[PATCH 1/3] Use cvt_op to save intermediate type operand instead of "subtle" vec_dest.

[PATCH 1/3] Use cvt_op to save intermediate type operand instead of "subtle" vec_dest.

[liuhongt]

When there're multiple operands in vec_oprnds0, vec_dest will be
overwrited to vectype_out, but in multi_step_cvt case, cvt_type is
expected. It caused an ICE when verify_gimple_in_cfg.

Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,} and aarch64-linux-gnu.
Ok for trunk?

gcc/ChangeLog:

	PR tree-optimization/110371
	PR tree-optimization/110018
	* tree-vect-stmts.cc (vectorizable_conversion): Use cvt_op to
	save intermediate type operand instead of "subtle" vec_dest
	for case NONE.

gcc/testsuite/ChangeLog:

	* gcc.target/aarch64/pr110371.c: New test.
---
 gcc/testsuite/gcc.target/aarch64/pr110371.c | 20 ++++++++++++++++++++
 gcc/tree-vect-stmts.cc                      | 14 ++++++++++----
 2 files changed, 30 insertions(+), 4 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/aarch64/pr110371.c

diff --git a/gcc/testsuite/gcc.target/aarch64/pr110371.c b/gcc/testsuite/gcc.target/aarch64/pr110371.c
new file mode 100644
index 00000000000..444e514e04f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/pr110371.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-options "-O3" } */
+
+typedef struct dest
+{
+  double m[3][3];
+} dest;
+
+typedef struct src
+{
+  int m[3][3];
+} src;
+
+void
+foo (dest *a, src* s)
+{
+  for (int i = 0; i != 3; i++)
+    for (int j = 0; j != 3; j++)
+      a->m[i][j] = s->m[i][j];
+}
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 85d1f3ae52c..1748555a625 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -5044,7 +5044,7 @@ vectorizable_conversion (vec_info *vinfo,
 			 gimple **vec_stmt, slp_tree slp_node,
 			 stmt_vector_for_cost *cost_vec)
 {
-  tree vec_dest;
+  tree vec_dest, cvt_op = NULL_TREE;
   tree scalar_dest;
   tree op0, op1 = NULL_TREE;
   loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
@@ -5568,6 +5568,13 @@ vectorizable_conversion (vec_info *vinfo,
     case NONE:
       vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies,
 			 op0, &vec_oprnds0);
+      /* vec_dest is intermediate type operand when multi_step_cvt.  */
+      if (multi_step_cvt)
+	{
+	  cvt_op = vec_dest;
+	  vec_dest = vec_dsts[0];
+	}
+
       FOR_EACH_VEC_ELT (vec_oprnds0, i, vop0)
 	{
 	  /* Arguments are ready, create the new vector stmt.  */
@@ -5575,12 +5582,11 @@ vectorizable_conversion (vec_info *vinfo,
 	  if (multi_step_cvt)
 	    {
 	      gcc_assert (multi_step_cvt == 1);
-	      new_stmt = vect_gimple_build (vec_dest, codecvt1, vop0);
-	      new_temp = make_ssa_name (vec_dest, new_stmt);
+	      new_stmt = vect_gimple_build (cvt_op, codecvt1, vop0);
+	      new_temp = make_ssa_name (cvt_op, new_stmt);
 	      gimple_assign_set_lhs (new_stmt, new_temp);
 	      vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi);
 	      vop0 = new_temp;
-	      vec_dest = vec_dsts[0];
 	    }
 	  new_stmt = vect_gimple_build (vec_dest, code1, vop0);
 	  new_temp = make_ssa_name (vec_dest, new_stmt);
-- 
2.39.1.388.g2fc9e9ca3c

[PATCH 2/3] Don't use intermiediate type for FIX_TRUNC_EXPR when ftrapping-math.

[liuhongt]

> > Hmm, good question.  GENERIC has a direct truncation to unsigned char
> > for example, the C standard generally says if the integral part cannot
> > be represented then the behavior is undefined.  So I think we should be
> > safe here (0x1.0p32 doesn't fit an int).
>
> We should be following Annex F (unspecified value plus "invalid" exception
> for out-of-range floating-to-integer conversions rather than undefined
> behavior).  But we don't achieve that very well at present (see bug 93806
> comments 27-29 for examples of how such conversions produce wobbly
> values).

