Update in V2 with more information in commit log: https://gcc.gnu.org/pipermail/gcc-patches/2023-December/640863.html
juzhe.zh...@rivai.ai From: Juzhe-Zhong Date: 2023-12-18 18:59 To: gcc-patches CC: kito.cheng; kito.cheng; jeffreyalaw; rdapp.gcc; Juzhe-Zhong Subject: [PATCH] RISC-V: Support one more overlap for wv instructions For 'wv' instructions, e.g. vwadd.wv vd,vs2,vs1. vs2 has same EEW as vd. vs1 has smaller than vd. So, vs2 can overlap with vd, but vs1 can only overlap highest-number of vd when LMUL of vs1 is greater than 1. We already have supported overlap for vs1 LMUL >= 1. But I forget vs1 LMUL < 1, vs2 can overlap vd even though vs1 totally can not overlap vd. Consider the reduction auto-vectorization: int64_t reduc_plus_int (int *__restrict a, int n) { int64_t r = 0; for (int i = 0; i < n; ++i) r += a[i]; return r; } Before this patch: reduc_plus_int: ble a1,zero,.L4 vsetvli a5,zero,e64,m1,ta,ma vmv.v.i v1,0 .L3: vsetvli a5,a1,e32,mf2,tu,ma slli a4,a5,2 sub a1,a1,a5 vle32.v v2,0(a0) vmv1r.v v3,v1 ----> This should be removed. add a0,a0,a4 vwadd.wv v1,v3,v2 ----> vs2 should be v1 bne a1,zero,.L3 li a5,0 vsetivli zero,1,e64,m1,ta,ma vmv.s.x v2,a5 vsetvli a5,zero,e64,m1,ta,ma vredsum.vs v1,v1,v2 vmv.x.s a0,v1 ret .L4: li a0,0 ret After this patch: reduc_plus_int: ble a1,zero,.L4 vsetvli a5,zero,e64,m1,ta,ma vmv.v.i v1,0 .L3: vsetvli a5,a1,e32,mf2,tu,ma slli a4,a5,2 sub a1,a1,a5 vle32.v v2,0(a0) add a0,a0,a4 vwadd.wv v1,v1,v2 bne a1,zero,.L3 li a5,0 vsetivli zero,1,e64,m1,ta,ma vmv.s.x v2,a5 vsetvli a5,zero,e64,m1,ta,ma vredsum.vs v1,v1,v2 vmv.x.s a0,v1 ret .L4: li a0,0 ret PR target/112432 gcc/ChangeLog: * config/riscv/riscv.md (none,W21,W42,W84,W43,W86,W87): Add W0. (none,W21,W42,W84,W43,W86,W87,W0): Ditto. * config/riscv/vector.md: Ditto. gcc/testsuite/ChangeLog: * gcc.target/riscv/rvv/base/pr112432-42.c: New test. --- gcc/config/riscv/riscv.md | 14 +++- gcc/config/riscv/vector.md | 84 +++++++++---------- .../gcc.target/riscv/rvv/base/pr112432-42.c | 30 +++++++ 3 files changed, 82 insertions(+), 46 deletions(-) create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/pr112432-42.c diff --git a/gcc/config/riscv/riscv.md b/gcc/config/riscv/riscv.md index eed997116b0..ee8b71c22aa 100644 --- a/gcc/config/riscv/riscv.md +++ b/gcc/config/riscv/riscv.md @@ -503,7 +503,7 @@ ;; Widening instructions have group-overlap constraints. Those are only ;; valid for certain register-group sizes. This attribute marks the ;; alternatives not matching the required register-group size as disabled. -(define_attr "group_overlap" "none,W21,W42,W84,W43,W86,W87" +(define_attr "group_overlap" "none,W21,W42,W84,W43,W86,W87,W0" (const_string "none")) (define_attr "group_overlap_valid" "no,yes" @@ -524,9 +524,9 @@ ;; According to RVV ISA: ;; The destination EEW is greater than the source EEW, the source EMUL is at least 1, - ;; and the overlap is in the highest-numbered part of the destination register group - ;; (e.g., when LMUL=8, vzext.vf4 v0, v6 is legal, but a source of v0, v2, or v4 is not). - ;; So the source operand should have LMUL >= 1. + ;; and the overlap is in the highest-numbered part of the destination register group + ;; (e.g., when LMUL=8, vzext.vf4 v0, v6 is legal, but a source of v0, v2, or v4 is not). + ;; So the source operand should have LMUL >= 1. (and (eq_attr "group_overlap" "W43") (match_test "riscv_get_v_regno_alignment (GET_MODE (operands[0])) != 4 && riscv_get_v_regno_alignment (GET_MODE (operands[3])) >= 1")) @@ -536,6 +536,12 @@ (match_test "riscv_get_v_regno_alignment (GET_MODE (operands[0])) != 8 && riscv_get_v_regno_alignment (GET_MODE (operands[3])) >= 1")) (const_string "no") + + ;; W21 supports highest-number overlap for source LMUL = 1. + ;; For 'wv' variant, we can also allow wide source operand overlaps dest operand. + (and (eq_attr "group_overlap" "W0") + (match_test "riscv_get_v_regno_alignment (GET_MODE (operands[0])) > 1")) + (const_string "no") ] (const_string "yes"))) diff --git a/gcc/config/riscv/vector.md b/gcc/config/riscv/vector.md index 7646615b12a..d475e14a823 100644 --- a/gcc/config/riscv/vector.md +++ b/gcc/config/riscv/vector.md @@ -3776,48 +3776,48 @@ (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) (define_insn "@pred_single_widen_sub<any_extend:su><mode>" - [(set (match_operand:VWEXTI 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, ?