RISC-V: Support neg VLS auto-vectorization
Checks
Commit Message
#include "riscv_vector.h"
#define DEF_OP_V(PREFIX, NUM, TYPE, OP) \
void __attribute__ ((noinline, noclone)) \
PREFIX##_##TYPE##NUM (TYPE *restrict a, TYPE *restrict b) \
{ \
for (int i = 0; i < NUM; ++i) \
a[i] = OP b[i]; \
}
DEF_OP_V (neg, 16, int32_t, -)
After this patch:
neg_int32_t16:
vsetivli zero,16,e32,mf2,ta,ma
vle32.v v1,0(a1)
vneg.v v1,v1
vse32.v v1,0(a0)
ret
gcc/ChangeLog:
* config/riscv/autovec-vls.md (<optab><mode>2): Add VLS neg.
* config/riscv/vector.md: Ditto.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/vls/def.h: Ditto.
* gcc.target/riscv/rvv/autovec/vls/neg-1.c: New test.
---
gcc/config/riscv/autovec-vls.md | 21 +++++++
gcc/config/riscv/vector.md | 10 ++--
.../gcc.target/riscv/rvv/autovec/vls/def.h | 8 +++
.../gcc.target/riscv/rvv/autovec/vls/neg-1.c | 57 +++++++++++++++++++
4 files changed, 91 insertions(+), 5 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/neg-1.c
Comments
LGTM :)
Juzhe-Zhong <juzhe.zhong@rivai.ai> 於 2023年8月8日 週二 11:10 寫道:
> #include "riscv_vector.h"
>
> #define DEF_OP_V(PREFIX, NUM, TYPE, OP)
> \
> void __attribute__ ((noinline, noclone))
> \
> PREFIX##_##TYPE##NUM (TYPE *restrict a, TYPE *restrict b)
> \
> {
> \
> for (int i = 0; i < NUM; ++i)
> \
> a[i] = OP b[i];
> \
> }
>
> DEF_OP_V (neg, 16, int32_t, -)
>
> After this patch:
>
> neg_int32_t16:
> vsetivli zero,16,e32,mf2,ta,ma
> vle32.v v1,0(a1)
> vneg.v v1,v1
> vse32.v v1,0(a0)
> ret
>
> gcc/ChangeLog:
>
> * config/riscv/autovec-vls.md (<optab><mode>2): Add VLS neg.
> * config/riscv/vector.md: Ditto.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/autovec/vls/def.h: Ditto.
> * gcc.target/riscv/rvv/autovec/vls/neg-1.c: New test.
>
> ---
> gcc/config/riscv/autovec-vls.md | 21 +++++++
> gcc/config/riscv/vector.md | 10 ++--
> .../gcc.target/riscv/rvv/autovec/vls/def.h | 8 +++
> .../gcc.target/riscv/rvv/autovec/vls/neg-1.c | 57 +++++++++++++++++++
> 4 files changed, 91 insertions(+), 5 deletions(-)
> create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/neg-1.c
>
> diff --git a/gcc/config/riscv/autovec-vls.md
> b/gcc/config/riscv/autovec-vls.md
> index 4a9f8c8d0bc..1b1d940d779 100644
> --- a/gcc/config/riscv/autovec-vls.md
> +++ b/gcc/config/riscv/autovec-vls.md
> @@ -181,3 +181,24 @@
> riscv_vector::RVV_BINOP, operands);
> DONE;
> })
> +
> +;;
> -------------------------------------------------------------------------------
> +;; ---- [INT] Unary operations
> +;;
> -------------------------------------------------------------------------------
> +;; Includes:
> +;; - vneg.v/vnot.v
> +;;
> -------------------------------------------------------------------------------
> +
> +(define_insn_and_split "<optab><mode>2"
> + [(set (match_operand:VLSI 0 "register_operand")
