diff options
| author | Thomas Schwinge <tschwinge@baylibre.com> | 2024-02-24 00:29:14 +0100 |
|---|---|---|
| committer | Thomas Schwinge <tschwinge@baylibre.com> | 2024-04-08 22:08:00 +0200 |
| commit | df7625c3af004a81c13d54bb8810e03932eeb59a (patch) | |
| tree | 7853130e3d74a569cd94e21f8059e3fb1d141618 /gcc/doc | |
| parent | 3fa8bff30ab58bd8b8018764d390ec2fcc8153bb (diff) | |
| download | gcc-df7625c3af004a81c13d54bb8810e03932eeb59a.zip gcc-df7625c3af004a81c13d54bb8810e03932eeb59a.tar.gz gcc-df7625c3af004a81c13d54bb8810e03932eeb59a.tar.bz2 | |
GCN: '--param=gcn-preferred-vectorization-factor=[default,32,64]'
..., and specify '--param=gcn-preferred-vectorization-factor=64' for
'gcc.target/gcn/[...]' test cases with 'scan-assembler' directives that
are specific to 64-lane vectors. This resolves regressions introduced
in commit 6dedafe166cc02ae87b6a0699ad61ce3ffc46803
"amdgcn: Prefer V32 on RDNA devices".
gcc/
* config/gcn/gcn.opt (--param=gcn-preferred-vectorization-factor):
New.
* config/gcn/gcn.cc (gcn_vectorize_preferred_simd_mode) Use it.
* doc/invoke.texi (Optimize Options): Document it.
gcc/testsuite/
* gcc.target/gcn/cond_fmaxnm_1.c: Specify
'--param=gcn-preferred-vectorization-factor=64'.
* gcc.target/gcn/cond_fmaxnm_2.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_3.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_4.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_5.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_6.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_7.c: Likewise.
* gcc.target/gcn/cond_fmaxnm_8.c: Likewise.
* gcc.target/gcn/cond_fminnm_1.c: Likewise.
* gcc.target/gcn/cond_fminnm_2.c: Likewise.
* gcc.target/gcn/cond_fminnm_3.c: Likewise.
* gcc.target/gcn/cond_fminnm_4.c: Likewise.
* gcc.target/gcn/cond_fminnm_5.c: Likewise.
* gcc.target/gcn/cond_fminnm_6.c: Likewise.
* gcc.target/gcn/cond_fminnm_7.c: Likewise.
* gcc.target/gcn/cond_fminnm_8.c: Likewise.
* gcc.target/gcn/cond_shift_3.c: Likewise.
* gcc.target/gcn/cond_shift_4.c: Likewise.
* gcc.target/gcn/cond_shift_8.c: Likewise.
* gcc.target/gcn/cond_shift_9.c: Likewise.
* gcc.target/gcn/cond_smax_1.c: Likewise.
* gcc.target/gcn/cond_smin_1.c: Likewise.
* gcc.target/gcn/cond_umax_1.c: Likewise.
* gcc.target/gcn/cond_umin_1.c: Likewise.
* gcc.target/gcn/simd-math-1.c: Likewise.
* gcc.target/gcn/simd-math-5-char.c: Likewise.
* gcc.target/gcn/simd-math-5-long.c: Likewise.
* gcc.target/gcn/simd-math-5-short.c: Likewise.
* gcc.target/gcn/simd-math-5.c: Likewise.
* gcc.target/gcn/smax_1.c: Likewise.
* gcc.target/gcn/smin_1.c: Likewise.
* gcc.target/gcn/umax_1.c: Likewise.
* gcc.target/gcn/umin_1.c: Likewise.
Diffstat (limited to 'gcc/doc')
| -rw-r--r-- | gcc/doc/invoke.texi | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi index f5f5d83..0b1230b 100644 --- a/gcc/doc/invoke.texi +++ b/gcc/doc/invoke.texi @@ -17017,6 +17017,14 @@ loop. The default value is four. @end table +The following choices of @var{name} are available on GCN targets: + +@table @gcctabopt +@item gcn-preferred-vectorization-factor +Preferred vectorization factor: @samp{default}, @samp{32}, @samp{64}. + +@end table + The following choices of @var{name} are available on i386 and x86_64 targets: @table @gcctabopt |