&vr, ?&vr") + [(set (match_operand:VWEXTI 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, &vr, &vr, &vr, ?&vr") (if_then_else:VWEXTI (unspec:<VM> - [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1") - (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") - (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") + [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1,vmWc1,vmWc1") + (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") + (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") (reg:SI VL_REGNUM) (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE) (minus:VWEXTI - (match_operand:VWEXTI 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr") + (match_operand:VWEXTI 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, 0, 0, vr, vr") (any_extend:VWEXTI - (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr"))) - (match_operand:VWEXTI 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0")))] + (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr, vr, vr"))) + (match_operand:VWEXTI 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0, vu, 0")))] "TARGET_VECTOR" "vwsub<any_extend:u>.wv\t%0,%3,%4%p1" [(set_attr "type" "viwalu") (set_attr "mode" "<V_DOUBLE_TRUNC>") - (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) + (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,W0,W0,none,none")]) (define_insn "@pred_single_widen_add<any_extend:su><mode>" - [(set (match_operand:VWEXTI 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, ?&vr, ?&vr") + [(set (match_operand:VWEXTI 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, &vr, &vr, &vr, ?&vr") (if_then_else:VWEXTI (unspec:<VM> - [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1") - (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") - (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") + [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1,vmWc1,vmWc1") + (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") + (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") (reg:SI VL_REGNUM) (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE) (plus:VWEXTI (any_extend:VWEXTI - (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr")) - (match_operand:VWEXTI 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr")) - (match_operand:VWEXTI 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0")))] + (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr, vr, vr")) + (match_operand:VWEXTI 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, 0, 0, vr, vr")) + (match_operand:VWEXTI 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0, vu, 0")))] "TARGET_VECTOR" "vwadd<any_extend:u>.wv\t%0,%3,%4%p1" [(set_attr "type" "viwalu") (set_attr "mode" "<V_DOUBLE_TRUNC>") - (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) + (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,W0,W0,none,none")]) (define_insn "@pred_single_widen_<plus_minus:optab><any_extend:su><mode>_scalar" [(set (match_operand:VWEXTI 0 "register_operand" "=vr, vr") @@ -7056,56 +7056,56 @@ (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) (define_insn "@pred_single_widen_add<mode>" - [(set (match_operand:VWEXTF 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, ?&vr, ?&vr") + [(set (match_operand:VWEXTF 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, &vr, &vr, &vr, ?