i386: Fix ICEs due to simplify_gen_subreg returning NULL [PR99025]

Hi!

In these patterns, we call simplify_gen_subreg on the input operand
to create paradoxical subregs that have 2x, 4x or 8x elements as the input
operand.  That works fine if the input operand is a REG, but when it is a
SUBREG, RTL doesn't allow SUBREG of SUBREG and so relies on simplify_subreg
actually simplifying it.  And e.g. if the input operand is a SUBREG that
changes the element mode (floating vs. non-floating) and then combined with
a paradoxical subreg (i.e. different size) this can easily fail, then
simplify_gen_subreg returns NULL but we still use it in instructions.

Fixed by forcing the operands into REG.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2021-02-10  Jakub Jelinek  <jakub@redhat.com>

	PR target/99025
	* config/i386/sse.md (fix<fixunssuffix>_truncv2sfv2di2,
	<insn>v8qiv8hi2, <insn>v8qiv8si2, <insn>v4qiv4si2, <insn>v4hiv4si2,
	<insn>v8qiv8di2, <insn>v4qiv4di2, <insn>v2qiv2di2, <insn>v4hiv4di2,
	<insn>v2hiv2di2, <insn>v2siv2di2): Force operands[1] into REG before
	calling simplify_gen_subreg on it.

	* gcc.target/i386/pr99025.c: New test.


	Jakub

On Wed, Feb 10, 2021 at 4:12 PM Jakub Jelinek <jakub@redhat.com> wrote:
>
> Hi!
>
> In these patterns, we call simplify_gen_subreg on the input operand
> to create paradoxical subregs that have 2x, 4x or 8x elements as the input
> operand.  That works fine if the input operand is a REG, but when it is a
> SUBREG, RTL doesn't allow SUBREG of SUBREG and so relies on simplify_subreg
> actually simplifying it.  And e.g. if the input operand is a SUBREG that
> changes the element mode (floating vs. non-floating) and then combined with
> a paradoxical subreg (i.e. different size) this can easily fail, then
> simplify_gen_subreg returns NULL but we still use it in instructions.
>
> Fixed by forcing the operands into REG.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2021-02-10  Jakub Jelinek  <jakub@redhat.com>
>
>         PR target/99025
>         * config/i386/sse.md (fix<fixunssuffix>_truncv2sfv2di2,
>         <insn>v8qiv8hi2, <insn>v8qiv8si2, <insn>v4qiv4si2, <insn>v4hiv4si2,
>         <insn>v8qiv8di2, <insn>v4qiv4di2, <insn>v2qiv2di2, <insn>v4hiv4di2,
>         <insn>v2hiv2di2, <insn>v2siv2di2): Force operands[1] into REG before
>         calling simplify_gen_subreg on it.
>
>         * gcc.target/i386/pr99025.c: New test.

OK.

Thanks,
Uros.

>
> --- gcc/config/i386/sse.md.jj   2021-02-10 07:52:32.673901634 +0100
> +++ gcc/config/i386/sse.md      2021-02-10 10:57:37.229665371 +0100
> @@ -6356,6 +6356,7 @@ (define_expand "fix<fixunssuffix>_truncv
>           (match_operand:V2SF 1 "register_operand")))]
>    "TARGET_AVX512DQ && TARGET_AVX512VL"
>  {
> +  operands[1] = force_reg (V2SFmode, operands[1]);
>    operands[1] = simplify_gen_subreg (V4SFmode, operands[1], V2SFmode, 0);
>    emit_insn (gen_avx512dq_fix<fixunssuffix>_truncv2sfv2di2
>              (operands[0], operands[1]));
> @@ -18013,6 +18014,7 @@ (define_expand "<insn>v8qiv8hi2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V8QImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V16QImode, operands[1], V8QImode, 0);
>        emit_insn (gen_sse4_1_<code>v8qiv8hi2 (operands[0], operands[1]));
>        DONE;
> @@ -18090,6 +18092,7 @@ (define_expand "<insn>v8qiv8si2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V8QImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V16QImode, operands[1], V8QImode, 0);
>        emit_insn (gen_avx2_<code>v8qiv8si2 (operands[0], operands[1]));
>        DONE;
> @@ -18153,6 +18156,7 @@ (define_expand "<insn>v4qiv4si2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V4QImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V16QImode, operands[1], V4QImode, 0);
>        emit_insn (gen_sse4_1_<code>v4qiv4si2 (operands[0], operands[1]));
>        DONE;
> @@ -18279,6 +18283,7 @@ (define_expand "<insn>v4hiv4si2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V4HImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V8HImode, operands[1], V4HImode, 0);
>        emit_insn (gen_sse4_1_<code>v4hiv4si2 (operands[0], operands[1]));
>        DONE;
> @@ -18366,6 +18371,7 @@ (define_expand "<insn>v8qiv8di2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V8QImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V16QImode, operands[1], V8QImode, 0);
>        emit_insn (gen_avx512f_<code>v8qiv8di2 (operands[0], operands[1]));
>        DONE;
> @@ -18427,6 +18433,7 @@ (define_expand "<insn>v4qiv4di2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V8QImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V16QImode, operands[1], V8QImode, 0);
>        emit_insn (gen_avx2_<code>v4qiv4di2 (operands[0], operands[1]));
>        DONE;
> @@ -18453,6 +18460,7 @@ (define_expand "<insn>v2qiv2di2"
>           (match_operand:V2QI 1 "register_operand")))]
>    "TARGET_SSE4_1"
>  {
> +  operands[1] = force_reg (V2QImode, operands[1]);
>    operands[1] = simplify_gen_subreg (V16QImode, operands[1], V2QImode, 0);
>    emit_insn (gen_sse4_1_<code>v2qiv2di2 (operands[0], operands[1]));
>    DONE;
> @@ -18525,6 +18533,7 @@ (define_expand "<insn>v4hiv4di2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V4HImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V8HImode, operands[1], V4HImode, 0);
>        emit_insn (gen_avx2_<code>v4hiv4di2 (operands[0], operands[1]));
>        DONE;
> @@ -18586,6 +18595,7 @@ (define_expand "<insn>v2hiv2di2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V2HImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V8HImode, operands[1], V2HImode, 0);
>        emit_insn (gen_sse4_1_<code>v2hiv2di2 (operands[0], operands[1]));
>        DONE;
> @@ -18737,6 +18747,7 @@ (define_expand "<insn>v2siv2di2"
>  {
>    if (!MEM_P (operands[1]))
>      {
> +      operands[1] = force_reg (V2SImode, operands[1]);
>        operands[1] = simplify_gen_subreg (V4SImode, operands[1], V2SImode, 0);
>        emit_insn (gen_sse4_1_<code>v2siv2di2 (operands[0], operands[1]));
>        DONE;
> --- gcc/testsuite/gcc.target/i386/pr99025.c.jj  2021-02-09 19:17:29.705924814 +0100
> +++ gcc/testsuite/gcc.target/i386/pr99025.c     2021-02-09 19:17:10.687137688 +0100
> @@ -0,0 +1,17 @@
> +/* PR target/99025 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -msse4" } */
> +
> +long v[16];
> +int w;
> +union U { float u; int r; } x;
> +
> +void
> +foo (float y)
> +{
> +  union U z;
> +  x.u = w;
> +  v[5] = x.r;
> +  z.u = y;
> +  v[6] = z.r;
> +}
>
>         Jakub
>