LGTM :)
Paul-Antoine Arras <par...@baylibre.com> 於 2025年9月9日 週二 22:48 寫道：
> This pattern enables the combine pass (or late-combine, depending on the
> case)
> to merge a float_extend'ed vec_duplicate into a minus RTL instruction. Both
> minus operands are widened.
>
> Before this patch, we have six instructions, e.g.:
> fcvt.d.s fa0,fa0
> vsetvli a5,zero,e64,m1,ta,ma
> vfmv.v.f v3,fa0
> vfwcvt.f.f.v v1,v2
> vsetvli zero,zero,e64,m1,ta,ma
> vfsub.vv v1,v1,v3
>
> After, we get only one:
> vfwsub.vf v1,v2,fa0
>
> gcc/ChangeLog:
>
> * config/riscv/autovec-opt.md (*vfwsub_vf_<mode>): New pattern to
> combine float_extend + vec_duplicate + vfwsub.vv into vfwsub.vf.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f16.c: Add vfwsub.vf.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f32.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f16.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f32.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f16.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f32.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f16.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f32.c: Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf_binop.h
> (DEF_VF_BINOP_WIDEN_CASE_0, DEF_VF_BINOP_WIDEN_CASE_1): Swap
> operands.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf_binop_widen_run.h:
> Likewise.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c: New
> test.
> * gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c: New
> test.
> ---
> gcc/config/riscv/autovec-opt.md | 23 +++++++++++++++++++
> .../riscv/rvv/autovec/vx_vf/vf-1-f16.c | 2 ++
> .../riscv/rvv/autovec/vx_vf/vf-1-f32.c | 2 ++
> .../riscv/rvv/autovec/vx_vf/vf-2-f16.c | 3 ++-
> .../riscv/rvv/autovec/vx_vf/vf-2-f32.c | 3 ++-
> .../riscv/rvv/autovec/vx_vf/vf-3-f16.c | 2 ++
> .../riscv/rvv/autovec/vx_vf/vf-3-f32.c | 2 ++
> .../riscv/rvv/autovec/vx_vf/vf-4-f16.c | 1 +
> .../riscv/rvv/autovec/vx_vf/vf-4-f32.c | 1 +
> .../riscv/rvv/autovec/vx_vf/vf_binop.h | 10 ++++----
> .../rvv/autovec/vx_vf/vf_binop_widen_run.h | 2 +-
> .../rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c | 20 ++++++++++++++++
> .../rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c | 16 +++++++++++++
> 13 files changed, 79 insertions(+), 8 deletions(-)
> create mode 100644
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c
> create mode 100644
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c
>
> diff --git gcc/config/riscv/autovec-opt.md gcc/config/riscv/autovec-opt.md
> index 5512c46fa8e..02f19bc6a42 100644
> --- gcc/config/riscv/autovec-opt.md
> +++ gcc/config/riscv/autovec-opt.md
> @@ -2215,6 +2215,29 @@ (define_insn_and_split "*vfwadd_wf_<mode>"
> [(set_attr "type" "vfwalu")]
> )
>
> +;; vfwsub.vf
> +(define_insn_and_split "*vfwsub_vf_<mode>"
> + [(set (match_operand:VWEXTF 0 "register_operand")
> + (minus:VWEXTF
> + (float_extend:VWEXTF
> + (match_operand:<V_DOUBLE_TRUNC> 1 "register_operand"))
> + (vec_duplicate:VWEXTF
> + (float_extend:<VEL>
> + (match_operand:<VSUBEL> 2 "register_operand")))))]
> + "TARGET_VECTOR && can_create_pseudo_p ()"
> + "#"
> + "&& 1"
> + [(const_int 0)]
> + {
> + riscv_vector::emit_vlmax_insn (code_for_pred_dual_widen_scalar (MINUS,
> +
> <MODE>mode),
> + riscv_vector::BINARY_OP_FRM_DYN,
> operands);
> +
> + DONE;
