffffff7fa3010 which
is invalid address. Add addr32 prefix to UNSPEC_VSIBADDR instructions
for x32 if there is no base register nor symbol.
This fixes 175.vpr and 254.gap in SPEC CPU 2000 on x32 with
-Ofast -funroll-loops -march=haswell
gcc/
PR target/89523
* config/i386/i386.c (ix86_print_operand): Handle 'M' to add
addr32 prefix to VSIB address for X32.
* config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
"%M2" to opcode.
(*avx512pf_gatherpf<mode>df_mask): Likewise.
(*avx512pf_scatterpf<mode>sf_mask): Likewise.
(*avx512pf_scatterpf<mode>df_mask): Likewise.
(*avx2_gathersi<mode>): Prepend "%M3" to opcode.
(*avx2_gathersi<mode>_2): Prepend "%M2" to opcode.
(*avx2_gatherdi<mode>): Prepend "%M3" to opcode.
(*avx2_gatherdi<mode>_2): Prepend "%M2" to opcode.
(*avx2_gatherdi<mode>_3): Prepend "%M3" to opcode.
(*avx2_gatherdi<mode>_4): Prepend "%M2" to opcode.`
(*avx512f_gathersi<mode>): Prepend "%M4" to opcode.
(*avx512f_gathersi<mode>_2): Prepend "%M3" to opcode.
(*avx512f_gatherdi<mode>): Prepend "%M4" to opcode.
(*avx512f_gatherdi<mode>_2): Prepend "%M3" to opcode.
(*avx512f_scattersi<mode>): Prepend "%M0" to opcode.
(*avx512f_scatterdi<mode>): Likewise.
gcc/testsuite/
PR target/89523
* gcc.target/i386/pr89523-1a.c: New test.
* gcc.target/i386/pr89523-1b.c: Likewise.
* gcc.target/i386/pr89523-2.c: Likewise.
* gcc.target/i386/pr89523-3.c: Likewise.
* gcc.target/i386/pr89523-4.c: Likewise.
* gcc.target/i386/pr89523-5.c: Likewise.
* gcc.target/i386/pr89523-6.c: Likewise.
* gcc.target/i386/pr89523-7.c: Likewise.
* gcc.target/i386/pr89523-8.c: Likewise.
* gcc.target/i386/pr89523-9.c: Likewise.
---
gcc/config/i386/i386.c | 21 ++++++++++
gcc/config/i386/sse.md | 46 +++++++++++-----------
gcc/testsuite/gcc.target/i386/pr89523-1a.c | 24 +++++++++++
gcc/testsuite/gcc.target/i386/pr89523-1b.c | 7 ++++
gcc/testsuite/gcc.target/i386/pr89523-2.c | 37 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-3.c | 36 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-4.c | 36 +++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-5.c | 39 ++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-6.c | 38 ++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-7.c | 42 ++++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-8.c | 41 +++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr89523-9.c | 30 ++++++++++++++
12 files changed, 374 insertions(+), 23 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1a.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-1b.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-2.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-3.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-4.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-5.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-6.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-7.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-8.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr89523-9.c
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index fe459071aaf..1f94a45909d 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -17813,6 +17813,7 @@ print_reg (rtx x, int code, FILE *file)
; -- print a semicolon (after prefixes due to bug in older gas).
~ -- print "i" if TARGET_AVX2, "f" otherwise.
^ -- print addr32 prefix if TARGET_64BIT and Pmode != word_mode
+ M -- print addr32 prefix for TARGET_X32 with VSIB address.
! -- print NOTRACK prefix for jxx/call/ret instructions if required.
