This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [AArch64] Tighten predicates on SIMD shift intrinsics
- From: James Greenhalgh <james dot greenhalgh at arm dot com>
- To: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Cc: Marcus Shawcroft <Marcus dot Shawcroft at arm dot com>
- Date: Fri, 19 Sep 2014 11:59:23 +0100
- Subject: Re: [AArch64] Tighten predicates on SIMD shift intrinsics
- Authentication-results: sourceware.org; auth=none
- References: <1410424192-5243-1-git-send-email-james dot greenhalgh at arm dot com>
*Ping*
Thanks,
James
On Thu, Sep 11, 2014 at 09:29:52AM +0100, James Greenhalgh wrote:
> gcc/
>
> 2014-09-11 James Greenhalgh <james.greenhalgh@arm.com>
>
> * config/aarch64/aarch64-protos.h (aarch64_simd_const_bounds): Change
> return type to bool.
> * config/aarch64/aarch64-simd.md (aarch64_<sur>q<r>shl<mode>): Use
> new predicates.
> (aarch64_<sur>shll2_n<mode>): Likewise.
> (aarch64_<sur>shr_n<mode>): Likewise.
> (aarch64_<sur>sra_n<mode>: Likewise.
> (aarch64_<sur>s<lr>i_n<mode>): Likewise.
> (aarch64_<sur>qshl<u>_n<mode>): Likewise.
> * config/aarch64/aarch64.c (aarch64_simd_const_bounds): Change
> return type to bool; don't print errors.
> * config/aarch64/iterators.md (ve_mode): New.
> (offsetlr): Remap to infix text for use in new predicates.
> * config/aarch64/predicates.md (aarch64_simd_shift_imm_qi): New.
> (aarch64_simd_shift_imm_hi): Likewise.
> (aarch64_simd_shift_imm_si): Likewise.
> (aarch64_simd_shift_imm_di): Likewise.
> (aarch64_simd_shift_imm_offset_qi): Likewise.
> (aarch64_simd_shift_imm_offset_hi): Likewise.
> (aarch64_simd_shift_imm_offset_si): Likewise.
> (aarch64_simd_shift_imm_offset_di): Likewise.
> (aarch64_simd_shift_imm_bitsize_qi): Likewise.
> (aarch64_simd_shift_imm_bitsize_hi): Likewise.
> (aarch64_simd_shift_imm_bitsize_si): Likewise.
> (aarch64_simd_shift_imm_bitsize_di): Likewise.
>
> gcc/testsuite/
>
> 2014-09-08 James Greenhalgh <james.greenhalgh@arm.com>
>
> * gcc.target/aarch64/simd/vqshlb_1.c: New.
>
> diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
> index 35f89ff..9de7af7 100644
> --- a/gcc/config/aarch64/aarch64-protos.h
> +++ b/gcc/config/aarch64/aarch64-protos.h
> @@ -205,6 +205,7 @@ bool aarch64_regno_ok_for_base_p (int, bool);
> bool aarch64_regno_ok_for_index_p (int, bool);
> bool aarch64_simd_check_vect_par_cnst_half (rtx op, enum machine_mode mode,
> bool high);
> +bool aarch64_simd_const_bounds (rtx, HOST_WIDE_INT, HOST_WIDE_INT);
> bool aarch64_simd_imm_scalar_p (rtx x, enum machine_mode mode);
> bool aarch64_simd_imm_zero_p (rtx, enum machine_mode);
> bool aarch64_simd_scalar_immediate_valid_for_move (rtx, enum machine_mode);
> @@ -255,7 +256,6 @@ void aarch64_emit_call_insn (rtx);
> /* Initialize builtins for SIMD intrinsics. */
> void init_aarch64_simd_builtins (void);
>
> -void aarch64_simd_const_bounds (rtx, HOST_WIDE_INT, HOST_WIDE_INT);
> void aarch64_simd_disambiguate_copy (rtx *, rtx *, rtx *, unsigned int);
>
> /* Emit code to place a AdvSIMD pair result in memory locations (with equal
> diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
> index 6a45e91512ffe1c8c2ecd2b1ba4336baf87f7256..9e688e310027c772cfe5ecd4a158796b143998c5 100644
> --- a/gcc/config/aarch64/aarch64-simd.md
> +++ b/gcc/config/aarch64/aarch64-simd.md
> @@ -3715,12 +3715,12 @@ (define_insn "aarch64_<sur>q<r>shl<mode>
> (define_insn "aarch64_<sur>shll_n<mode>"
> [(set (match_operand:<VWIDE> 0 "register_operand" "=w")
> (unspec:<VWIDE> [(match_operand:VDW 1 "register_operand" "w")
> - (match_operand:SI 2 "immediate_operand" "i")]
> + (match_operand:SI 2
> + "aarch64_simd_shift_imm_bitsize_<ve_mode>" "i")]
> VSHLL))]
> "TARGET_SIMD"
> "*
> int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[2], 0, bit_width + 1);
> if (INTVAL (operands[2]) == bit_width)
> {
> return \"shll\\t%0.<Vwtype>, %1.<Vtype>, %2\";
> @@ -3741,7 +3741,6 @@ (define_insn "aarch64_<sur>shll2_n<mode>
> "TARGET_SIMD"
> "*
> int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[2], 0, bit_width + 1);
> if (INTVAL (operands[2]) == bit_width)
> {
> return \"shll2\\t%0.<Vwtype>, %1.<Vtype>, %2\";
> @@ -3757,13 +3756,11 @@ (define_insn "aarch64_<sur>shll2_n<mode>
> (define_insn "aarch64_<sur>shr_n<mode>"
> [(set (match_operand:VSDQ_I_DI 0 "register_operand" "=w")
> (unspec:VSDQ_I_DI [(match_operand:VSDQ_I_DI 1 "register_operand" "w")
