Fold _mm{, 256, 512}_abs_{epi8, epi16, epi32, epi64} into gimple ABS_EXPR.
Checks
Commit Message
Also for 64-bit vector abs intrinsics _mm_abs_{pi8,pi16,pi32}.
Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}.
Ok for trunk?
gcc/ChangeLog:
PR target/109900
* config/i386/i386.cc (ix86_gimple_fold_builtin): Fold
_mm{,256,512}_abs_{epi8,epi16,epi32,epi64} and
_mm_abs_{pi8,pi16,pi32} into gimple ABS_EXPR.
(ix86_masked_all_ones): Handle 64-bit mask.
* config/i386/i386-builtin.def: Replace icode of related
non-mask simd abs builtins with CODE_FOR_nothing.
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr109900.c: New test.
---
gcc/config/i386/i386-builtin.def | 18 ++---
gcc/config/i386/i386.cc | 86 +++++++++++++++------
gcc/testsuite/gcc.target/i386/pr109900.c | 95 ++++++++++++++++++++++++
3 files changed, 166 insertions(+), 33 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr109900.c
Comments
On Mon, May 22, 2023 at 3:35 PM liuhongt <hongtao.liu@intel.com> wrote:
>
> Also for 64-bit vector abs intrinsics _mm_abs_{pi8,pi16,pi32}.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}.
> Ok for trunk?
Ready push to main trunk.
>
> gcc/ChangeLog:
>
> PR target/109900
> * config/i386/i386.cc (ix86_gimple_fold_builtin): Fold
> _mm{,256,512}_abs_{epi8,epi16,epi32,epi64} and
> _mm_abs_{pi8,pi16,pi32} into gimple ABS_EXPR.
> (ix86_masked_all_ones): Handle 64-bit mask.
> * config/i386/i386-builtin.def: Replace icode of related
> non-mask simd abs builtins with CODE_FOR_nothing.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/i386/pr109900.c: New test.
> ---
> gcc/config/i386/i386-builtin.def | 18 ++---
> gcc/config/i386/i386.cc | 86 +++++++++++++++------
> gcc/testsuite/gcc.target/i386/pr109900.c | 95 ++++++++++++++++++++++++
> 3 files changed, 166 insertions(+), 33 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/i386/pr109900.c
>
> diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def
> index f7b10a6ab1e..c91e3809c75 100644
> --- a/gcc/config/i386/i386-builtin.def
> +++ b/gcc/config/i386/i386-builtin.def
> @@ -899,12 +899,12 @@ BDESC (OPTION_MASK_ISA_SSE3, 0, CODE_FOR_sse3_hsubv4sf3, "__builtin_ia32_hsubps"
> BDESC (OPTION_MASK_ISA_SSE3, 0, CODE_FOR_sse3_hsubv2df3, "__builtin_ia32_hsubpd", IX86_BUILTIN_HSUBPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF)
>
> /* SSSE3 */
> -BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv16qi2, "__builtin_ia32_pabsb128", IX86_BUILTIN_PABSB128, UNKNOWN, (int) V16QI_FTYPE_V16QI)
> -BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv8qi2, "__builtin_ia32_pabsb", IX86_BUILTIN_PABSB, UNKNOWN, (int) V8QI_FTYPE_V8QI)
> -BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv8hi2, "__builtin_ia32_pabsw128", IX86_BUILTIN_PABSW128, UNKNOWN, (int) V8HI_FTYPE_V8HI)
> -BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv4hi2, "__builtin_ia32_pabsw", IX86_BUILTIN_PABSW, UNKNOWN, (int) V4HI_FTYPE_V4HI)
> -BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv4si2, "__builtin_ia32_pabsd128", IX86_BUILTIN_PABSD128, UNKNOWN, (int) V4SI_FTYPE_V4SI)
> -BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv2si2, "__builtin_ia32_pabsd", IX86_BUILTIN_PABSD, UNKNOWN, (int) V2SI_FTYPE_V2SI)
> +BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb128", IX86_BUILTIN_PABSB128, UNKNOWN, (int) V16QI_FTYPE_V16QI)
> +BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb", IX86_BUILTIN_PABSB, UNKNOWN, (int) V8QI_FTYPE_V8QI)
> +BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw128", IX86_BUILTIN_PABSW128, UNKNOWN, (int) V8HI_FTYPE_V8HI)
> +BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw", IX86_BUILTIN_PABSW, UNKNOWN, (int) V4HI_FTYPE_V4HI)
> +BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd128", IX86_BUILTIN_PABSD128, UNKNOWN, (int) V4SI_FTYPE_V4SI)
> +BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd", IX86_BUILTIN_PABSD, UNKNOWN, (int) V2SI_FTYPE_V2SI)