> + (any_int_unop:VLSI
> + (match_operand:VLSI 1 "register_operand")))]
> + "TARGET_VECTOR && can_create_pseudo_p ()"
> + "#"
> + "&& 1"
> + [(const_int 0)]
> +{
> + insn_code icode = code_for_pred (<CODE>, <MODE>mode);
> + riscv_vector::emit_vlmax_insn (icode, riscv_vector::RVV_UNOP, operands);
> + DONE;
> +})
> diff --git a/gcc/config/riscv/vector.md b/gcc/config/riscv/vector.md
> index 9b51e0089d7..65b5fe456ed 100644
> --- a/gcc/config/riscv/vector.md
> +++ b/gcc/config/riscv/vector.md
> @@ -3386,8 +3386,8 @@
> ;;
> -------------------------------------------------------------------------------
>
> (define_insn "@pred_<optab><mode>"
> - [(set (match_operand:VI 0 "register_operand" "=vd,vd, vr, vr")
> - (if_then_else:VI
> + [(set (match_operand:V_VLSI 0 "register_operand" "=vd,vd, vr,
> vr")
> + (if_then_else:V_VLSI
> (unspec:<VM>
> [(match_operand:<VM> 1 "vector_mask_operand" "vm,vm,Wc1,Wc1")
> (match_operand 4 "vector_length_operand" "rK,rK, rK, rK")
> @@ -3396,9 +3396,9 @@
> (match_operand 7 "const_int_operand" " i, i, i, i")
> (reg:SI VL_REGNUM)
> (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
> - (any_int_unop:VI
> - (match_operand:VI 3 "register_operand" "vr,vr, vr, vr"))
> - (match_operand:VI 2 "vector_merge_operand" "vu, 0, vu,
> 0")))]
> + (any_int_unop:V_VLSI
> + (match_operand:V_VLSI 3 "register_operand" "vr,vr, vr,
> vr"))
> + (match_operand:V_VLSI 2 "vector_merge_operand" "vu, 0, vu,
> 0")))]
> "TARGET_VECTOR"
> "v<insn>.v\t%0,%3%p1"
> [(set_attr "type" "vialu")
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h
> b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h
> index 33916ff0698..2a5baef747b 100644
> --- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/def.h
> @@ -142,3 +142,11 @@ typedef double v512df __attribute__ ((vector_size
> (4096)));
> for (int i = 0; i < NUM; ++i)
> \
> a[i] = b[i] OP 7;
> \
> }
> +
> +#define DEF_OP_V(PREFIX, NUM, TYPE, OP)
> \
> + void __attribute__ ((noinline, noclone))
> \
> + PREFIX##_##TYPE##NUM (TYPE *restrict a, TYPE *restrict b)
> \
> + {
> \
> + for (int i = 0; i < NUM; ++i)
> \
> + a[i] = OP b[i];
> \
> + }
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/neg-1.c
> b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/neg-1.c
> new file mode 100644
> index 00000000000..0d723d70e8c
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/vls/neg-1.c
> @@ -0,0 +1,57 @@
> +/* { dg-do compile } */
> +/* { dg-options "-march=rv64gcv_zvfh_zvl4096b -mabi=lp64d -O3
> -fno-schedule-insns -fno-schedule-insns2 --param=riscv-autovec-lmul=m8" } */
> +
> +#include "def.h"
> +
> +DEF_OP_V (neg, 1, int8_t, -)
> +DEF_OP_V (neg, 2, int8_t, -)
> +DEF_OP_V (neg, 4, int8_t, -)
> +DEF_OP_V (neg, 8, int8_t, -)
> +DEF_OP_V (neg, 16, int8_t, -)