*/
@@ -18360,6 +18361,26 @@ ix86_print_operand (FILE *file, rtx x, int code)
putc (TARGET_AVX2 ? 'i' : 'f', file);
return;
+ case 'M':
+ if (TARGET_X32)
+ {
+ /* NB: 32-bit indices in VSIB address are sign-extended
+ to 64 bits. In x32, if 32-bit address 0xf7fa3010 is
+ sign-extended to 0xfffffffff7fa3010 which is invalid
+ address. Add addr32 prefix if there is no base
+ register nor symbol. */
+ bool ok;
+ struct ix86_address parts;
+ ok = ix86_decompose_address (x, &parts);
+ gcc_assert (ok && parts.index == NULL_RTX);
+ if (parts.base == NULL_RTX
+ && (parts.disp == NULL_RTX
+ || !symbolic_operand (parts.disp,
+ GET_MODE (parts.disp))))
+ fputs ("addr32 ", file);
+ }
+ return;
+
case '^':
if (TARGET_64BIT && Pmode != word_mode)
fputs ("addr32 ", file);
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index d4c01407f4a..86bf8661808 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -17486,9 +17486,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vgatherpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vgatherpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17533,9 +17533,9 @@
case 3:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vgatherpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
- return "vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vgatherpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17581,10 +17581,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vscatterpf0<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vscatterpf1<ssemodesuffix>ps\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -17630,10 +17630,10 @@
case 7:
/* %X5 so that we don't emit any *WORD PTR for -masm=intel, as
gas changed what it requires incompatibly. */
- return "vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vscatterpf0<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
case 2:
case 6:
- return "vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
+ return "%M2vscatterpf1<ssemodesuffix>pd\t{%5%{%0%}|%X5%{%0%}}";
default:
gcc_unreachable ();
}
@@ -20377,7 +20377,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
+ "%M3v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %7, %0|%0, %7, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20397,7 +20397,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
+ "%M2v<sseintprefix>gatherd<ssemodesuffix>\t{%1, %6, %0|%0, %6, %1}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20438,7 +20438,7 @@
UNSPEC_GATHER))
(clobber (match_scratch:VEC_GATHER_MODE 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
+ "%M3v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %2|%2, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20460,8 +20460,8 @@
"TARGET_AVX2"
{
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
+ return "%M2v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %x0|%x0, %6, %4}";
+ return "%M2v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
@@ -20485,7 +20485,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
+ "%M3v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %7, %0|%0, %7, %5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20508,7 +20508,7 @@
(const_int 2) (const_int 3)])))
(clobber (match_scratch:VI4F_256 1 "=&x"))]
"TARGET_AVX2"
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
+ "%M2v<sseintprefix>gatherq<ssemodesuffix>\t{%4, %6, %0|%0, %6, %4}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "vex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20548,7 +20548,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
+ "%M4v<sseintprefix>gatherd<ssemodesuffix>\t{%6, %0%{%2%}|%0%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20569,7 +20569,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
+ "%M3v<sseintprefix>gatherd<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20610,7 +20610,7 @@
"TARGET_AVX512F"
;; %X6 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
+ "%M4v<sseintprefix>gatherq<ssemodesuffix>\t{%6, %1%{%2%}|%1%{%2%}, %X6}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20635,11 +20635,11 @@
if (<MODE>mode != <VEC_GATHER_SRCDI>mode)
{
if (<MODE_SIZE> != 64)
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
+ return "%M3v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %x0%{%1%}|%x0%{%1%}, %X5}";
else
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
+ return "%M3v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %t0%{%1%}|%t0%{%1%}, %X5}";
}
- return "v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
+ return "%M3v<sseintprefix>gatherq<ssemodesuffix>\t{%5, %0%{%1%}|%0%{%1%}, %X5}";
}
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
@@ -20678,7 +20678,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%M0v<sseintprefix>scatterd<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
@@ -20716,7 +20716,7 @@
"TARGET_AVX512F"
;; %X5 so that we don't emit any *WORD PTR for -masm=intel, as
;; gas changed what it requires incompatibly.
