> Is that for the case where the target cannot do a widen_mul? Or why
> is this important?
No, the below pattern will generate mul/widen_mul depending on types.
679 #define DEF_SAT_U_MUL_FMT_1(NT, WT) \
680 NT __attribute__((noinline)) \
681 sat_u_mul_##NT##_from_##WT##_fmt_1 (NT a, NT b) \
682 { \
683 WT x = (WT)a * (WT)b; \
684 NT max = -1; \
685 if (x > (WT)(max)) \
686 return max; \
687 else \
688 return (NT)x; \
689 }
If DEF_SAT_U_MUL_FMT_1(uint32_t, uint128_t), then widen_mul
If DEF_SAT_U_MUL_FMT_1(uint32_t, uint64_t), then mul.
And the layout not that the same between the widen_mul and mul, thus I added
the mul based part,
to make sure the other type combines are covered.
Pan
-----Original Message-----
From: Richard Biener <richard.guent...@gmail.com>
Sent: Monday, July 28, 2025 6:51 PM
To: Li, Pan2 <pan2...@intel.com>
Cc: gcc-patches@gcc.gnu.org; juzhe.zh...@rivai.ai; kito.ch...@gmail.com;
jeffreya...@gmail.com; rdapp....@gmail.com; Chen, Ken <ken.c...@intel.com>;
Liu, Hongtao <hongtao....@intel.com>
Subject: Re: [PATCH v1 0/2] Add mul based unsigned SAT_MUL for form 1
On Sat, Jul 26, 2025 at 3:13 PM <pan2...@intel.com> wrote:
>
> From: Pan Li <pan2...@intel.com>
>
> This patch series would like to support the unsigned SAT_MUL with
> the help of mul, instead of the widen_mul. Aka:
Is that for the case where the target cannot do a widen_mul? Or why
is this important?
Richard.
> NT __attribute__((noinline))
> sat_u_mul_##NT##_fmt_1 (NT a, NT b)
> {
> uint64_t x = (uint64_t)a * (uint64_t)b;
> NT max = -1;
> if (x > (uint64_t)(max))
> return max;
> else
> return (NT)x;
> }
>
> Take NT as uint32_t as example, we will have:
>
> Before this patch:
> 15 | <bb 2> [local count: 1073741824]:
> 16 | _1 = (long unsigned int) a_4(D);
> 17 | _2 = (long unsigned int) b_5(D);
> 18 | x_6 = _1 * _2;
> 19 | _7 = MIN_EXPR <x_6, 4294967295>;
> 20 | _3 = (uint32_t) _7;
> 21 | return _3;
>
> After this patch:
> 9 | <bb 2> [local count: 1073741824]:
> 10 | _3 = .SAT_MUL (a_4(D), b_5(D)); [tail call]
> 11 | return _3;
>
> The below test suites are passed for this patch:
> 1. The rv64gcv fully regression tests.
> 2. The rv32gcv regression tests for sat_mul.
> 3. The x86 bootstrap tests.
> 4. The x86 fully regression tests.
>
> Pan Li (2):
> Match: Introduce mul based pattern for unsigned SAT_MUL
> RISC-V: Add test cases for mul based unsigned scalar SAT_MUL
>
> gcc/match.pd | 12 ++++++++++++
> .../riscv/sat/sat_u_mul-1-u16-from-u32.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-1-u8-from-u16.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-1-u8-from-u32.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-2-u16-from-u64.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-2-u32-from-u64.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-2-u8-from-u64.c | 11 +++++++++++
> .../riscv/sat/sat_u_mul-run-1-u16-from-u32.c | 16 ++++++++++++++++
> .../riscv/sat/sat_u_mul-run-1-u16-from-u64.c | 2 +-
> .../riscv/sat/sat_u_mul-run-1-u32-from-u64.c | 2 +-
> .../riscv/sat/sat_u_mul-run-1-u8-from-u16.c | 16 ++++++++++++++++
> .../riscv/sat/sat_u_mul-run-1-u8-from-u32.c | 16 ++++++++++++++++
> .../riscv/sat/sat_u_mul-run-1-u8-from-u64.c | 2 +-
> 13 files changed, 129 insertions(+), 3 deletions(-)
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-1-u16-from-u32.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-1-u8-from-u16.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-1-u8-from-u32.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-2-u16-from-u64.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-2-u32-from-u64.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-2-u8-from-u64.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-run-1-u16-from-u32.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-run-1-u8-from-u16.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/sat/sat_u_mul-run-1-u8-from-u32.c
>
> --
> 2.43.0
>
