[1/2,Middle-end] Simplify (trunc)copysign((extend)a, (extend)b) to .COPYSIGN (a, b).

a and b are same type as the truncation type and has less precision
than extend type.

Bootstrapped and regtested on x86-64-pc-linux-gnu{-m32,}.
Ok for trunk?

gcc/ChangeLog:

	PR target/102464
	* match.pd: simplify (trunc)copysign((extend)a, (extend)b) to
	.COPYSIGN (a,b) when a and b are same type as the truncation
	type and has less precision than extend type.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/pr102464-copysign-1.c: New test.
---
 gcc/match.pd                                  | 13 +++
 .../gcc.target/i386/pr102464-copysign-1.c     | 80 +++++++++++++++++++
 2 files changed, 93 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/i386/pr102464-copysign-1.c

On Thu, Nov 4, 2021 at 7:45 AM liuhongt <hongtao.liu@intel.com> wrote:
>
> a and b are same type as the truncation type and has less precision
> than extend type.
>
> Bootstrapped and regtested on x86-64-pc-linux-gnu{-m32,}.
> Ok for trunk?

OK.

Richard.

> gcc/ChangeLog:
>
>         PR target/102464
>         * match.pd: simplify (trunc)copysign((extend)a, (extend)b) to
>         .COPYSIGN (a,b) when a and b are same type as the truncation
>         type and has less precision than extend type.
>
> gcc/testsuite/ChangeLog:
>
>         * gcc.target/i386/pr102464-copysign-1.c: New test.
> ---
>  gcc/match.pd                                  | 13 +++
>  .../gcc.target/i386/pr102464-copysign-1.c     | 80 +++++++++++++++++++
>  2 files changed, 93 insertions(+)
>  create mode 100644 gcc/testsuite/gcc.target/i386/pr102464-copysign-1.c
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 0734c45700c..f63079023d0 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -6169,6 +6169,19 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
>           && direct_internal_fn_supported_p (as_internal_fn (tos),
>                                              type, OPTIMIZE_FOR_BOTH))
>         (tos @0))))
> +
> +/* Simplify (trunc) copysign ((extend)x, (extend)y) to copysignf (x, y),
> +   x,y is float value, similar for _Float16/double.  */
> +(for copysigns (COPYSIGN_ALL)
> + (simplify
> +  (convert (copysigns (convert@2 @0) (convert @1)))
> +   (if (optimize
> +       && types_match (type, TREE_TYPE (@0))
> +       && types_match (type, TREE_TYPE (@1))
> +       && TYPE_PRECISION (type) < TYPE_PRECISION (TREE_TYPE (@2))
> +       && direct_internal_fn_supported_p (IFN_COPYSIGN,
> +                                         type, OPTIMIZE_FOR_BOTH))
> +    (IFN_COPYSIGN @0 @1))))
>  #endif
>
>  (for froms (XFLOORL XCEILL XROUNDL XRINTL)
> diff --git a/gcc/testsuite/gcc.target/i386/pr102464-copysign-1.c b/gcc/testsuite/gcc.target/i386/pr102464-copysign-1.c
> new file mode 100644
> index 00000000000..95a39509738
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr102464-copysign-1.c
> @@ -0,0 +1,80 @@
> +/* PR target/102464.  */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512fp16 -mavx512vl -ftree-vectorize -mfpmath=sse -fdump-tree-optimized" } */
> +
> +#include<math.h>
> +void foo1 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  for (int i = 0; i != 8; i++)
> +    a[i] = copysignf (b[i], c[i]);
> +}
> +
> +void foo2 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  for (int i = 0; i != 8; i++)
> +    a[i] = copysign (b[i], c[i]);
> +}
> +
> +void foo3 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  for (int i = 0; i != 8; i++)
> +    a[i] = copysignl (b[i], c[i]);
> +}
> +
> +void foo4 (float* __restrict a, float* b, float* c)
> +{
> +  for (int i = 0; i != 4; i++)
> +    a[i] = copysign (b[i], c[i]);
> +}
> +
> +void foo5 (float* __restrict a, float* b, float* c)
> +{
> +  for (int i = 0; i != 4; i++)
> +    a[i] = copysignl (b[i], c[i]);
> +}
> +
> +void foo6 (double* __restrict a, double* b, double* c)
> +{
> +  for (int i = 0; i != 4; i++)
> +    a[i] = copysignl (b[i], c[i]);
> +}
> +
> +void foo7 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  a[0] = copysignf (b[0], c[0]);
> +}
> +
> +void foo8 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  a[0] = copysign (b[0], c[0]);
> +}
> +
> +void foo9 (_Float16* __restrict a, _Float16* b, _Float16* c)
> +{
> +  a[0] = copysignl (b[0], c[0]);
> +}
> +
> +void foo10 (float* __restrict a, float* b, float* c)
> +{
> +  a[0] = copysign (b[0], c[0]);
> +}
> +
> +void foo11 (float* __restrict a, float* b, float* c)
> +{
> +  a[0] = copysignl (b[0], c[0]);
> +}
> +
> +void foo12 (double* __restrict a, double* b, double* c)
> +{
> +  a[0] = copysignl (b[0], c[0]);
> +}
> +
> +/* { dg-final { scan-assembler-not "vcvtsh2s\[sd\]" } } */
> +/* { dg-final { scan-assembler-not "vcvtss2sd" } } */
> +/* { dg-final { scan-assembler-not "fld" } } */
> +/* { dg-final { scan-assembler-not "vcvtph2p\[sd\]" } } */
> +/* { dg-final { scan-assembler-not "vcvtps2pd" } } */
> +/* { dg-final { scan-assembler-not "extendhfxf" } } */
> +/* { dg-final { scan-assembler-not "\\\{1to8\\\}" } } */
> +/* { dg-final { scan-tree-dump-times "\.COPYSIGN" 12 "optimized" } } */
> +/* { dg-final { scan-assembler-times "vpternlog" 12 } } */
> --
> 2.18.1
>

On Thu, 4 Nov 2021, liuhongt via Gcc-patches wrote:

> a and b are same type as the truncation type and has less precision
> than extend type.

Note that this is not safe with -fsignaling-nans, so needs to be disabled 
for that option (if there isn't already logic somewhere with that effect), 
because the extend will convert a signaling NaN to quiet (raising 
"invalid"), but copysign won't, so this transformation could result in a 
signaling NaN being wrongly returned when the original code would never 
have returned a signaling NaN.