> +DEF_OP_V (neg, 32, int8_t, -)
> +DEF_OP_V (neg, 64, int8_t, -)
> +DEF_OP_V (neg, 128, int8_t, -)
> +DEF_OP_V (neg, 256, int8_t, -)
> +DEF_OP_V (neg, 512, int8_t, -)
> +DEF_OP_V (neg, 1024, int8_t, -)
> +DEF_OP_V (neg, 2048, int8_t, -)
> +DEF_OP_V (neg, 4096, int8_t, -)
> +
> +DEF_OP_V (neg, 1, int16_t, -)
> +DEF_OP_V (neg, 2, int16_t, -)
> +DEF_OP_V (neg, 4, int16_t, -)
> +DEF_OP_V (neg, 8, int16_t, -)
> +DEF_OP_V (neg, 16, int16_t, -)
> +DEF_OP_V (neg, 32, int16_t, -)
> +DEF_OP_V (neg, 64, int16_t, -)
> +DEF_OP_V (neg, 128, int16_t, -)
> +DEF_OP_V (neg, 256, int16_t, -)
> +DEF_OP_V (neg, 512, int16_t, -)
> +DEF_OP_V (neg, 1024, int16_t, -)
> +DEF_OP_V (neg, 2048, int16_t, -)
> +
> +DEF_OP_V (neg, 1, int32_t, -)
> +DEF_OP_V (neg, 2, int32_t, -)
> +DEF_OP_V (neg, 4, int32_t, -)
> +DEF_OP_V (neg, 8, int32_t, -)
> +DEF_OP_V (neg, 16, int32_t, -)
> +DEF_OP_V (neg, 32, int32_t, -)
> +DEF_OP_V (neg, 64, int32_t, -)
> +DEF_OP_V (neg, 128, int32_t, -)
> +DEF_OP_V (neg, 256, int32_t, -)
> +DEF_OP_V (neg, 512, int32_t, -)
> +DEF_OP_V (neg, 1024, int32_t, -)
> +
> +DEF_OP_V (neg, 1, int64_t, -)
> +DEF_OP_V (neg, 2, int64_t, -)
> +DEF_OP_V (neg, 4, int64_t, -)
> +DEF_OP_V (neg, 8, int64_t, -)
> +DEF_OP_V (neg, 16, int64_t, -)
> +DEF_OP_V (neg, 32, int64_t, -)
> +DEF_OP_V (neg, 64, int64_t, -)
> +DEF_OP_V (neg, 128, int64_t, -)
> +DEF_OP_V (neg, 256, int64_t, -)
> +DEF_OP_V (neg, 512, int64_t, -)
> +
> +/* { dg-final { scan-assembler-times {vneg\.v\s+v[0-9]+,\s*v[0-9]+} 42 }
> } */
> +/* { dg-final { scan-assembler-not {csrr} } } */
> --
> 2.36.1
>
>
Committed to the trunk, thanks Kito and Juzhe.
@@ -181,3 +181,24 @@
riscv_vector::RVV_BINOP, operands);
DONE;
})
+
+;; -------------------------------------------------------------------------------
+;; ---- [INT] Unary operations
+;; -------------------------------------------------------------------------------
+;; Includes:
+;; - vneg.v/vnot.v
+;; -------------------------------------------------------------------------------
+
+(define_insn_and_split "<optab><mode>2"
+ [(set (match_operand:VLSI 0 "register_operand")
+ (any_int_unop:VLSI
+ (match_operand:VLSI 1 "register_operand")))]
+ "TARGET_VECTOR && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(const_int 0)]
+{
+ insn_code icode = code_for_pred (<CODE>, <MODE>mode);
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::RVV_UNOP, operands);
+ DONE;
+})
@@ -3386,8 +3386,8 @@
;; -------------------------------------------------------------------------------
(define_insn "@pred_<optab><mode>"
- [(set (match_operand:VI 0 "register_operand" "=vd,vd, vr, vr")
- (if_then_else:VI
+ [(set (match_operand:V_VLSI 0 "register_operand" "=vd,vd, vr, vr")
+ (if_then_else:V_VLSI
(unspec:<VM>
[(match_operand:<VM> 1 "vector_mask_operand" "vm,vm,Wc1,Wc1")
