I'd like to ping the following patch.
Peter
On 10/27/21 8:37 PM, Peter Bergner via Gcc-patches wrote:
> PR102976 shows a test case where we generate wrong code when building
> a vector pair from 2 vector registers. The bug here is that with unlucky
> register assignments, we can clobber one of the input operands before
> we write both registers of the output operand. The solution is to use
> early-clobbers in the assemble pair and accumulator patterns.
>
> This passed bootstrap and regtesting with no regressions and our
> OpenBLAS team has confirmed it fixes the issues they reported.
> Ok for mainline?
>
> Ok for GCC 11 too after a few days on trunk?
>
> Peter
>
>
> gcc/
> PR target/102976
> * config/rs6000/mma.md (*vsx_assemble_pair): Add early-clobber for
> output operand.
> (*mma_assemble_acc): Likewise.
>
> gcc/testsuite/
> PR target/102976
> * gcc.target/powerpc/pr102976.c: New test.
>
> diff --git a/gcc/config/rs6000/mma.md b/gcc/config/rs6000/mma.md
> index 1990a2183f6..f0ea99963f7 100644
> --- a/gcc/config/rs6000/mma.md
> +++ b/gcc/config/rs6000/mma.md
> @@ -339,7 +339,7 @@ (define_expand "vsx_assemble_pair"
> })
>
> (define_insn_and_split "*vsx_assemble_pair"
> - [(set (match_operand:OO 0 "vsx_register_operand" "=wa")
> + [(set (match_operand:OO 0 "vsx_register_operand" "=&wa")
> (unspec:OO [(match_operand:V16QI 1 "mma_assemble_input_operand" "mwa")
> (match_operand:V16QI 2 "mma_assemble_input_operand" "mwa")]
> UNSPEC_MMA_ASSEMBLE))]
> @@ -405,7 +405,7 @@ (define_expand "mma_assemble_acc"
> })
>
> (define_insn_and_split "*mma_assemble_acc"
> - [(set (match_operand:XO 0 "fpr_reg_operand" "=d")
> + [(set (match_operand:XO 0 "fpr_reg_operand" "=&d")
> (unspec:XO [(match_operand:V16QI 1 "mma_assemble_input_operand" "mwa")
> (match_operand:V16QI 2 "mma_assemble_input_operand" "mwa")
> (match_operand:V16QI 3 "mma_assemble_input_operand" "mwa")
> diff --git a/gcc/testsuite/gcc.target/powerpc/pr102976.c
> b/gcc/testsuite/gcc.target/powerpc/pr102976.c
> new file mode 100644
> index 00000000000..a8de8f056f1
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/powerpc/pr102976.c
> @@ -0,0 +1,14 @@
> +/* { dg-require-effective-target power10_ok } */
> +/* { dg-options "-O2 -mdejagnu-cpu=power10" } */
> +
> +#include <altivec.h>
> +void
> +bug (__vector_pair *dst)
> +{
> + register vector unsigned char vec0 asm ("vs44");
> + register vector unsigned char vec1 asm ("vs32");
> + __builtin_vsx_build_pair (dst, vec0, vec1);
> +}
> +
> +/* { dg-final { scan-assembler-times {xxlor[^,]*,44,44} 1 } } */
> +/* { dg-final { scan-assembler-times {xxlor[^,]*,32,32} 1 } } */
>
