Thanks Richard for comments.
> OK.
Given almost all of the Saturation match pattern(s) are merged, I'd like to
start the refinement about "common" match predictions to avoid duplication
as we discussed in previous, after I draft the strided ld/st IFN patch(es).
Pan
-----Original Message-----
From: Richard Biener <richard.guent...@gmail.com>
Sent: Monday, October 21, 2024 8:29 PM
To: Li, Pan2 <pan2...@intel.com>
Cc: gcc-patches@gcc.gnu.org; tamar.christ...@arm.com; juzhe.zh...@rivai.ai;
kito.ch...@gmail.com; jeffreya...@gmail.com; rdapp....@gmail.com
Subject: Re: [PATCH 01/11] Match: Support form 1 for vector signed integer
SAT_TRUNC
On Mon, Oct 14, 2024 at 1:12 PM <pan2...@intel.com> wrote:
>
> From: Pan Li <pan2...@intel.com>
>
> This patch would like to support the form 1 of the vector signed
> integer SAT_TRUNC. Aka below example:
>
> Form 1:
> #define DEF_VEC_SAT_S_TRUNC_FMT_1(NT, WT, NT_MIN, NT_MAX) \
> void __attribute__((noinline)) \
> vec_sat_s_trunc_##NT##_##WT##_fmt_1 (NT *out, WT *in, unsigned limit) \
> { \
> unsigned i; \
> for (i = 0; i < limit; i++) \
> { \
> WT x = in[i]; \
> NT trunc = (NT)x; \
> out[i] = (WT)NT_MIN <= x && x <= (WT)NT_MAX \
> ? trunc \
> : x < 0 ? NT_MIN : NT_MAX; \
> } \
> }
>
> DEF_VEC_SAT_S_TRUNC_FMT_1(int32_t, int64_t, INT32_MIN, INT32_MAX)
>
> Before this patch:
> 48 │ _87 = .SELECT_VL (ivtmp_85, POLY_INT_CST [2, 2]);
> 49 │ ivtmp_64 = _87 * 8;
> 50 │ vect_x_14.10_67 = .MASK_LEN_LOAD (vectp_in.8_65, 64B, { -1, ... },
> _87, 0);
> 51 │ vect_trunc_15.21_78 = (vector([2,2]) int) vect_x_14.10_67;
> 52 │ _61 = VIEW_CONVERT_EXPR<vector([2,2]) unsigned
> long>(vect_x_14.10_67);
> 53 │ _32 = _61 >> 63;
> 54 │ vect_patt_52.16_73 = (vector([2,2]) int) _32;
> 55 │ vect__46.17_74 = VIEW_CONVERT_EXPR<vector([2,2]) unsigned
> int>(vect_patt_52.16_73);
> 56 │ vect__47.18_75 = -vect__46.17_74;
> 57 │ vect__21.19_76 = VIEW_CONVERT_EXPR<vector([2,2])
> int>(vect__47.18_75);
> 58 │ vect_x.11_68 = VIEW_CONVERT_EXPR<vector([2,2]) unsigned
> long>(vect_x_14.10_67);
> 59 │ vect__5.12_69 = vect_x.11_68 + { 2147483648, ... };
> 60 │ mask__34.13_70 = vect__5.12_69 > { 4294967295, ... };
> 61 │ _25 = .COND_XOR (mask__34.13_70, vect__21.19_76, { 2147483647, ...
> }, vect_trunc_15.21_78);
> 62 │ ivtmp_80 = _87 * 4;
> 63 │ .MASK_LEN_STORE (vectp_out.23_81, 32B, { -1, ... }, _87, 0, _25);
> 64 │ vectp_in.8_66 = vectp_in.8_65 + ivtmp_64;
> 65 │ vectp_out.23_82 = vectp_out.23_81 + ivtmp_80;
> 66 │ ivtmp_86 = ivtmp_85 - _87;
>
> After this patch:
> 38 │ _77 = .SELECT_VL (ivtmp_75, POLY_INT_CST [2, 2]);
> 39 │ ivtmp_65 = _77 * 8;
> 40 │ vect_x_14.10_68 = .MASK_LEN_LOAD (vectp_in.8_66, 64B, { -1, ... },
> _77, 0);
> 41 │ vect_patt_53.11_69 = .SAT_TRUNC (vect_x_14.10_68);
> 42 │ ivtmp_70 = _77 * 4;
> 43 │ .MASK_LEN_STORE (vectp_out.12_71, 32B, { -1, ... }, _77, 0,
> vect_patt_53.11_69);
> 44 │ vectp_in.8_67 = vectp_in.8_66 + ivtmp_65;
> 45 │ vectp_out.12_72 = vectp_out.12_71 + ivtmp_70;
> 46 │ ivtmp_76 = ivtmp_75 - _77;
>
> The below test suites are passed for this patch.
> * The rv64gcv fully regression test.
> * The x86 bootstrap test.
> * The x86 fully regression test.
OK.
Richard.
> gcc/ChangeLog:
>
> * match.pd: Refine matching for vector signed SAT_TRUNC form 1.
>
> Signed-off-by: Pan Li <pan2...@intel.com>
> ---
> gcc/match.pd | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index 2ab76f9d055..36195d0c920 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -3493,7 +3493,9 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> SAT_S_TRUNC(X) = (unsigned)X + NT_MAX + 1 > Unsigned_MAX ? (NT)X. */
> (match (signed_integer_sat_trunc @0)
> (cond^ (gt (plus:c (convert@4 @0) INTEGER_CST@1) INTEGER_CST@2)
> - (bit_xor:c (negate (convert (lt @0 integer_zerop))) INTEGER_CST@3)
> + (bit_xor:c (nop_convert?
> + (negate (nop_convert? (convert (lt @0 integer_zerop)))))
> + INTEGER_CST@3)
> (convert @0))
> (if (INTEGRAL_TYPE_P (type) && !TYPE_UNSIGNED (type)
> && !TYPE_UNSIGNED (TREE_TYPE (@0)) && TYPE_UNSIGNED (TREE_TYPE (@4)))
> --
> 2.43.0
>