That would mean guarding this with !flag_trapping_math would be the appropriate
thing to do.

Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,} and aarch64-linux-gnu.
Ok for trunk?

gcc/ChangeLog:

	PR tree-optimization/110371
	PR tree-optimization/110018
	* tree-vect-stmts.cc (vectorizable_conversion): Don't use
	intermiediate type for FIX_TRUNC_EXPR when ftrapping-math.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/pr110018-1.c: Add -fno-trapping-math to dg-options.
	* gcc.target/i386/pr110018-2.c: Ditto.
---
 gcc/testsuite/gcc.target/i386/pr110018-1.c | 2 +-
 gcc/testsuite/gcc.target/i386/pr110018-2.c | 2 +-
 gcc/tree-vect-stmts.cc                     | 3 ++-
 3 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/gcc/testsuite/gcc.target/i386/pr110018-1.c b/gcc/testsuite/gcc.target/i386/pr110018-1.c
index b6a3be7b7a2..24eeca60f6f 100644
--- a/gcc/testsuite/gcc.target/i386/pr110018-1.c
+++ b/gcc/testsuite/gcc.target/i386/pr110018-1.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq" } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq -fno-trapping-math" } */
 /* { dg-final { scan-assembler-times {(?n)vcvttp[dsh]2[dqw]} 5 } } */
 /* { dg-final { scan-assembler-times {(?n)vcvt[dqw]*2p[dsh]} 5 } } */
 
diff --git a/gcc/testsuite/gcc.target/i386/pr110018-2.c b/gcc/testsuite/gcc.target/i386/pr110018-2.c
index a663e074698..9a2d9e17894 100644
--- a/gcc/testsuite/gcc.target/i386/pr110018-2.c
+++ b/gcc/testsuite/gcc.target/i386/pr110018-2.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq" } */
+/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq -fno-trapping-math" } */
 /* { dg-final { scan-assembler-times {(?n)vcvttp[dsh]2[dqw]} 5 } } */
 /* { dg-final { scan-assembler-times {(?n)vcvt[dqw]*2p[dsh]} 5 } } */
 
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 1748555a625..bf61461939b 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -5263,7 +5263,8 @@ vectorizable_conversion (vec_info *vinfo,
       if ((code == FLOAT_EXPR
 	   && GET_MODE_SIZE (lhs_mode) > GET_MODE_SIZE (rhs_mode))
 	  || (code == FIX_TRUNC_EXPR
-	      && GET_MODE_SIZE (rhs_mode) > GET_MODE_SIZE (lhs_mode)))
+	      && GET_MODE_SIZE (rhs_mode) > GET_MODE_SIZE (lhs_mode)
+	      && !flag_trapping_math))
 	{
 	  bool float_expr_p = code == FLOAT_EXPR;
 	  scalar_mode imode = float_expr_p ? rhs_mode : lhs_mode;
-- 
2.39.1.388.g2fc9e9ca3c

Re: [PATCH 2/3] Don't use intermiediate type for FIX_TRUNC_EXPR when ftrapping-math.

[Richard Biener]

On Mon, Jun 26, 2023 at 3:31 AM liuhongt via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> > > Hmm, good question.  GENERIC has a direct truncation to unsigned char
> > > for example, the C standard generally says if the integral part cannot
> > > be represented then the behavior is undefined.  So I think we should be
> > > safe here (0x1.0p32 doesn't fit an int).
> >
> > We should be following Annex F (unspecified value plus "invalid" exception
> > for out-of-range floating-to-integer conversions rather than undefined
> > behavior).  But we don't achieve that very well at present (see bug 93806
> > comments 27-29 for examples of how such conversions produce wobbly
> > values).
>
> That would mean guarding this with !flag_trapping_math would be the appropriate
> thing to do.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,} and aarch64-linux-gnu.
> Ok for trunk?