>
> BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_ssse3_phaddwv8hi3, "__builtin_ia32_phaddw128", IX86_BUILTIN_PHADDW128, UNKNOWN, (int) V8HI_FTYPE_V8HI_V8HI)
> BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_phaddwv4hi3, "__builtin_ia32_phaddw", IX86_BUILTIN_PHADDW, UNKNOWN, (int) V4HI_FTYPE_V4HI_V4HI)
> @@ -1178,9 +1178,9 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_vec_pack_sfix_v4df, "__builtin_ia32_vec_
>
> /* AVX2 */
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_mpsadbw, "__builtin_ia32_mpsadbw256", IX86_BUILTIN_MPSADBW256, UNKNOWN, (int) V32QI_FTYPE_V32QI_V32QI_INT)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv32qi2, "__builtin_ia32_pabsb256", IX86_BUILTIN_PABSB256, UNKNOWN, (int) V32QI_FTYPE_V32QI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv16hi2, "__builtin_ia32_pabsw256", IX86_BUILTIN_PABSW256, UNKNOWN, (int) V16HI_FTYPE_V16HI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv8si2, "__builtin_ia32_pabsd256", IX86_BUILTIN_PABSD256, UNKNOWN, (int) V8SI_FTYPE_V8SI)
> +BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb256", IX86_BUILTIN_PABSB256, UNKNOWN, (int) V32QI_FTYPE_V32QI)
> +BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw256", IX86_BUILTIN_PABSW256, UNKNOWN, (int) V16HI_FTYPE_V16HI)
> +BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd256", IX86_BUILTIN_PABSD256, UNKNOWN, (int) V8SI_FTYPE_V8SI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packssdw, "__builtin_ia32_packssdw256", IX86_BUILTIN_PACKSSDW256, UNKNOWN, (int) V16HI_FTYPE_V8SI_V8SI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packsswb, "__builtin_ia32_packsswb256", IX86_BUILTIN_PACKSSWB256, UNKNOWN, (int) V32QI_FTYPE_V16HI_V16HI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packusdw, "__builtin_ia32_packusdw256", IX86_BUILTIN_PACKUSDW256, UNKNOWN, (int) V16HI_FTYPE_V8SI_V8SI)
> diff --git a/gcc/config/i386/i386.cc b/gcc/config/i386/i386.cc
> index 6a4b3326219..7d1d1bb0760 100644
> --- a/gcc/config/i386/i386.cc
> +++ b/gcc/config/i386/i386.cc
> @@ -17928,6 +17928,8 @@ ix86_masked_all_ones (unsigned HOST_WIDE_INT elems, tree arg_mask)
> return false;
>
> unsigned HOST_WIDE_INT mask = TREE_INT_CST_LOW (arg_mask);
> + if (elems == HOST_BITS_PER_WIDE_INT)
> + return mask == HOST_WIDE_INT_M1U;
> if ((mask | (HOST_WIDE_INT_M1U << elems)) != HOST_WIDE_INT_M1U)
> return false;
>
> @@ -18407,7 +18409,7 @@ ix86_fold_builtin (tree fndecl, int n_args,
> bool
> ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> {
> - gimple *stmt = gsi_stmt (*gsi);
> + gimple *stmt = gsi_stmt (*gsi), *g;
> tree fndecl = gimple_call_fndecl (stmt);
> gcc_checking_assert (fndecl && fndecl_built_in_p (fndecl, BUILT_IN_MD));
> int n_args = gimple_call_num_args (stmt);
> @@ -18420,6 +18422,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> unsigned HOST_WIDE_INT count;
> bool is_vshift;
> unsigned HOST_WIDE_INT elems;
> + location_t loc;
>
> /* Don't fold when there's isa mismatch. */
> if (!ix86_check_builtin_isa_match (fn_code, NULL, NULL))
> @@ -18455,8 +18458,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> if (!expr_not_equal_to (arg0, wi::zero (prec)))
> return false;
>
> - location_t loc = gimple_location (stmt);
> - gimple *g = gimple_build_call (decl, 1, arg0);
> + loc = gimple_location (stmt);
> + g = gimple_build_call (decl, 1, arg0);
> gimple_set_location (g, loc);
> tree lhs = make_ssa_name (integer_type_node);
> gimple_call_set_lhs (g, lhs);
> @@ -18478,8 +18481,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> arg0 = gimple_call_arg (stmt, 0);
> if (idx < TYPE_PRECISION (TREE_TYPE (arg0)))
> break;
> - location_t loc = gimple_location (stmt);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> + loc = gimple_location (stmt);
> + g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> return true;