- "v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
+ "%M0v<sseintprefix>scatterq<ssemodesuffix>\t{%3, %5%{%1%}|%X5%{%1%}, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix" "evex")
(set_attr "mode" "<sseinsnmode>")])
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1a.c b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
new file mode 100644
index 00000000000..0d0edab0363
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-1a.c
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-maddress-mode=short -mx32 -Ofast -funroll-loops -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+void foo (void);
+
+extern float *ncost;
+
+float
+bar (int type, int num)
+{
+ int i;
+ float cost;
+
+ cost = 0;
+ for (i = 0; i < num; i++)
+ if (type)
+ cost += ncost[i];
+ else
+ foo ();
+ return (cost);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-1b.c b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
new file mode 100644
index 00000000000..6a5c1d43625
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-1b.c
@@ -0,0 +1,7 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-maddress-mode=long -mx32 -Ofast -funroll-loops -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+#include "pr89523-1a.c"
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-2.c b/gcc/testsuite/gcc.target/i386/pr89523-2.c
new file mode 100644
index 00000000000..2ffbffe5e40
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-2.c
@@ -0,0 +1,37 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+ __v2df x = x;
+
+ return (__m128d) __builtin_ia32_gathersiv2df (x,
+ __base,
+ (__v4si)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-3.c b/gcc/testsuite/gcc.target/i386/pr89523-3.c
new file mode 100644
index 00000000000..fc3631b694b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-3.c
@@ -0,0 +1,36 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i64gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+
+ return (__m128d) __builtin_ia32_gatherdiv2df (__zero,
+ __base,
+ (__v2di)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+double *base;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i64gather_pd (base, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-4.c b/gcc/testsuite/gcc.target/i386/pr89523-4.c
new file mode 100644
index 00000000000..3436e5dcae3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-4.c
@@ -0,0 +1,36 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -march=haswell" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+typedef double __v2df __attribute__ ((__vector_size__ (16)));
+typedef int __v4si __attribute__ ((__vector_size__ (16)));
+typedef long long __v2di __attribute__ ((__vector_size__ (16)));
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
+
+extern __inline __m128d
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm_i32gather_pd (double const *__base, __m128i __index, const int __scale)
+{
+ __v2df __zero = { 0.0, 0.0 };
+ __v2df __mask = __builtin_ia32_cmpeqpd (__zero, __zero);
+ __v2df x = x;
+
+ return (__m128d) __builtin_ia32_gathersiv2df (x,
+ __base,
+ (__v4si)__index,
+ __mask,
+ __scale);
+}
+
+__m128d x;
+__m128i idx;
+
+void extern
+avx2_test (void)
+{
+ x = _mm_i32gather_pd ((void *) 0, idx, 1);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-5.c b/gcc/testsuite/gcc.target/i386/pr89523-5.c
new file mode 100644
index 00000000000..6a769c7a249
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-5.c
@@ -0,0 +1,39 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler "\tvgather" } } */
+/* { dg-final { scan-assembler-not "addr32 vgather" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
+ int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
+ __scale, __hint);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
+ void const *__addr, int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
+ __hint);
+}
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+void *base;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, base, 8, 3);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, base, 8, 3);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-6.c b/gcc/testsuite/gcc.target/i386/pr89523-6.c
new file mode 100644
index 00000000000..82f795e085c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-6.c
@@ -0,0 +1,38 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512pf" } */
+/* { dg-final { scan-assembler-not "\tvgather" } } */
+/* { dg-final { scan-assembler "addr32 vgather" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
+ int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
+ __scale, __hint);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
+ void const *__addr, int __scale, int __hint)
+{
+ __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
+ __hint);
+}
+
+volatile __m256i idx;
+volatile __mmask8 m8;
+
+void extern
+avx512pf_test (void)
+{
+ _mm512_prefetch_i32gather_pd (idx, (void *) 0, 8, 3);
+ _mm512_mask_prefetch_i32gather_pd (idx, m8, (void *) 0, 8, 3);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-7.c b/gcc/testsuite/gcc.target/i386/pr89523-7.c
new file mode 100644
index 00000000000..030b00d268a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-7.c
@@ -0,0 +1,42 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