> + }
> + [(set_attr "type" "vfwalu")]
> +)
> +
> ;; vfadd.vf
> (define_insn_and_split "*vfadd_vf_<mode>"
> [(set (match_operand:V_VLSF 0 "register_operand")
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f16.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f16.c
> index fed5d3b6001..20e809010d8 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f16.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f16.c
> @@ -27,6 +27,7 @@ DEF_VF_BINOP_CASE_2_WRAP (_Float16, MAX_FUNC_0_WRAP
> (_Float16), max)
> DEF_VF_BINOP_CASE_2_WRAP (_Float16, MAX_FUNC_1_WRAP (_Float16), max)
> DEF_VF_BINOP_WIDEN_CASE_0 (_Float16, float, *, mul)
> DEF_VF_BINOP_WIDEN_CASE_0 (_Float16, float, +, add)
> +DEF_VF_BINOP_WIDEN_CASE_0 (_Float16, float, -, sub)
> DEF_VF_BINOP_WIDEN_CASE_2 (_Float16, float, +, add)
>
> /* { dg-final { scan-assembler-times {vfmadd.vf} 1 } } */
> @@ -50,4 +51,5 @@ DEF_VF_BINOP_WIDEN_CASE_2 (_Float16, float, +, add)
> /* { dg-final { scan-assembler-times {vfmax.vf} 2 } } */
> /* { dg-final { scan-assembler-times {vfwmul.vf} 1 } } */
> /* { dg-final { scan-assembler-times {vfwadd.vf} 1 } } */
> +/* { dg-final { scan-assembler-times {vfwsub.vf} 1 } } */
> /* { dg-final { scan-assembler-times {vfwadd.wf} 1 } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f32.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f32.c
> index 82d64d11c87..8ecd7d0fa00 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f32.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-1-f32.c
> @@ -27,6 +27,7 @@ DEF_VF_BINOP_CASE_2_WRAP (float, MAX_FUNC_0_WRAP
> (float), max)
> DEF_VF_BINOP_CASE_2_WRAP (float, MAX_FUNC_1_WRAP (float), max)
> DEF_VF_BINOP_WIDEN_CASE_0 (float, double, *, mul)
> DEF_VF_BINOP_WIDEN_CASE_0 (float, double, +, add)
> +DEF_VF_BINOP_WIDEN_CASE_0 (float, double, -, sub)
> DEF_VF_BINOP_WIDEN_CASE_2 (float, double, +, add)
>
> /* { dg-final { scan-assembler-times {vfmadd.vf} 1 } } */
> @@ -50,4 +51,5 @@ DEF_VF_BINOP_WIDEN_CASE_2 (float, double, +, add)
> /* { dg-final { scan-assembler-times {vfmax.vf} 2 } } */
> /* { dg-final { scan-assembler-times {vfwmul.vf} 1 } } */
> /* { dg-final { scan-assembler-times {vfwadd.vf} 1 } } */
> +/* { dg-final { scan-assembler-times {vfwsub.vf} 1 } } */
> /* { dg-final { scan-assembler-times {vfwadd.wf} 1 } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f16.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f16.c
> index eef86749c50..8fe361f4f70 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f16.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f16.c
> @@ -24,5 +24,6 @@
> /* { dg-final { scan-assembler-not {vfmax.vf} } } */
> /* { dg-final { scan-assembler-not {vfwmul.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler-not {vfwsub.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.wf} } } */
> -/* { dg-final { scan-assembler-times {fcvt.s.h} 7 } } */
> +/* { dg-final { scan-assembler-times {fcvt.s.h} 8 } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f32.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f32.c
> index c5e93060e02..a1eaaa8b47f 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f32.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-2-f32.c
> @@ -24,5 +24,6 @@
> /* { dg-final { scan-assembler-not {vfmax.vf} } } */