OK.

Thanks,
Richard.

> gcc/ChangeLog:
>
>         PR tree-optimization/110371
>         PR tree-optimization/110018
>         * tree-vect-stmts.cc (vectorizable_conversion): Don't use
>         intermiediate type for FIX_TRUNC_EXPR when ftrapping-math.
>
> gcc/testsuite/ChangeLog:
>
>         * gcc.target/i386/pr110018-1.c: Add -fno-trapping-math to dg-options.
>         * gcc.target/i386/pr110018-2.c: Ditto.
> ---
>  gcc/testsuite/gcc.target/i386/pr110018-1.c | 2 +-
>  gcc/testsuite/gcc.target/i386/pr110018-2.c | 2 +-
>  gcc/tree-vect-stmts.cc                     | 3 ++-
>  3 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/gcc/testsuite/gcc.target/i386/pr110018-1.c b/gcc/testsuite/gcc.target/i386/pr110018-1.c
> index b6a3be7b7a2..24eeca60f6f 100644
> --- a/gcc/testsuite/gcc.target/i386/pr110018-1.c
> +++ b/gcc/testsuite/gcc.target/i386/pr110018-1.c
> @@ -1,5 +1,5 @@
>  /* { dg-do compile } */
> -/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq" } */
> +/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq -fno-trapping-math" } */
>  /* { dg-final { scan-assembler-times {(?n)vcvttp[dsh]2[dqw]} 5 } } */
>  /* { dg-final { scan-assembler-times {(?n)vcvt[dqw]*2p[dsh]} 5 } } */
>
> diff --git a/gcc/testsuite/gcc.target/i386/pr110018-2.c b/gcc/testsuite/gcc.target/i386/pr110018-2.c
> index a663e074698..9a2d9e17894 100644
> --- a/gcc/testsuite/gcc.target/i386/pr110018-2.c
> +++ b/gcc/testsuite/gcc.target/i386/pr110018-2.c
> @@ -1,5 +1,5 @@
>  /* { dg-do compile } */
> -/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq" } */
> +/* { dg-options "-mavx512fp16 -mavx512vl -O2 -mavx512dq -fno-trapping-math" } */
>  /* { dg-final { scan-assembler-times {(?n)vcvttp[dsh]2[dqw]} 5 } } */
>  /* { dg-final { scan-assembler-times {(?n)vcvt[dqw]*2p[dsh]} 5 } } */
>
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index 1748555a625..bf61461939b 100644
> --- a/gcc/tree-vect-stmts.cc
> +++ b/gcc/tree-vect-stmts.cc
> @@ -5263,7 +5263,8 @@ vectorizable_conversion (vec_info *vinfo,
>        if ((code == FLOAT_EXPR
>            && GET_MODE_SIZE (lhs_mode) > GET_MODE_SIZE (rhs_mode))
>           || (code == FIX_TRUNC_EXPR
> -             && GET_MODE_SIZE (rhs_mode) > GET_MODE_SIZE (lhs_mode)))
> +             && GET_MODE_SIZE (rhs_mode) > GET_MODE_SIZE (lhs_mode)
> +             && !flag_trapping_math))
>         {
>           bool float_expr_p = code == FLOAT_EXPR;
>           scalar_mode imode = float_expr_p ? rhs_mode : lhs_mode;
> --
> 2.39.1.388.g2fc9e9ca3c
>

[PATCH 3/3] [aarch64] Adjust testcase to match assembly output after r14-2007.