> @@ -18494,9 +18497,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> arg1 = gimple_call_arg (stmt, 1);
> if (integer_all_onesp (arg1) && gimple_call_lhs (stmt))
> {
> - location_t loc = gimple_location (stmt);
> + loc = gimple_location (stmt);
> arg0 = gimple_call_arg (stmt, 0);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> + g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> return true;
> @@ -18527,7 +18530,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> arg2 = gimple_call_arg (stmt, 2);
> if (gimple_call_lhs (stmt))
> {
> - location_t loc = gimple_location (stmt);
> + loc = gimple_location (stmt);
> tree type = TREE_TYPE (arg2);
> gimple_seq stmts = NULL;
> if (VECTOR_FLOAT_TYPE_P (type))
> @@ -18541,9 +18544,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> tree cmp_type = truth_type_for (type);
> tree cmp = gimple_build (&stmts, LT_EXPR, cmp_type, arg2, zero_vec);
> gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
> - VEC_COND_EXPR, cmp,
> - arg1, arg0);
> + g = gimple_build_assign (gimple_call_lhs (stmt),
> + VEC_COND_EXPR, cmp,
> + arg1, arg0);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> }
> @@ -18579,7 +18582,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> arg1 = gimple_call_arg (stmt, 1);
> if (gimple_call_lhs (stmt))
> {
> - location_t loc = gimple_location (stmt);
> + loc = gimple_location (stmt);
> tree type = TREE_TYPE (arg0);
> tree zero_vec = build_zero_cst (type);
> tree minus_one_vec = build_minus_one_cst (type);
> @@ -18587,9 +18590,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> gimple_seq stmts = NULL;
> tree cmp = gimple_build (&stmts, tcode, cmp_type, arg0, arg1);
> gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
> - gimple* g = gimple_build_assign (gimple_call_lhs (stmt),
> - VEC_COND_EXPR, cmp,
> - minus_one_vec, zero_vec);
> + g = gimple_build_assign (gimple_call_lhs (stmt),
> + VEC_COND_EXPR, cmp,
> + minus_one_vec, zero_vec);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> }
> @@ -18794,8 +18797,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> if (count == 0)
> {
> /* Just return the first argument for shift by 0. */
> - location_t loc = gimple_location (stmt);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> + loc = gimple_location (stmt);
> + g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> return true;
> @@ -18805,9 +18808,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> {
> /* For shift counts equal or greater than precision, except for
> arithmetic right shift the result is zero. */
> - location_t loc = gimple_location (stmt);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
> - build_zero_cst (TREE_TYPE (arg0)));
> + loc = gimple_location (stmt);
> + g = gimple_build_assign (gimple_call_lhs (stmt),
> + build_zero_cst (TREE_TYPE (arg0)));
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> return true;
> @@ -18836,7 +18839,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> return false;
>
> machine_mode imode = GET_MODE_INNER (TYPE_MODE (TREE_TYPE (arg0)));
> - location_t loc = gimple_location (stmt);
> + loc = gimple_location (stmt);
> tree itype = (imode == E_DFmode
> ? long_long_integer_type_node : integer_type_node);
> tree vtype = build_vector_type (itype, elems);
> @@ -18867,9 +18870,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
>
> tree perm_mask = elts.build ();
> arg1 = gimple_call_arg (stmt, 1);
> - gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
> - VEC_PERM_EXPR,
> - arg0, arg1, perm_mask);
> + g = gimple_build_assign (gimple_call_lhs (stmt),
> + VEC_PERM_EXPR,
> + arg0, arg1, perm_mask);
> gimple_set_location (g, loc);
> gsi_replace (gsi, g, false);
> return true;
> @@ -18877,6 +18880,41 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> // Do not error yet, the constant could be propagated later?