> - (match_operand:SI 2 "immediate_operand" "i")]
> + (match_operand:SI 2
> + "aarch64_simd_shift_imm_offset_<ve_mode>" "i")]
> VRSHR_N))]
> "TARGET_SIMD"
> - "*
> - int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[2], 1, bit_width + 1);
> - return \"<sur>shr\\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2\";"
> + "<sur>shr\\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2"
> [(set_attr "type" "neon_sat_shift_imm<q>")]
> )
>
> @@ -3773,13 +3770,11 @@ (define_insn "aarch64_<sur>sra_n<mode>"
> [(set (match_operand:VSDQ_I_DI 0 "register_operand" "=w")
> (unspec:VSDQ_I_DI [(match_operand:VSDQ_I_DI 1 "register_operand" "0")
> (match_operand:VSDQ_I_DI 2 "register_operand" "w")
> - (match_operand:SI 3 "immediate_operand" "i")]
> + (match_operand:SI 3
> + "aarch64_simd_shift_imm_offset_<ve_mode>" "i")]
> VSRA))]
> "TARGET_SIMD"
> - "*
> - int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[3], 1, bit_width + 1);
> - return \"<sur>sra\\t%<v>0<Vmtype>, %<v>2<Vmtype>, %3\";"
> + "<sur>sra\\t%<v>0<Vmtype>, %<v>2<Vmtype>, %3"
> [(set_attr "type" "neon_shift_acc<q>")]
> )
>
> @@ -3789,14 +3784,11 @@ (define_insn "aarch64_<sur>s<lr>i_n<mode
> [(set (match_operand:VSDQ_I_DI 0 "register_operand" "=w")
> (unspec:VSDQ_I_DI [(match_operand:VSDQ_I_DI 1 "register_operand" "0")
> (match_operand:VSDQ_I_DI 2 "register_operand" "w")
> - (match_operand:SI 3 "immediate_operand" "i")]
> + (match_operand:SI 3
> + "aarch64_simd_shift_imm_<offsetlr><ve_mode>" "i")]
> VSLRI))]
> "TARGET_SIMD"
> - "*
> - int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[3], 1 - <VSLRI:offsetlr>,
> - bit_width - <VSLRI:offsetlr> + 1);
> - return \"s<lr>i\\t%<v>0<Vmtype>, %<v>2<Vmtype>, %3\";"
> + "s<lr>i\\t%<v>0<Vmtype>, %<v>2<Vmtype>, %3"
> [(set_attr "type" "neon_shift_imm<q>")]
> )
>
> @@ -3805,13 +3797,11 @@ (define_insn "aarch64_<sur>s<lr>i_n<mode
> (define_insn "aarch64_<sur>qshl<u>_n<mode>"
> [(set (match_operand:VSDQ_I 0 "register_operand" "=w")
> (unspec:VSDQ_I [(match_operand:VSDQ_I 1 "register_operand" "w")
> - (match_operand:SI 2 "immediate_operand" "i")]
> + (match_operand:SI 2
> + "aarch64_simd_shift_imm_<ve_mode>" "i")]
> VQSHL_N))]
> "TARGET_SIMD"
> - "*
> - int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[2], 0, bit_width);
> - return \"<sur>qshl<u>\\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2\";"
> + "<sur>qshl<u>\\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2"
> [(set_attr "type" "neon_sat_shift_imm<q>")]
> )
>
> @@ -3821,13 +3811,11 @@ (define_insn "aarch64_<sur>qshl<u>_n<mod
> (define_insn "aarch64_<sur>q<r>shr<u>n_n<mode>"
> [(set (match_operand:<VNARROWQ> 0 "register_operand" "=w")
> (unspec:<VNARROWQ> [(match_operand:VSQN_HSDI 1 "register_operand" "w")
> - (match_operand:SI 2 "immediate_operand" "i")]
> + (match_operand:SI 2
> + "aarch64_simd_shift_imm_offset_<ve_mode>" "i")]
> VQSHRN_N))]
> "TARGET_SIMD"
> - "*
> - int bit_width = GET_MODE_UNIT_SIZE (<MODE>mode) * BITS_PER_UNIT;
> - aarch64_simd_const_bounds (operands[2], 1, bit_width + 1);
> - return \"<sur>q<r>shr<u>n\\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2\";"
> + "<sur>q<r>shr<u>n\\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
> [(set_attr "type" "neon_sat_shift_imm_narrow_q")]
> )
>
> diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
> index e020bd3..b87db36 100644
> --- a/gcc/config/aarch64/aarch64.c
> +++ b/gcc/config/aarch64/aarch64.c
> @@ -7931,14 +7931,13 @@ aarch64_simd_lane_bounds (rtx operand, HOST_WIDE_INT low, HOST_WIDE_INT high)
> error ("lane out of range");
> }
>
> -void
> +bool
> aarch64_simd_const_bounds (rtx operand, HOST_WIDE_INT low, HOST_WIDE_INT high)
> {
> gcc_assert (CONST_INT_P (operand));
> - HOST_WIDE_INT lane = INTVAL (operand);
> + HOST_WIDE_INT constant = INTVAL (operand);
>
> - if (lane < low || lane >= high)
> - error ("constant out of range");
> + return (constant >= low && constant <= high);
> }
>
> /* Emit code to reinterpret one AdvSIMD type as another,
> diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
> index a00283a04b3c916422ea65c5260e39db63d1a4b3..79f3ba58c28638924294c630ee86379cb1352053 100644
> --- a/gcc/config/aarch64/iterators.md
> +++ b/gcc/config/aarch64/iterators.md
> @@ -538,6 +538,14 @@ (define_mode_attr v_cmp_result [(V8QI "v
> (V2DF "v2di") (DF "di")
> (SF "si")])
>
> +;; Lower case element modes (as used in shift immediate patterns).