(match_operand 4 "vector_length_operand" "rK,rK, rK, rK")
@@ -3396,9 +3396,9 @@
(match_operand 7 "const_int_operand" " i, i, i, i")
(reg:SI VL_REGNUM)
(reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
- (any_int_unop:VI
- (match_operand:VI 3 "register_operand" "vr,vr, vr, vr"))
- (match_operand:VI 2 "vector_merge_operand" "vu, 0, vu, 0")))]
+ (any_int_unop:V_VLSI
+ (match_operand:V_VLSI 3 "register_operand" "vr,vr, vr, vr"))
+ (match_operand:V_VLSI 2 "vector_merge_operand" "vu, 0, vu, 0")))]
"TARGET_VECTOR"
"v<insn>.v\t%0,%3%p1"
[(set_attr "type" "vialu")
@@ -142,3 +142,11 @@ typedef double v512df __attribute__ ((vector_size (4096)));
for (int i = 0; i < NUM; ++i) \
a[i] = b[i] OP 7; \
}
+
+#define DEF_OP_V(PREFIX, NUM, TYPE, OP) \
+ void __attribute__ ((noinline, noclone)) \
+ PREFIX##_##TYPE##NUM (TYPE *restrict a, TYPE *restrict b) \
+ { \
+ for (int i = 0; i < NUM; ++i) \
+ a[i] = OP b[i]; \
+ }
new file mode 100644
@@ -0,0 +1,57 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv_zvfh_zvl4096b -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2 --param=riscv-autovec-lmul=m8" } */
+
+#include "def.h"
+
+DEF_OP_V (neg, 1, int8_t, -)
+DEF_OP_V (neg, 2, int8_t, -)
+DEF_OP_V (neg, 4, int8_t, -)
+DEF_OP_V (neg, 8, int8_t, -)
+DEF_OP_V (neg, 16, int8_t, -)
+DEF_OP_V (neg, 32, int8_t, -)
+DEF_OP_V (neg, 64, int8_t, -)
+DEF_OP_V (neg, 128, int8_t, -)
+DEF_OP_V (neg, 256, int8_t, -)
+DEF_OP_V (neg, 512, int8_t, -)
+DEF_OP_V (neg, 1024, int8_t, -)
+DEF_OP_V (neg, 2048, int8_t, -)
+DEF_OP_V (neg, 4096, int8_t, -)
+
+DEF_OP_V (neg, 1, int16_t, -)
+DEF_OP_V (neg, 2, int16_t, -)
+DEF_OP_V (neg, 4, int16_t, -)
+DEF_OP_V (neg, 8, int16_t, -)
+DEF_OP_V (neg, 16, int16_t, -)
+DEF_OP_V (neg, 32, int16_t, -)
+DEF_OP_V (neg, 64, int16_t, -)
+DEF_OP_V (neg, 128, int16_t, -)
+DEF_OP_V (neg, 256, int16_t, -)
+DEF_OP_V (neg, 512, int16_t, -)
+DEF_OP_V (neg, 1024, int16_t, -)
+DEF_OP_V (neg, 2048, int16_t, -)
+
+DEF_OP_V (neg, 1, int32_t, -)
+DEF_OP_V (neg, 2, int32_t, -)
+DEF_OP_V (neg, 4, int32_t, -)
+DEF_OP_V (neg, 8, int32_t, -)
+DEF_OP_V (neg, 16, int32_t, -)
+DEF_OP_V (neg, 32, int32_t, -)
+DEF_OP_V (neg, 64, int32_t, -)
+DEF_OP_V (neg, 128, int32_t, -)
+DEF_OP_V (neg, 256, int32_t, -)
+DEF_OP_V (neg, 512, int32_t, -)
+DEF_OP_V (neg, 1024, int32_t, -)
+
+DEF_OP_V (neg, 1, int64_t, -)
+DEF_OP_V (neg, 2, int64_t, -)
+DEF_OP_V (neg, 4, int64_t, -)
+DEF_OP_V (neg, 8, int64_t, -)
+DEF_OP_V (neg, 16, int64_t, -)
+DEF_OP_V (neg, 32, int64_t, -)
+DEF_OP_V (neg, 64, int64_t, -)
+DEF_OP_V (neg, 128, int64_t, -)
+DEF_OP_V (neg, 256, int64_t, -)
+DEF_OP_V (neg, 512, int64_t, -)
+
+/* { dg-final { scan-assembler-times {vneg\.v\s+v[0-9]+,\s*v[0-9]+} 42 } } */
+/* { dg-final { scan-assembler-not {csrr} } } */