> break;
>
> + case IX86_BUILTIN_PABSB:
> + case IX86_BUILTIN_PABSB128:
> + case IX86_BUILTIN_PABSB256:
> + case IX86_BUILTIN_PABSB512:
> + case IX86_BUILTIN_PABSW:
> + case IX86_BUILTIN_PABSW128:
> + case IX86_BUILTIN_PABSW256:
> + case IX86_BUILTIN_PABSW512:
> + case IX86_BUILTIN_PABSD:
> + case IX86_BUILTIN_PABSD128:
> + case IX86_BUILTIN_PABSD256:
> + case IX86_BUILTIN_PABSD512:
> + case IX86_BUILTIN_PABSQ128:
> + case IX86_BUILTIN_PABSQ256:
> + case IX86_BUILTIN_PABSQ512:
> + case IX86_BUILTIN_PABSB128_MASK:
> + case IX86_BUILTIN_PABSB256_MASK:
> + case IX86_BUILTIN_PABSW128_MASK:
> + case IX86_BUILTIN_PABSW256_MASK:
> + case IX86_BUILTIN_PABSD128_MASK:
> + case IX86_BUILTIN_PABSD256_MASK:
> + gcc_assert (n_args >= 1);
> + if (!gimple_call_lhs (stmt))
> + break;
> + arg0 = gimple_call_arg (stmt, 0);
> + elems = TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0));
> + /* For masked ABS, only optimize if the mask is all ones. */
> + if (n_args > 1
> + && !ix86_masked_all_ones (elems, gimple_call_arg (stmt, n_args - 1)))
> + break;
> + loc = gimple_location (stmt);
> + g = gimple_build_assign (gimple_call_lhs (stmt), ABS_EXPR, arg0);
> + gsi_replace (gsi, g, false);
> + return true;
> +
> default:
> break;
> }
> diff --git a/gcc/testsuite/gcc.target/i386/pr109900.c b/gcc/testsuite/gcc.target/i386/pr109900.c
> new file mode 100644
> index 00000000000..f87e8cc11de
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr109900.c
> @@ -0,0 +1,95 @@
> +#include <immintrin.h>
> +/* { dg-do compile } */
> +/* { dg-options "-mavx512bw -O2 -mavx512vl -fdump-tree-optimized" } */
> +/* { dg-final { scan-tree-dump-not "builtin_ia32_pabs" "optimized" } } */
> +
> +
> +__m64
> +absb_64 ()
> +{
> + return _mm_abs_pi8(_mm_set1_pi8 (-1));
> +}
> +
> +__m128i
> +absb_128 ()
> +{
> + return _mm_abs_epi8(_mm_set1_epi8 (-1));
> +}
> +
> +__m256i
> +absb_256 ()
> +{
> + return _mm256_abs_epi8(_mm256_set1_epi8 (-1));
> +}
> +
> +__m512i
> +absb_512 ()
> +{
> + return _mm512_abs_epi8(_mm512_set1_epi8 (-1));
> +}
> +
> +__m64
> +absw_64 ()
> +{
> + return _mm_abs_pi16(_mm_set1_pi16 (-1));
> +}
> +
> +__m128i
> +absw_128 ()
> +{
> + return _mm_abs_epi16(_mm_set1_epi16 (-1));
> +}
> +
> +__m256i
> +absw_256 ()
> +{
> + return _mm256_abs_epi16(_mm256_set1_epi16 (-1));
> +}
> +
> +__m512i
> +absw_512 ()
> +{
> + return _mm512_abs_epi16(_mm512_set1_epi16 (-1));
> +}
> +
> +__m64
> +absd_64 ()
> +{
> + return _mm_abs_pi32(_mm_set1_pi32 (-1));
> +}
> +
> +__m128i
> +absd_128 ()
> +{
> + return _mm_abs_epi32(_mm_set1_epi32 (-1));
> +}
> +
> +__m256i
> +absd_256 ()
> +{
> + return _mm256_abs_epi32(_mm256_set1_epi32 (-1));
> +}
> +
> +__m512i
> +absd_512 ()
> +{
> + return _mm512_abs_epi32(_mm512_set1_epi32 (-1));
> +}
> +
> +__m128i
> +absq_128 ()
> +{
> + return _mm_abs_epi64(_mm_set1_epi64x (-1));
> +}
> +
> +__m256i
> +absq_256 ()
> +{
> + return _mm256_abs_epi64(_mm256_set1_epi64x (-1));
> +}
> +
> +__m512i
> +absq_512 ()
> +{
> + return _mm512_abs_epi64(_mm512_set1_epi64 (-1));
> +}
> --
> 2.39.1.388.g2fc9e9ca3c