+ (__v8si) __index, (__v8df) __v1, __scale);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_i32scatter_pd (void *__addr, __mmask8 __mask,
+ __m256i __index, __m512d __v1, int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, __mask, (__v8si) __index,
+ (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m256i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i32scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-8.c b/gcc/testsuite/gcc.target/i386/pr89523-8.c
new file mode 100644
index 00000000000..465c985c2b7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-8.c
@@ -0,0 +1,41 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler "\tvscatter" } } */
+/* { dg-final { scan-assembler-not "addr32 vscatter" } } */
+
+typedef long long __v8di __attribute__ ((__vector_size__ (64)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i64scatter_pd (void *__addr, __m512i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scatterdiv8df (__addr, (__mmask8) 0xFF,
+ (__v8di) __index, (__v8df) __v1, __scale);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_mask_i64scatter_pd (void *__addr, __mmask8 __mask,
+ __m512i __index, __m512d __v1, int __scale)
+{
+ __builtin_ia32_scatterdiv8df (__addr, __mask, (__v8di) __index,
+ (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m512i idx;
+volatile __mmask8 m8;
+double *addr;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i64scatter_pd (addr, idx, src, 8);
+ _mm512_mask_i64scatter_pd (addr, m8, idx, src, 8);
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr89523-9.c b/gcc/testsuite/gcc.target/i386/pr89523-9.c
new file mode 100644
index 00000000000..e9323126bd6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr89523-9.c
@@ -0,0 +1,30 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-require-effective-target maybe_x32 } */
+/* { dg-options "-mx32 -O2 -mavx512f" } */
+/* { dg-final { scan-assembler-not "\tvscatter" } } */
+/* { dg-final { scan-assembler "addr32 vscatter" } } */
+
+typedef int __v8si __attribute__ ((__vector_size__ (32)));
+typedef double __v8df __attribute__ ((__vector_size__ (64)));
+typedef long long __m256i __attribute__ ((__vector_size__ (32),
+ __may_alias__));
+typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));
+typedef unsigned char __mmask8;
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+_mm512_i32scatter_pd (void *__addr, __m256i __index, __m512d __v1,
+ int __scale)
+{
+ __builtin_ia32_scattersiv8df (__addr, (__mmask8) 0xFF,
+ (__v8si) __index, (__v8df) __v1, __scale);
+}
+
+volatile __m512d src;
+volatile __m256i idx;
+
+void extern
+avx512f_test (void)
+{
+ _mm512_i32scatter_pd ((void *) 0, idx, src, 8);
+}
--
2.20.1
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: V2 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions
2019-03-14 8:34 ` H.J. Lu
@ 2019-03-14 8:47 ` Uros Bizjak
0 siblings, 0 replies; 8+ messages in thread
From: Uros Bizjak @ 2019-03-14 8:47 UTC (permalink / raw)
To: gcc-patches; +Cc: H.J. Lu
On Thu, Mar 14, 2019 at 9:22 AM H.J. Lu <hjl.tools@gmail.com> wrote:
> > > We can't use the %X5 since %X5 is used on operands.
> >
> > So, please introduce some other modifier ("X" was not to be taken
> > literally, but *some* letter). Why are you overloading 'P'?
>
> Here is the updated patch with the 'M' modifier.
>
>
> gcc/
>
> PR target/89523
> * config/i386/i386.c (ix86_print_operand): Handle 'M' to add
> addr32 prefix to VSIB address for X32.
> * config/i386/sse.md (*avx512pf_gatherpf<mode>sf_mask): Prepend
> "%M2" to opcode.
> (*avx512pf_gatherpf<mode>df_mask): Likewise.
> (*avx512pf_scatterpf<mode>sf_mask): Likewise.
> (*avx512pf_scatterpf<mode>df_mask): Likewise.
> (*avx2_gathersi<mode>): Prepend "%M3" to opcode.
> (*avx2_gathersi<mode>_2): Prepend "%M2" to opcode.
> (*avx2_gatherdi<mode>): Prepend "%M3" to opcode.
> (*avx2_gatherdi<mode>_2): Prepend "%M2" to opcode.
> (*avx2_gatherdi<mode>_3): Prepend "%M3" to opcode.
> (*avx2_gatherdi<mode>_4): Prepend "%M2" to opcode.`
> (*avx512f_gathersi<mode>): Prepend "%M4" to opcode.
> (*avx512f_gathersi<mode>_2): Prepend "%M3" to opcode.
> (*avx512f_gatherdi<mode>): Prepend "%M4" to opcode.
> (*avx512f_gatherdi<mode>_2): Prepend "%M3" to opcode.
> (*avx512f_scattersi<mode>): Prepend "%M0" to opcode.
> (*avx512f_scatterdi<mode>): Likewise.
>
> gcc/testsuite/
>
> PR target/89523
> * gcc.target/i386/pr89523-1a.c: New test.
> * gcc.target/i386/pr89523-1b.c: Likewise.
> * gcc.target/i386/pr89523-2.c: Likewise.
> * gcc.target/i386/pr89523-3.c: Likewise.
> * gcc.target/i386/pr89523-4.c: Likewise.
> * gcc.target/i386/pr89523-5.c: Likewise.
> * gcc.target/i386/pr89523-6.c: Likewise.
> * gcc.target/i386/pr89523-7.c: Likewise.
> * gcc.target/i386/pr89523-8.c: Likewise.
> * gcc.target/i386/pr89523-9.c: Likewise.
The patch looks safe and relatively non-intrusive, so OK for gcc-9 and
backports.
Thanks,
Uros.
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2019-03-14 8:45 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-02-28 19:22 [PATCH] x32: Add addr32 prefix to UNSPEC_VSIBADDR instructions H.J. Lu
2019-03-03 17:27 ` Uros Bizjak
2019-03-03 21:18 ` H.J. Lu
2019-03-03 21:34 ` Uros Bizjak
2019-03-04 13:54 ` V2 " H.J. Lu
2019-03-04 14:09 ` Uros Bizjak
2019-03-14 8:34 ` H.J. Lu
2019-03-14 8:47 ` Uros Bizjak
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).