&vr") (if_then_else:VWEXTF (unspec:<VM> - [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1") - (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") - (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 9 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") + [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1,vmWc1,vmWc1") + (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") + (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 9 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") (reg:SI VL_REGNUM) (reg:SI VTYPE_REGNUM) (reg:SI FRM_REGNUM)] UNSPEC_VPREDICATE) (plus:VWEXTF (float_extend:VWEXTF - (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr")) - (match_operand:VWEXTF 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr")) - (match_operand:VWEXTF 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0")))] + (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr, vr, vr")) + (match_operand:VWEXTF 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, 0, 0, vr, vr")) + (match_operand:VWEXTF 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0, vu, 0")))] "TARGET_VECTOR" "vfwadd.wv\t%0,%3,%4%p1" [(set_attr "type" "vfwalu") (set_attr "mode" "<V_DOUBLE_TRUNC>") (set (attr "frm_mode") (symbol_ref "riscv_vector::get_frm_mode (operands[9])")) - (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) + (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,W0,W0,none,none")]) (define_insn "@pred_single_widen_sub<mode>" - [(set (match_operand:VWEXTF 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, ?&vr, ?&vr") + [(set (match_operand:VWEXTF 0 "register_operand" "=vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, vd, vr, &vr, &vr, &vr, ?&vr") (if_then_else:VWEXTF (unspec:<VM> - [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1") - (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") - (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") - (match_operand 9 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i") + [(match_operand:<VM> 1 "vector_mask_operand" " vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1, vm,Wc1,vmWc1,vmWc1,vmWc1,vmWc1") + (match_operand 5 "vector_length_operand" " rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK, rK") + (match_operand 6 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 7 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 8 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") + (match_operand 9 "const_int_operand" " i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i") (reg:SI VL_REGNUM) (reg:SI VTYPE_REGNUM) (reg:SI FRM_REGNUM)] UNSPEC_VPREDICATE) (minus:VWEXTF - (match_operand:VWEXTF 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr") + (match_operand:VWEXTF 3 "register_operand" " vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, vr, 0, 0, vr, vr") (float_extend:VWEXTF - (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr"))) - (match_operand:VWEXTF 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0")))] + (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84, vr, vr, vr, vr"))) + (match_operand:VWEXTF 2 "vector_merge_operand" " vu, vu, 0, 0, vu, vu, 0, 0, vu, vu, 0, 0, vu, 0, vu, 0")))] "TARGET_VECTOR" "vfwsub.wv\t%0,%3,%4%p1" [(set_attr "type" "vfwalu") (set_attr "mode" "<V_DOUBLE_TRUNC>") (set (attr "frm_mode") (symbol_ref "riscv_vector::get_frm_mode (operands[9])")) - (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,none,none")]) + (set_attr "group_overlap" "W21,W21,W21,W21,W42,W42,W42,W42,W84,W84,W84,W84,W0,W0,none,none")]) (define_insn "@pred_single_widen_<plus_minus:optab><mode>_scalar" [(set (match_operand:VWEXTF 0 "register_operand" "=vr, vr") diff --git a/gcc/testsuite/gcc.target/riscv/rvv/base/pr112432-42.c b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112432-42.c new file mode 100644 index 00000000000..1ee5b20a899 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112432-42.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -ffast-math" } */ + +#include <stdint-gcc.h> + +int64_t +reduc_plus_int (int *__restrict a, int n) +{ + int64_t r = 0; + for (int i = 0; i < n; ++i) + r += a[i]; + return r; +} + +double +reduc_plus_float (float *__restrict a, int n) +{ + double r = 0; + for (int i = 0; i < n; ++i) + r += a[i]; + return r; +} + +/* { dg-final { scan-assembler-not {vmv1r} } } */ +/* { dg-final { scan-assembler-not {vmv2r} } } */ +/* { dg-final { scan-assembler-not {vmv4r} } } */ +/* { dg-final { scan-assembler-not {vmv8r} } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ +/* { dg-final { scan-assembler-times {vwadd\.wv} 1 } } */ +/* { dg-final { scan-assembler-times {vfwadd\.wv} 1 } } */ -- 2.36.3