>
@@ -899,12 +899,12 @@ BDESC (OPTION_MASK_ISA_SSE3, 0, CODE_FOR_sse3_hsubv4sf3, "__builtin_ia32_hsubps"
BDESC (OPTION_MASK_ISA_SSE3, 0, CODE_FOR_sse3_hsubv2df3, "__builtin_ia32_hsubpd", IX86_BUILTIN_HSUBPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF)
/* SSSE3 */
-BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv16qi2, "__builtin_ia32_pabsb128", IX86_BUILTIN_PABSB128, UNKNOWN, (int) V16QI_FTYPE_V16QI)
-BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv8qi2, "__builtin_ia32_pabsb", IX86_BUILTIN_PABSB, UNKNOWN, (int) V8QI_FTYPE_V8QI)
-BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv8hi2, "__builtin_ia32_pabsw128", IX86_BUILTIN_PABSW128, UNKNOWN, (int) V8HI_FTYPE_V8HI)
-BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv4hi2, "__builtin_ia32_pabsw", IX86_BUILTIN_PABSW, UNKNOWN, (int) V4HI_FTYPE_V4HI)
-BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_absv4si2, "__builtin_ia32_pabsd128", IX86_BUILTIN_PABSD128, UNKNOWN, (int) V4SI_FTYPE_V4SI)
-BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_absv2si2, "__builtin_ia32_pabsd", IX86_BUILTIN_PABSD, UNKNOWN, (int) V2SI_FTYPE_V2SI)
+BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb128", IX86_BUILTIN_PABSB128, UNKNOWN, (int) V16QI_FTYPE_V16QI)
+BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb", IX86_BUILTIN_PABSB, UNKNOWN, (int) V8QI_FTYPE_V8QI)
+BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw128", IX86_BUILTIN_PABSW128, UNKNOWN, (int) V8HI_FTYPE_V8HI)
+BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw", IX86_BUILTIN_PABSW, UNKNOWN, (int) V4HI_FTYPE_V4HI)
+BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd128", IX86_BUILTIN_PABSD128, UNKNOWN, (int) V4SI_FTYPE_V4SI)
+BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd", IX86_BUILTIN_PABSD, UNKNOWN, (int) V2SI_FTYPE_V2SI)
BDESC (OPTION_MASK_ISA_SSSE3, 0, CODE_FOR_ssse3_phaddwv8hi3, "__builtin_ia32_phaddw128", IX86_BUILTIN_PHADDW128, UNKNOWN, (int) V8HI_FTYPE_V8HI_V8HI)
BDESC (OPTION_MASK_ISA_SSSE3 | OPTION_MASK_ISA_MMX, 0, CODE_FOR_ssse3_phaddwv4hi3, "__builtin_ia32_phaddw", IX86_BUILTIN_PHADDW, UNKNOWN, (int) V4HI_FTYPE_V4HI_V4HI)
@@ -1178,9 +1178,9 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_vec_pack_sfix_v4df, "__builtin_ia32_vec_
/* AVX2 */
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_mpsadbw, "__builtin_ia32_mpsadbw256", IX86_BUILTIN_MPSADBW256, UNKNOWN, (int) V32QI_FTYPE_V32QI_V32QI_INT)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv32qi2, "__builtin_ia32_pabsb256", IX86_BUILTIN_PABSB256, UNKNOWN, (int) V32QI_FTYPE_V32QI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv16hi2, "__builtin_ia32_pabsw256", IX86_BUILTIN_PABSW256, UNKNOWN, (int) V16HI_FTYPE_V16HI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_absv8si2, "__builtin_ia32_pabsd256", IX86_BUILTIN_PABSD256, UNKNOWN, (int) V8SI_FTYPE_V8SI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsb256", IX86_BUILTIN_PABSB256, UNKNOWN, (int) V32QI_FTYPE_V32QI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsw256", IX86_BUILTIN_PABSW256, UNKNOWN, (int) V16HI_FTYPE_V16HI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_nothing, "__builtin_ia32_pabsd256", IX86_BUILTIN_PABSD256, UNKNOWN, (int) V8SI_FTYPE_V8SI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packssdw, "__builtin_ia32_packssdw256", IX86_BUILTIN_PACKSSDW256, UNKNOWN, (int) V16HI_FTYPE_V8SI_V8SI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packsswb, "__builtin_ia32_packsswb256", IX86_BUILTIN_PACKSSWB256, UNKNOWN, (int) V32QI_FTYPE_V16HI_V16HI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_packusdw, "__builtin_ia32_packusdw256", IX86_BUILTIN_PACKUSDW256, UNKNOWN, (int) V16HI_FTYPE_V8SI_V8SI)
@@ -17928,6 +17928,8 @@ ix86_masked_all_ones (unsigned HOST_WIDE_INT elems, tree arg_mask)
return false;
unsigned HOST_WIDE_INT mask = TREE_INT_CST_LOW (arg_mask);
+ if (elems == HOST_BITS_PER_WIDE_INT)
+ return mask == HOST_WIDE_INT_M1U;
if ((mask | (HOST_WIDE_INT_M1U << elems)) != HOST_WIDE_INT_M1U)
return false;
@@ -18407,7 +18409,7 @@ ix86_fold_builtin (tree fndecl, int n_args,
bool
ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
{
- gimple *stmt = gsi_stmt (*gsi);
+ gimple *stmt = gsi_stmt (*gsi), *g;
tree fndecl = gimple_call_fndecl (stmt);
gcc_checking_assert (fndecl && fndecl_built_in_p (fndecl, BUILT_IN_MD));
int n_args = gimple_call_num_args (stmt);
@@ -18420,6 +18422,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
unsigned HOST_WIDE_INT count;
bool is_vshift;
unsigned HOST_WIDE_INT elems;
+ location_t loc;
/* Don't fold when there's isa mismatch. */
if (!ix86_check_builtin_isa_match (fn_code, NULL, NULL))
@@ -18455,8 +18458,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
if (!expr_not_equal_to (arg0, wi::zero (prec)))
return false;
- location_t loc = gimple_location (stmt);
- gimple *g = gimple_build_call (decl, 1, arg0);
+ loc = gimple_location (stmt);
+ g = gimple_build_call (decl, 1, arg0);
gimple_set_location (g, loc);
tree lhs = make_ssa_name (integer_type_node);
gimple_call_set_lhs (g, lhs);
@@ -18478,8 +18481,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
arg0 = gimple_call_arg (stmt, 0);
if (idx < TYPE_PRECISION (TREE_TYPE (arg0)))
break;
- location_t loc = gimple_location (stmt);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
+ loc = gimple_location (stmt);
+ g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
return true;
@@ -18494,9 +18497,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
arg1 = gimple_call_arg (stmt, 1);
if (integer_all_onesp (arg1) && gimple_call_lhs (stmt))
{
- location_t loc = gimple_location (stmt);
+ loc = gimple_location (stmt);
arg0 = gimple_call_arg (stmt, 0);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
+ g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
return true;
@@ -18527,7 +18530,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
arg2 = gimple_call_arg (stmt, 2);
if (gimple_call_lhs (stmt))
{
- location_t loc = gimple_location (stmt);
+ loc = gimple_location (stmt);
tree type = TREE_TYPE (arg2);
gimple_seq stmts = NULL;
if (VECTOR_FLOAT_TYPE_P (type))