> +(define_mode_attr ve_mode [(V8QI "qi") (V16QI "qi")
> + (V4HI "hi") (V8HI "hi")
> + (V2SI "si") (V4SI "si")
> + (DI "di") (V2DI "di")
> + (QI "qi") (HI "hi")
> + (SI "si")])
> +
> ;; Vm for lane instructions is restricted to FP_LO_REGS.
> (define_mode_attr vwx [(V4HI "x") (V8HI "x") (HI "x")
> (V2SI "w") (V4SI "w") (SI "w")])
> @@ -1007,8 +1015,9 @@ (define_int_attr addsub [(UNSPEC_SHADD "
> (UNSPEC_RADDHN2 "add")
> (UNSPEC_RSUBHN2 "sub")])
>
> -(define_int_attr offsetlr [(UNSPEC_SSLI "1") (UNSPEC_USLI "1")
> - (UNSPEC_SSRI "0") (UNSPEC_USRI "0")])
> +(define_int_attr offsetlr [(UNSPEC_SSLI "") (UNSPEC_USLI "")
> + (UNSPEC_SSRI "offset_")
> + (UNSPEC_USRI "offset_")])
>
> ;; Standard pattern names for floating-point rounding instructions.
> (define_int_attr frint_pattern [(UNSPEC_FRINTZ "btrunc")
> diff --git a/gcc/config/aarch64/predicates.md b/gcc/config/aarch64/predicates.md
> index 8191169e89b1eaf04c00ea709af70412d2cee361..3b65b5e03f3e762d62bb7b2d59e9a05dad34162b 100644
> --- a/gcc/config/aarch64/predicates.md
> +++ b/gcc/config/aarch64/predicates.md
> @@ -279,3 +279,56 @@ (define_special_predicate "aarch64_simd_
> {
> return aarch64_const_vec_all_same_int_p (op, -1);
> })
> +
> +;; Predicates used by the various SIMD shift operations. These
> +;; fall in to 3 categories.
> +;; Shifts with a range 0-(bit_size - 1) (aarch64_simd_shift_imm)
> +;; Shifts with a range 1-bit_size (aarch64_simd_shift_imm_offset)
> +;; Shifts with a range 0-bit_size (aarch64_simd_shift_imm_bitsize)
> +(define_predicate "aarch64_simd_shift_imm_qi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 7)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_hi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 15)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_si"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 31)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_di"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 63)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_offset_qi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 1, 8)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_offset_hi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 1, 16)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_offset_si"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 1, 32)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_offset_di"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 1, 64)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_bitsize_qi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 8)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_bitsize_hi"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 16)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_bitsize_si"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 32)")))
> +
> +(define_predicate "aarch64_simd_shift_imm_bitsize_di"
> + (and (match_code "const_int")
> + (match_test "aarch64_simd_const_bounds (op, 0, 64)")))
> diff --git a/gcc/testsuite/gcc.target/aarch64/simd/vqshlb_1.c b/gcc/testsuite/gcc.target/aarch64/simd/vqshlb_1.c
> new file mode 100644
> index 0000000..ae741de
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/simd/vqshlb_1.c
> @@ -0,0 +1,21 @@
> +/* { dg-do run } */
> +/* { dg-options "-O3" } */
> +
> +#include "arm_neon.h"
> +
> +extern void abort ();
> +
> +int
> +main (int argc, char **argv)
> +{
> + int8_t arg1 = -1;
> + int8_t arg2 = 127;
> + int8_t exp = -128;
> + int8_t got = vqshlb_s8 (arg1, arg2);
> +
> + if (exp != got)
> + abort ();
> +
> + return 0;
> +}
> +