> /* { dg-final { scan-assembler-not {vfwmul.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler-not {vfwsub.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.wf} } } */
> -/* { dg-final { scan-assembler-times {fcvt.d.s} 7 } } */
> +/* { dg-final { scan-assembler-times {fcvt.d.s} 8 } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f16.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f16.c
> index f0c6594533e..f799437d3ca 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f16.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f16.c
> @@ -31,6 +31,7 @@ DEF_VF_BINOP_CASE_3_WRAP (_Float16, MAX_FUNC_1_WRAP
> (_Float16), max,
> VF_BINOP_FUNC_BODY_X128)
> DEF_VF_BINOP_WIDEN_CASE_1 (_Float16, float, *, mul)
> DEF_VF_BINOP_WIDEN_CASE_1 (_Float16, float, +, add)
> +DEF_VF_BINOP_WIDEN_CASE_1 (_Float16, float, -, sub)
> DEF_VF_BINOP_WIDEN_CASE_3 (_Float16, float, +, add)
>
> /* { dg-final { scan-assembler {vfmadd.vf} } } */
> @@ -54,4 +55,5 @@ DEF_VF_BINOP_WIDEN_CASE_3 (_Float16, float, +, add)
> /* { dg-final { scan-assembler {vfmax.vf} } } */
> /* { dg-final { scan-assembler {vfwmul.vf} } } */
> /* { dg-final { scan-assembler {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler {vfwsub.vf} } } */
> /* { dg-final { scan-assembler {vfwadd.wf} } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f32.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f32.c
> index 60617c3ec9b..bb987e1edc0 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f32.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-3-f32.c
> @@ -31,6 +31,7 @@ DEF_VF_BINOP_CASE_3_WRAP (float, MAX_FUNC_1_WRAP
> (float), max,
> VF_BINOP_FUNC_BODY_X128)
> DEF_VF_BINOP_WIDEN_CASE_1 (float, double, *, mul)
> DEF_VF_BINOP_WIDEN_CASE_1 (float, double, +, add)
> +DEF_VF_BINOP_WIDEN_CASE_1 (float, double, -, sub)
> DEF_VF_BINOP_WIDEN_CASE_3 (float, double, +, add)
>
> /* { dg-final { scan-assembler {vfmadd.vf} } } */
> @@ -54,4 +55,5 @@ DEF_VF_BINOP_WIDEN_CASE_3 (float, double, +, add)
> /* { dg-final { scan-assembler {vfmax.vf} } } */
> /* { dg-final { scan-assembler {vfwmul.vf} } } */
> /* { dg-final { scan-assembler {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler {vfwsub.vf} } } */
> /* { dg-final { scan-assembler {vfwadd.wf} } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f16.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f16.c
> index 0650265b6c0..50a4968718b 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f16.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f16.c
> @@ -24,5 +24,6 @@
> /* { dg-final { scan-assembler-not {vfmax.vf} } } */
> /* { dg-final { scan-assembler-not {vfwmul.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler-not {vfwsub.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.wf} } } */
> /* { dg-final { scan-assembler {fcvt.s.h} } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f32.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f32.c
> index b43699deb83..2e7ef538215 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f32.c
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf-4-f32.c
> @@ -24,5 +24,6 @@
> /* { dg-final { scan-assembler-not {vfmax.vf} } } */
> /* { dg-final { scan-assembler-not {vfwmul.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.vf} } } */
> +/* { dg-final { scan-assembler-not {vfwsub.vf} } } */