@@ -18541,9 +18544,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
tree cmp_type = truth_type_for (type);
tree cmp = gimple_build (&stmts, LT_EXPR, cmp_type, arg2, zero_vec);
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
- VEC_COND_EXPR, cmp,
- arg1, arg0);
+ g = gimple_build_assign (gimple_call_lhs (stmt),
+ VEC_COND_EXPR, cmp,
+ arg1, arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
}
@@ -18579,7 +18582,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
arg1 = gimple_call_arg (stmt, 1);
if (gimple_call_lhs (stmt))
{
- location_t loc = gimple_location (stmt);
+ loc = gimple_location (stmt);
tree type = TREE_TYPE (arg0);
tree zero_vec = build_zero_cst (type);
tree minus_one_vec = build_minus_one_cst (type);
@@ -18587,9 +18590,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
gimple_seq stmts = NULL;
tree cmp = gimple_build (&stmts, tcode, cmp_type, arg0, arg1);
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
- gimple* g = gimple_build_assign (gimple_call_lhs (stmt),
- VEC_COND_EXPR, cmp,
- minus_one_vec, zero_vec);
+ g = gimple_build_assign (gimple_call_lhs (stmt),
+ VEC_COND_EXPR, cmp,
+ minus_one_vec, zero_vec);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
}
@@ -18794,8 +18797,8 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
if (count == 0)
{
/* Just return the first argument for shift by 0. */
- location_t loc = gimple_location (stmt);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
+ loc = gimple_location (stmt);
+ g = gimple_build_assign (gimple_call_lhs (stmt), arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
return true;
@@ -18805,9 +18808,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
{
/* For shift counts equal or greater than precision, except for
arithmetic right shift the result is zero. */
- location_t loc = gimple_location (stmt);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
- build_zero_cst (TREE_TYPE (arg0)));
+ loc = gimple_location (stmt);
+ g = gimple_build_assign (gimple_call_lhs (stmt),
+ build_zero_cst (TREE_TYPE (arg0)));
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
return true;
@@ -18836,7 +18839,7 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
return false;
machine_mode imode = GET_MODE_INNER (TYPE_MODE (TREE_TYPE (arg0)));
- location_t loc = gimple_location (stmt);
+ loc = gimple_location (stmt);
tree itype = (imode == E_DFmode
? long_long_integer_type_node : integer_type_node);
tree vtype = build_vector_type (itype, elems);
@@ -18867,9 +18870,9 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
tree perm_mask = elts.build ();
arg1 = gimple_call_arg (stmt, 1);
- gimple *g = gimple_build_assign (gimple_call_lhs (stmt),
- VEC_PERM_EXPR,
- arg0, arg1, perm_mask);
+ g = gimple_build_assign (gimple_call_lhs (stmt),
+ VEC_PERM_EXPR,
+ arg0, arg1, perm_mask);
gimple_set_location (g, loc);
gsi_replace (gsi, g, false);
return true;
@@ -18877,6 +18880,41 @@ ix86_gimple_fold_builtin (gimple_stmt_iterator *gsi)
// Do not error yet, the constant could be propagated later?