[liuhongt]

The new assembly looks better than original one, so I adjust those testcases.
Ok for trunk?

gcc/testsuite/ChangeLog:

	PR tree-optimization/110371
	PR tree-optimization/110018
	* gcc.target/aarch64/sve/unpack_fcvt_signed_1.c: Scan scvt +
	sxtw instead of scvt + zip1 + zip2.
	* gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c: Scan scvt +
	uxtw instead of ucvtf + zip1 + zip2.
---
 gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c | 6 +++---
 .../gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c         | 5 ++---
 2 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
index 0f96dc2ff00..5edc288ce35 100644
--- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
+++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
@@ -10,6 +10,6 @@ unpack_double_int_plus8 (double *d, int32_t *s, int size)
     d[i] = s[i] + 8;
 }
 
-/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
-/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
-/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tsxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
+
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
index 70465f91eba..ecd72176177 100644
--- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
+++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
@@ -10,6 +10,5 @@ unpack_double_int_plus9 (double *d, uint32_t *s, int size)
     d[i] = (double) (s[i] + 9);
 }
 
-/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
-/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
-/* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
+/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
+/* { dg-final { scan-assembler-times {\tuxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
-- 
2.39.1.388.g2fc9e9ca3c

Re: [PATCH 3/3] [aarch64] Adjust testcase to match assembly output after r14-2007.

[Hongtao Liu]

On Mon, Jun 26, 2023 at 9:31 AM liuhongt via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> The new assembly looks better than original one, so I adjust those testcases.
> Ok for trunk?
>
> gcc/testsuite/ChangeLog:
>
>         PR tree-optimization/110371
>         PR tree-optimization/110018
>         * gcc.target/aarch64/sve/unpack_fcvt_signed_1.c: Scan scvt +
>         sxtw instead of scvt + zip1 + zip2.
>         * gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c: Scan scvt +
>         uxtw instead of ucvtf + zip1 + zip2.
> ---
>  gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c | 6 +++---
>  .../gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c         | 5 ++---
>  2 files changed, 5 insertions(+), 6 deletions(-)
>
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> index 0f96dc2ff00..5edc288ce35 100644
> --- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> @@ -10,6 +10,6 @@ unpack_double_int_plus8 (double *d, int32_t *s, int size)
>      d[i] = s[i] + 8;
>  }
>
> -/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
> +/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +/* { dg-final { scan-assembler-times {\tsxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> index 70465f91eba..ecd72176177 100644
> --- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> @@ -10,6 +10,5 @@ unpack_double_int_plus9 (double *d, uint32_t *s, int size)
>      d[i] = (double) (s[i] + 9);
>  }
>
> -/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
> +/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +/* { dg-final { scan-assembler-times {\tuxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> --
> 2.39.1.388.g2fc9e9ca3c
>


--
BR,
Hongtao

Re: [PATCH 3/3] [aarch64] Adjust testcase to match assembly output after r14-2007.

[Richard Sandiford]

liuhongt <hongtao.liu@intel.com> writes:
> The new assembly looks better than original one, so I adjust those testcases.

The new loops are shorter, but they process only half the amount of data
per iteration.

The problem is that the new vectoriser code generates multiple statements
but only costs one.  I'll post a fix soon.

Thanks,
Richard

> Ok for trunk?
>
> gcc/testsuite/ChangeLog:
>
> 	PR tree-optimization/110371
> 	PR tree-optimization/110018
> 	* gcc.target/aarch64/sve/unpack_fcvt_signed_1.c: Scan scvt +
> 	sxtw instead of scvt + zip1 + zip2.
> 	* gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c: Scan scvt +
> 	uxtw instead of ucvtf + zip1 + zip2.
> ---
>  gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c | 6 +++---
>  .../gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c         | 5 ++---
>  2 files changed, 5 insertions(+), 6 deletions(-)
>
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> index 0f96dc2ff00..5edc288ce35 100644
> --- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_signed_1.c
> @@ -10,6 +10,6 @@ unpack_double_int_plus8 (double *d, int32_t *s, int size)
>      d[i] = s[i] + 8;
>  }
>  
> -/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
> +/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +/* { dg-final { scan-assembler-times {\tsxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> index 70465f91eba..ecd72176177 100644
> --- a/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/unpack_fcvt_unsigned_1.c
> @@ -10,6 +10,5 @@ unpack_double_int_plus9 (double *d, uint32_t *s, int size)
>      d[i] = (double) (s[i] + 9);
>  }
>  
> -/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.s, z[0-9]+\.s, z[0-9]+\.s\n} 1 } } */
> -/* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.s\n} 2 } } */
> +/* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */
> +/* { dg-final { scan-assembler-times {\tuxtw\tz[0-9]+\.d, p[0-7]/m, z[0-9]+\.d\n} 1 } } */

Re: [PATCH 1/3] Use cvt_op to save intermediate type operand instead of "subtle" vec_dest.

