On Fri, Dec 20, 2024 at 3:38 PM Jakub Jelinek <[email protected]> wrote:
>
> On Fri, Dec 20, 2024 at 02:55:51PM +0100, Christoph Müllner wrote:
> > A recent bugfix (eee2891312) for PR117830 also addressed PR118149.
> > This patch adds two test cases for PR118149.
> > These tests are different than other tests in that one of the
> > vec-perm selectors contains indices in descending order (1, 1, 0, 0),
> > which is the root cause for the ICE observed in PR118149.
> >
> > PR 118149
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.dg/tree-ssa/pr118149-2.c: New test.
> > * gcc.dg/tree-ssa/pr118149.c: New test.
> >
> > Signed-off-by: Christoph Müllner <[email protected]>
>
> Have you actually tested it with say
> make check-gcc
> RUNTESTFLAGS="--target_board=unix\{-m64,-m32,-m32/-mno-mmx/-mno-sse}
> tree-ssa.exp='pr118149* vector-11.c'"
> on x86_64? While the -2.c possibly could work, I don't see how the last
> testcase could on i686-linux (i.e. that -m32 -mno-mmx -mno-sse).
> So I think it needs
> /* { dg-additional-options "-msse2" { target { i?86-*-* x86_64-*-* } } } */
> Or so.
No, I haven't tested like this (only with defaults on x86-64 and aarch64).
I've seen that you added these flags to vector-7, vector-8 and
vector-9 recently.
I copied the DG lines from vector-10 which does not blend and therefore does not
need this line. Btw, vector-11 also needs this DG line.
I will do the following
* Add these options to vector-11.c, pr118149.c and pr118149-2.c
* Retest with the mentioned RUNTESTFLAGS
* Send a patch to fix existing vector-11 patch
* Send a v2 of this patch
> Also, why are you using dg-additional-options in tree-ssa/ ? I think the
> default there is just -pedantic-errors which you don't really need for these
> tests, so just dg-options?
I will change accordingly for all vector-* tests and satd-hadamard.c
and include that in the patch that fixes vector-11.c.
Thanks!
>
> > gcc/testsuite/gcc.dg/tree-ssa/pr118149-2.c | 36 ++++++++++++++++++++++
> > gcc/testsuite/gcc.dg/tree-ssa/pr118149.c | 19 ++++++++++++
> > 2 files changed, 55 insertions(+)
> > create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr118149-2.c
> > create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr118149.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr118149-2.c
> > b/gcc/testsuite/gcc.dg/tree-ssa/pr118149-2.c
> > new file mode 100644
> > index 00000000000..b6ceea8fdb7
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr118149-2.c
> > @@ -0,0 +1,36 @@
> > +/* { dg-do compile } */
> > +/* { dg-additional-options "-O3 -fdump-tree-forwprop1-details -Wno-psabi"
> > } */
> > +
> > +typedef int vec __attribute__((vector_size (4 * sizeof (float))));
> > +
> > +void f1 (vec *p_v_in, vec *p_v_out_1, vec *p_v_out_2)
> > +{
> > + vec sel00 = { 1, 1, 3, 3 };
> > + vec sel01 = { 0, 0, 2, 2 };
> > + vec sel10 = { 3, 3, 2, 2 };
> > + vec sel11 = { 1, 1, 0, 0 };
> > + vec sel = { 0, 1, 6, 7 };
> > + vec v_1, v_2, v_x, v_y, v_out_1, v_out_2;
> > + vec v_in = *p_v_in;
> > +
> > + /* First vec perm sequence. */
> > + v_1 = __builtin_shuffle (v_in, v_in, sel00);
> > + v_2 = __builtin_shuffle (v_in, v_in, sel01);
> > + v_x = v_2 - v_1;
> > + v_y = v_1 + v_2;
> > + v_out_1 = __builtin_shuffle (v_y, v_x, sel);
> > +
> > + /* Second vec perm sequence. */
> > + v_1 = __builtin_shuffle (v_in, v_in, sel10);
> > + v_2 = __builtin_shuffle (v_in, v_in, sel11);
> > + v_x = v_2 - v_1;
> > + v_y = v_1 + v_2;
> > + v_out_2 = __builtin_shuffle (v_y, v_x, sel);
> > +
> > + *p_v_out_1 = v_out_1;
> > + *p_v_out_2 = v_out_2;
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "Vec perm simplify sequences have been
> > blended" "forwprop1" { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
> > +/* { dg-final { scan-tree-dump "VEC_PERM_EXPR.*{ 0, 0, 6, 6 }" "forwprop1"
> > { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
> > +/* { dg-final { scan-tree-dump "VEC_PERM_EXPR.*{ 1, 1, 7, 7 }" "forwprop1"
> > { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
> > diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr118149.c
> > b/gcc/testsuite/gcc.dg/tree-ssa/pr118149.c
> > new file mode 100644
> > index 00000000000..a87463dfa07
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr118149.c
> > @@ -0,0 +1,19 @@
> > +/* { dg-do compile } */
> > +/* { dg-additional-options "-O3 -fdump-tree-forwprop4-details -Wno-psabi"
> > } */
> > +
> > +float *fastconv_parse_dst;
> > +
> > +void fastconv_parse ()
> > +{
> > + float r3k = fastconv_parse_dst[1] - fastconv_parse_dst[3],
> > + i0k = fastconv_parse_dst[4] + fastconv_parse_dst[6],
> > + i1k = fastconv_parse_dst[4] - fastconv_parse_dst[6],
> > + i2k = fastconv_parse_dst[5] + fastconv_parse_dst[7];
> > + fastconv_parse_dst[1] = fastconv_parse_dst[0];
> > + fastconv_parse_dst[4] = fastconv_parse_dst[5] = i0k - i2k;
> > + fastconv_parse_dst[6] = fastconv_parse_dst[7] = i1k + r3k;
> > +}
> > +
> > +/* { dg-final { scan-tree-dump "Vec perm simplify sequences have been
> > blended" "forwprop1" { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
> > +/* { dg-final { scan-tree-dump "VEC_PERM_EXPR.*{ 0, 0, 6, 6 }" "forwprop1"
> > { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
> > +/* { dg-final { scan-tree-dump "VEC_PERM_EXPR.*{ 1, 1, 7, 7 }" "forwprop1"
> > { target { aarch64*-*-* i?86-*-* x86_64-*-* } } } } */
>
> Jakub
>
