Re: [PATCH] i386: avoid zero extension for crc32q

[Uros Bizjak via Gcc-patches]

On Tue, Aug 23, 2022 at 6:10 PM Alexander Monakov via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> The crc32q instruction takes 64-bit operands, but ignores high 32 bits
> of the destination operand, and zero-extends the result from 32 bits.
>
> Let's model this in the RTL pattern to avoid zero-extension when the
> _mm_crc32_u64 intrinsic is used with a 32-bit type.
>
>         PR target/106453
>
> gcc/ChangeLog:
>
>         * config/i386/i386.md (sse4_2_crc32di): Model that only low 32
>         bits of operand 0 are consumed, and the result is zero-extended
>         to 64 bits.
>
> gcc/testsuite/ChangeLog:
>
>         * gcc.target/i386/pr106453.c: New test.

OK with a nit and a couple of changes to the testcase dg-directives.

> ---
>  gcc/config/i386/i386.md                  |  6 +++---
>  gcc/testsuite/gcc.target/i386/pr106453.c | 13 +++++++++++++
>  2 files changed, 16 insertions(+), 3 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.target/i386/pr106453.c
>
> diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> index 58fcc382f..b5760bb23 100644
> --- a/gcc/config/i386/i386.md
> +++ b/gcc/config/i386/i386.md
> @@ -23823,10 +23823,10 @@
>
>  (define_insn "sse4_2_crc32di"
>    [(set (match_operand:DI 0 "register_operand" "=r")
> -       (unspec:DI
> -         [(match_operand:DI 1 "register_operand" "0")
> +       (zero_extend:DI (unspec:SI
> +         [(match_operand:SI 1 "register_operand" "0")
>            (match_operand:DI 2 "nonimmediate_operand" "rm")]
> -         UNSPEC_CRC32))]
> +         UNSPEC_CRC32)))]

Usually the (unspec) part comes in the next line.

>    "TARGET_64BIT && TARGET_CRC32"
>    "crc32{q}\t{%2, %0|%0, %2}"
>    [(set_attr "type" "sselog1")
> diff --git a/gcc/testsuite/gcc.target/i386/pr106453.c 
> b/gcc/testsuite/gcc.target/i386/pr106453.c
> new file mode 100644
> index 000000000..bab5b1cb2
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr106453.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-msse4.2 -O2 -fdump-rtl-final" } */
> +/* { dg-final { scan-rtl-dump-not "zero_extendsidi" "final" } } */

This part can use scan-asembler-not directive, with a -dp compiler option:

+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -mcrc32 -dp" } */
+/* { dg-final { scan-assembler-not "zero_extendsidi" } } */

Also, the mainline compiler can use -mcrc32.

Please find all these suggestions implemented in the attached patch.

Thanks,
Uros.

diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 1aef1af594d..57771ed84f5 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -23823,10 +23823,11 @@ (define_insn "sse4_2_crc32<mode>"
 
 (define_insn "sse4_2_crc32di"
   [(set (match_operand:DI 0 "register_operand" "=r")
-       (unspec:DI
-         [(match_operand:DI 1 "register_operand" "0")
-          (match_operand:DI 2 "nonimmediate_operand" "rm")]
-         UNSPEC_CRC32))]
+       (zero_extend:DI
+         (unspec:SI
+           [(match_operand:SI 1 "register_operand" "0")
+            (match_operand:DI 2 "nonimmediate_operand" "rm")]
+           UNSPEC_CRC32)))]
   "TARGET_64BIT && TARGET_CRC32"
   "crc32{q}\t{%2, %0|%0, %2}"
   [(set_attr "type" "sselog1")
diff --git a/gcc/testsuite/gcc.target/i386/pr106453.c 
b/gcc/testsuite/gcc.target/i386/pr106453.c
new file mode 100644
index 00000000000..bd2e7282cf6
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr106453.c
@@ -0,0 +1,13 @@
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -mcrc32 -dp" } */
+/* { dg-final { scan-assembler-not "zero_extendsidi" } } */
+
+#include <immintrin.h>
+#include <stdint.h>
+
+uint32_t f(uint32_t c, uint64_t *p, size_t n)
+{
+    for (size_t i = 0; i < n; i++)
+        c = _mm_crc32_u64(c, p[i]);
+    return c;
+}