On Thu, Aug 23, 2018 at 11:09 AM Richard Sandiford
<richard.sandif...@arm.com> wrote:
>
> The "new" VEC_PERM_EXPR handling makes it easy to support bswap
> for variable-length vectors.
>
> Tested on aarch64-linux-gnu (with and without SVE), aarch64_be-elf
> and x86_64-linux-gnu. OK to install?
OK.
Richard.
> Richard
>
>
> 2018-08-23 Richard Sandiford <richard.sandif...@arm.com>
>
> gcc/
> * tree-vect-stmts.c (vectorizable_bswap): Handle variable-length
> vectors.
>
> gcc/testsuite/
> * gcc.target/aarch64/sve/bswap_1.c: New test.
> * gcc.target/aarch64/sve/bswap_2.c: Likewise.
> * gcc.target/aarch64/sve/bswap_3.c: Likewise.
>
> Index: gcc/tree-vect-stmts.c
> ===================================================================
> --- gcc/tree-vect-stmts.c 2018-08-23 09:59:35.245682525 +0100
> +++ gcc/tree-vect-stmts.c 2018-08-23 10:07:30.233601466 +0100
> @@ -2961,13 +2961,10 @@ vectorizable_bswap (stmt_vec_info stmt_i
> vec_info *vinfo = stmt_info->vinfo;
> loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
> unsigned ncopies;
> - unsigned HOST_WIDE_INT nunits, num_bytes;
>
> op = gimple_call_arg (stmt, 0);
> vectype = STMT_VINFO_VECTYPE (stmt_info);
> -
> - if (!TYPE_VECTOR_SUBPARTS (vectype).is_constant (&nunits))
> - return false;
> + poly_uint64 nunits = TYPE_VECTOR_SUBPARTS (vectype);
>
> /* Multiple types in SLP are handled by creating the appropriate number of
> vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
> @@ -2983,11 +2980,11 @@ vectorizable_bswap (stmt_vec_info stmt_i
> if (! char_vectype)
> return false;
>
> - if (!TYPE_VECTOR_SUBPARTS (char_vectype).is_constant (&num_bytes))
> + poly_uint64 num_bytes = TYPE_VECTOR_SUBPARTS (char_vectype);
> + unsigned word_bytes;
> + if (!constant_multiple_p (num_bytes, nunits, &word_bytes))
> return false;
>
> - unsigned word_bytes = num_bytes / nunits;
> -
> /* The encoding uses one stepped pattern for each byte in the word. */
> vec_perm_builder elts (num_bytes, word_bytes, 3);
> for (unsigned i = 0; i < 3; ++i)
> Index: gcc/testsuite/gcc.target/aarch64/sve/bswap_1.c
> ===================================================================
> --- /dev/null 2018-07-26 10:26:13.137955424 +0100
> +++ gcc/testsuite/gcc.target/aarch64/sve/bswap_1.c 2018-08-23
> 10:07:30.233601466 +0100
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -ftree-vectorize" } */
> +
> +#include <stdint.h>
> +
> +void
> +f (uint16_t *a, uint16_t *b)
> +{
> + for (int i = 0; i < 100; ++i)
> + a[i] = __builtin_bswap16 (b[i]);
> +}
> +
> +/* { dg-final { scan-assembler-times {\trevb\tz[0-9]+\.h, p[0-7]/m,
> z[0-9]+\.h\n} 1 { xfail aarch64_big_endian } } } */
> Index: gcc/testsuite/gcc.target/aarch64/sve/bswap_2.c
> ===================================================================
> --- /dev/null 2018-07-26 10:26:13.137955424 +0100
> +++ gcc/testsuite/gcc.target/aarch64/sve/bswap_2.c 2018-08-23
> 10:07:30.233601466 +0100
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -ftree-vectorize" } */
> +
> +#include <stdint.h>
> +
> +void
> +f (uint32_t *a, uint32_t *b)
> +{
> + for (int i = 0; i < 100; ++i)
> + a[i] = __builtin_bswap32 (b[i]);
> +}
> +
> +/* { dg-final { scan-assembler-times {\trevb\tz[0-9]+\.s, p[0-7]/m,
> z[0-9]+\.s\n} 1 { xfail aarch64_big_endian } } } */
> Index: gcc/testsuite/gcc.target/aarch64/sve/bswap_3.c
> ===================================================================
> --- /dev/null 2018-07-26 10:26:13.137955424 +0100
> +++ gcc/testsuite/gcc.target/aarch64/sve/bswap_3.c 2018-08-23
> 10:07:30.233601466 +0100
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -ftree-vectorize" } */
> +
> +#include <stdint.h>
> +
> +void
> +f (uint64_t *a, uint64_t *b)
> +{
> + for (int i = 0; i < 100; ++i)
> + a[i] = __builtin_bswap64 (b[i]);
> +}
> +
> +/* { dg-final { scan-assembler-times {\trevb\tz[0-9]+\.d, p[0-7]/m,
> z[0-9]+\.d\n} 1 { xfail aarch64_big_endian } } } */