[Richard Biener]

On Mon, 26 Jun 2023, liuhongt wrote:

> When there're multiple operands in vec_oprnds0, vec_dest will be
> overwrited to vectype_out, but in multi_step_cvt case, cvt_type is
> expected. It caused an ICE when verify_gimple_in_cfg.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,} and aarch64-linux-gnu.
> Ok for trunk?

OK.

Richard.

> gcc/ChangeLog:
> 
> 	PR tree-optimization/110371
> 	PR tree-optimization/110018
> 	* tree-vect-stmts.cc (vectorizable_conversion): Use cvt_op to
> 	save intermediate type operand instead of "subtle" vec_dest
> 	for case NONE.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* gcc.target/aarch64/pr110371.c: New test.
> ---
>  gcc/testsuite/gcc.target/aarch64/pr110371.c | 20 ++++++++++++++++++++
>  gcc/tree-vect-stmts.cc                      | 14 ++++++++++----
>  2 files changed, 30 insertions(+), 4 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.target/aarch64/pr110371.c
> 
> diff --git a/gcc/testsuite/gcc.target/aarch64/pr110371.c b/gcc/testsuite/gcc.target/aarch64/pr110371.c
> new file mode 100644
> index 00000000000..444e514e04f
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/pr110371.c
> @@ -0,0 +1,20 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O3" } */
> +
> +typedef struct dest
> +{
> +  double m[3][3];
> +} dest;
> +
> +typedef struct src
> +{
> +  int m[3][3];
> +} src;
> +
> +void
> +foo (dest *a, src* s)
> +{
> +  for (int i = 0; i != 3; i++)
> +    for (int j = 0; j != 3; j++)
> +      a->m[i][j] = s->m[i][j];
> +}
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index 85d1f3ae52c..1748555a625 100644
> --- a/gcc/tree-vect-stmts.cc
> +++ b/gcc/tree-vect-stmts.cc
> @@ -5044,7 +5044,7 @@ vectorizable_conversion (vec_info *vinfo,
>  			 gimple **vec_stmt, slp_tree slp_node,
>  			 stmt_vector_for_cost *cost_vec)
>  {
> -  tree vec_dest;
> +  tree vec_dest, cvt_op = NULL_TREE;
>    tree scalar_dest;
>    tree op0, op1 = NULL_TREE;
>    loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
> @@ -5568,6 +5568,13 @@ vectorizable_conversion (vec_info *vinfo,
>      case NONE:
>        vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies,
>  			 op0, &vec_oprnds0);
> +      /* vec_dest is intermediate type operand when multi_step_cvt.  */
> +      if (multi_step_cvt)
> +	{
> +	  cvt_op = vec_dest;
> +	  vec_dest = vec_dsts[0];
> +	}
> +
>        FOR_EACH_VEC_ELT (vec_oprnds0, i, vop0)
>  	{
>  	  /* Arguments are ready, create the new vector stmt.  */
> @@ -5575,12 +5582,11 @@ vectorizable_conversion (vec_info *vinfo,
>  	  if (multi_step_cvt)
>  	    {
>  	      gcc_assert (multi_step_cvt == 1);
> -	      new_stmt = vect_gimple_build (vec_dest, codecvt1, vop0);
> -	      new_temp = make_ssa_name (vec_dest, new_stmt);
> +	      new_stmt = vect_gimple_build (cvt_op, codecvt1, vop0);
> +	      new_temp = make_ssa_name (cvt_op, new_stmt);
>  	      gimple_assign_set_lhs (new_stmt, new_temp);
>  	      vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi);
>  	      vop0 = new_temp;
> -	      vec_dest = vec_dsts[0];
>  	    }
>  	  new_stmt = vect_gimple_build (vec_dest, code1, vop0);
>  	  new_temp = make_ssa_name (vec_dest, new_stmt);
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)