break;
+ case IX86_BUILTIN_PABSB:
+ case IX86_BUILTIN_PABSB128:
+ case IX86_BUILTIN_PABSB256:
+ case IX86_BUILTIN_PABSB512:
+ case IX86_BUILTIN_PABSW:
+ case IX86_BUILTIN_PABSW128:
+ case IX86_BUILTIN_PABSW256:
+ case IX86_BUILTIN_PABSW512:
+ case IX86_BUILTIN_PABSD:
+ case IX86_BUILTIN_PABSD128:
+ case IX86_BUILTIN_PABSD256:
+ case IX86_BUILTIN_PABSD512:
+ case IX86_BUILTIN_PABSQ128:
+ case IX86_BUILTIN_PABSQ256:
+ case IX86_BUILTIN_PABSQ512:
+ case IX86_BUILTIN_PABSB128_MASK:
+ case IX86_BUILTIN_PABSB256_MASK:
+ case IX86_BUILTIN_PABSW128_MASK:
+ case IX86_BUILTIN_PABSW256_MASK:
+ case IX86_BUILTIN_PABSD128_MASK:
+ case IX86_BUILTIN_PABSD256_MASK:
+ gcc_assert (n_args >= 1);
+ if (!gimple_call_lhs (stmt))
+ break;
+ arg0 = gimple_call_arg (stmt, 0);
+ elems = TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0));
+ /* For masked ABS, only optimize if the mask is all ones. */
+ if (n_args > 1
+ && !ix86_masked_all_ones (elems, gimple_call_arg (stmt, n_args - 1)))
+ break;
+ loc = gimple_location (stmt);
+ g = gimple_build_assign (gimple_call_lhs (stmt), ABS_EXPR, arg0);
+ gsi_replace (gsi, g, false);
+ return true;
+
default:
break;
}
new file mode 100644
@@ -0,0 +1,95 @@
+#include <immintrin.h>
+/* { dg-do compile } */
+/* { dg-options "-mavx512bw -O2 -mavx512vl -fdump-tree-optimized" } */
+/* { dg-final { scan-tree-dump-not "builtin_ia32_pabs" "optimized" } } */
+
+
+__m64
+absb_64 ()
+{
+ return _mm_abs_pi8(_mm_set1_pi8 (-1));
+}
+
+__m128i
+absb_128 ()
+{
+ return _mm_abs_epi8(_mm_set1_epi8 (-1));
+}
+
+__m256i
+absb_256 ()
+{
+ return _mm256_abs_epi8(_mm256_set1_epi8 (-1));
+}
+
+__m512i
+absb_512 ()
+{
+ return _mm512_abs_epi8(_mm512_set1_epi8 (-1));
+}
+
+__m64
+absw_64 ()
+{
+ return _mm_abs_pi16(_mm_set1_pi16 (-1));
+}
+
+__m128i
+absw_128 ()
+{
+ return _mm_abs_epi16(_mm_set1_epi16 (-1));
+}
+
+__m256i
+absw_256 ()
+{
+ return _mm256_abs_epi16(_mm256_set1_epi16 (-1));
+}
+
+__m512i
+absw_512 ()
+{
+ return _mm512_abs_epi16(_mm512_set1_epi16 (-1));
+}
+
+__m64
+absd_64 ()
+{
+ return _mm_abs_pi32(_mm_set1_pi32 (-1));
+}
+
+__m128i
+absd_128 ()
+{
+ return _mm_abs_epi32(_mm_set1_epi32 (-1));
+}
+
+__m256i
+absd_256 ()
+{
+ return _mm256_abs_epi32(_mm256_set1_epi32 (-1));
+}
+
+__m512i
+absd_512 ()
+{
+ return _mm512_abs_epi32(_mm512_set1_epi32 (-1));
+}
+
+__m128i
+absq_128 ()
+{
+ return _mm_abs_epi64(_mm_set1_epi64x (-1));
+}
+
+__m256i
+absq_256 ()
+{
+ return _mm256_abs_epi64(_mm256_set1_epi64x (-1));
+}
+
+__m512i
+absq_512 ()
+{
+ return _mm512_abs_epi64(_mm512_set1_epi64 (-1));
+}