> /* { dg-final { scan-assembler-not {vfwadd.wf} } } */
> /* { dg-final { scan-assembler {fcvt.d.s} } } */
> diff --git gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop.h
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop.h
> index 2a55c9c6df9..479a6fa7222 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop.h
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop.h
> @@ -37,7 +37,7 @@
> unsigned n)
> \
> {
> \
> for (unsigned i = 0; i < n; i++)
> \
> - out[i] = (T2) f OP (T2) in[i];
> \
> + out[i] = (T2) in[i] OP (T2) f;
> \
> }
> #define DEF_VF_BINOP_WIDEN_CASE_0_WRAP(T1, T2, OP, NAME)
> \
> DEF_VF_BINOP_WIDEN_CASE_0 (T1, T2, OP, NAME)
> @@ -246,10 +246,10 @@ DEF_MAX_1 (double)
> {
> \
> for (int i = 0; i < n; i++)
> \
> {
> \
> - dst[i] = (TYPE2) * a OP (TYPE2) b[i];
> \
> - dst2[i] = (TYPE2) * a2 OP (TYPE2) b[i];
> \
> - dst3[i] = (TYPE2) * a2 OP (TYPE2) a[i];
> \
> - dst4[i] = (TYPE2) * a OP (TYPE2) b2[i];
> \
> + dst[i] = (TYPE2) b[i] OP (TYPE2) * a;
> \
> + dst2[i] = (TYPE2) b[i] OP (TYPE2) * a2;
> \
> + dst3[i] = (TYPE2) a[i] OP (TYPE2) * a2;
> \
> + dst4[i] = (TYPE2) b2[i] OP (TYPE2) * a;
> \
> }
> \
> }
>
> diff --git
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop_widen_run.h
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop_widen_run.h
> index 8748cda21aa..1c9dc8c5e7b 100644
> --- gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop_widen_run.h
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_binop_widen_run.h
> @@ -28,7 +28,7 @@ int main ()
> TEST_RUN (T1, T2, NAME, out, in, f, N);
>
> for (int i = 0; i < N; i++)
> - assert (out[i] == ((T2) f OP (T2) in[i]));
> + assert (out[i] == ((T2) in[i] OP (T2) f));
>
> return 0;
> }
> diff --git
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c
> new file mode 100644
> index 00000000000..6269073b940
> --- /dev/null
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f16.c
> @@ -0,0 +1,20 @@
> +/* { dg-do run { target { riscv_v } } } */
> +/* { dg-require-effective-target riscv_v_ok } */
> +/* { dg-require-effective-target riscv_zvfh_ok } */
> +/* { dg-add-options "riscv_v" } */
> +/* { dg-add-options "riscv_zvfh" } */
> +/* { dg-additional-options "--param=fpr2vr-cost=0" } */
> +
> +#include "vf_binop.h"
> +
> +#define T1 _Float16
> +#define T2 float
> +#define NAME sub
> +#define OP -
> +
> +DEF_VF_BINOP_WIDEN_CASE_0_WRAP (T1, T2, OP, NAME)
> +
> +#define TEST_RUN(T1, T2, NAME, out, in, f, n)
> RUN_VF_BINOP_WIDEN_CASE_0_WRAP(T1, T2, NAME, out, in, f, n)
> +#define LIMIT -32768
> +
> +#include "vf_binop_widen_run.h"
> diff --git
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c
> gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c
> new file mode 100644
> index 00000000000..9e23db1f8af
> --- /dev/null
> +++ gcc/testsuite/gcc.target/riscv/rvv/autovec/vx_vf/vf_vfwsub-run-1-f32.c
> @@ -0,0 +1,16 @@
> +/* { dg-do run { target { riscv_v } } } */
> +/* { dg-additional-options "--param=fpr2vr-cost=0" } */
> +
> +#include "vf_binop.h"
> +
> +#define T1 float
> +#define T2 double
> +#define NAME sub
> +#define OP -
> +
> +DEF_VF_BINOP_WIDEN_CASE_0_WRAP (T1, T2, OP, NAME)
> +
> +#define TEST_RUN(T1, T2, NAME, out, in, f, n)
> RUN_VF_BINOP_WIDEN_CASE_0_WRAP(T1, T2, NAME, out, in, f, n)
> +#define LIMIT -2147483648
> +
> +#include "vf_binop_widen_run.h"
> --
> 2.39.5
>
>
