This patch is to enhance vwmul.vv combine optimizations.
Consider this following code:
void
vwadd_int16_t_int8_t (int16_t *__restrict dst, int16_t *__restrict dst2,
int16_t *__restrict dst3, int16_t *__restrict dst4,
int8_t *__restrict a, int8_t *__restrict b,
int8_t *__restrict a2, int8_t *__restrict b2, int n)
{
for (int i = 0; i < n; i++)
{
dst[i] = (int16_t) a[i] * (int16_t) b[i];
dst2[i] = (int16_t) a2[i] * (int16_t) b[i];
dst3[i] = (int16_t) a2[i] * (int16_t) a[i];
dst4[i] = (int16_t) a[i] * (int16_t) b2[i];
}
}
In such complicate case, the operand is not single used, used by multiple statements.
GCC combine optimization will iterate the combination of the operands.
Also, we add another pattern of vwmulsu.vv to enhance the vwmulsu.vv optimization.
Currently, we have format:
(mult: (sign_extend) (zero_extend)) in vector.md for intrinsics calling.
Now, we add a new vwmulsu.ww with this format:
(mult: (zero_extend) (sign_extend))
To handle this following cases (sign and unsigned widening multiplication mixing codes):
void
vwadd_int16_t_int8_t (int16_t *__restrict dst, int16_t *__restrict dst2,
int16_t *__restrict dst3, int16_t *__restrict dst4,
int8_t *__restrict a, uint8_t *__restrict b,
uint8_t *__restrict a2, int8_t *__restrict b2, int n)
{
for (int i = 0; i < n; i++)
{
dst[i] = (int16_t) a[i] * (int16_t) b[i];
dst2[i] = (int16_t) a2[i] * (int16_t) b[i];
dst3[i] = (int16_t) a2[i] * (int16_t) a[i];
dst4[i] = (int16_t) a[i] * (int16_t) b2[i];
}
}
Before this patch:
...
vsext.vf2 v6,v1
add t0,a0,t4
vzext.vf2 v4,v1
vmul.vv v2,v4,v6
add t0,a1,t4
vzext.vf2 v2,v1
vmul.vv v4,v2,v4
add t0,a2,t4
vmul.vv v2,v2,v6
add t0,a3,t4
sub t6,t6,t1
vsext.vf2 v2,v1
vmul.vv v2,v2,v6
...
After this patch:
...
add t0,a0,t3
vwmulsu.vv v2,v1,v3
add t0,a1,t3
vwmulu.vv v4,v3,v2
add t0,a2,t3
vwmulsu.vv v3,v1,v2
add t0,a3,t3
sub t4,t4,t1
vwmul.vv v2,v1,v3
...
gcc/ChangeLog:
* config/riscv/vector.md: Add vector-opt.md.
* config/riscv/autovec-opt.md: New file.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/widen/widen-7.c: New test.
* gcc.target/riscv/rvv/autovec/widen/widen-complicate-3.c: New test.
* gcc.target/riscv/rvv/autovec/widen/widen-complicate-4.c: New test.
* gcc.target/riscv/rvv/autovec/widen/widen_run-7.c: New test.
--- /dev/null
+;; Machine description for optimization of RVV auto-vectorization.
+;; Copyright (C) 2023 Free Software Foundation, Inc.
+;; Contributed by Juzhe Zhong (juzhe.zhong@rivai.ai), RiVAI Technologies Ltd.
+
+;; This file is part of GCC.
+
+;; GCC is free software; you can redistribute it and/or modify
+;; it under the terms of the GNU General Public License as published by
+;; the Free Software Foundation; either version 3, or (at your option)
+;; any later version.
+
+;; GCC is distributed in the hope that it will be useful,
+;; but WITHOUT ANY WARRANTY; without even the implied warranty of
+;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+;; GNU General Public License for more details.
+
+;; You should have received a copy of the GNU General Public License
+;; along with GCC; see the file COPYING3. If not see
+;; <http://www.gnu.org/licenses/>.
+
+;; We don't have vwmul.wv instruction like vwadd.wv in RVV.
+;; This pattern is an intermediate RTL IR as a pseudo vwmul.wv to enhance
+;; optimization of instructions combine.
+(define_insn_and_split "@pred_single_widen_mul<any_extend:su><mode>"
+ [(set (match_operand:VWEXTI 0 "register_operand" "=&vr,&vr")
+ (if_then_else:VWEXTI
+ (unspec:<VM>
+ [(match_operand:<VM> 1 "vector_mask_operand" "vmWc1,vmWc1")
+ (match_operand 5 "vector_length_operand" " rK, rK")
+ (match_operand 6 "const_int_operand" " i, i")
+ (match_operand 7 "const_int_operand" " i, i")
+ (match_operand 8 "const_int_operand" " i, i")
+ (reg:SI VL_REGNUM)
+ (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
+ (mult:VWEXTI
+ (any_extend:VWEXTI
+ (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" " vr, vr"))
+ (match_operand:VWEXTI 3 "register_operand" " vr, vr"))
+ (match_operand:VWEXTI 2 "vector_merge_operand" " vu, 0")))]
+ "TARGET_VECTOR"
+ "#"
+ "&& can_create_pseudo_p ()"
+ [(const_int 0)]
+ {
+ insn_code icode = code_for_pred_vf2 (<CODE>, <MODE>mode);
+ rtx tmp = gen_reg_rtx (<MODE>mode);
+ rtx ops[] = {tmp, operands[4]};
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::RVV_UNOP, ops);
+
+ emit_insn (gen_pred (MULT, <MODE>mode, operands[0], operands[1], operands[2],
+ operands[3], tmp, operands[5], operands[6],
+ operands[7], operands[8]));
+ DONE;
+ }
+ [(set_attr "type" "viwmul")
+ (set_attr "mode" "<MODE>")])
+
+;; This pattern it to enchance the instruction combine optimizations for complicate
+;; sign and unsigned widening multiplication operations.
+(define_insn "*pred_widen_mulsu<mode>"
+ [(set (match_operand:VWEXTI 0 "register_operand" "=&vr,&vr")
+ (if_then_else:VWEXTI
+ (unspec:<VM>
+ [(match_operand:<VM> 1 "vector_mask_operand" "vmWc1,vmWc1")
+ (match_operand 5 "vector_length_operand" " rK, rK")
+ (match_operand 6 "const_int_operand" " i, i")
+ (match_operand 7 "const_int_operand" " i, i")
+ (match_operand 8 "const_int_operand" " i, i")
+ (reg:SI VL_REGNUM)
+ (reg:SI VTYPE_REGNUM)] UNSPEC_VPREDICATE)
+ (mult:VWEXTI
+ (zero_extend:VWEXTI
+ (match_operand:<V_DOUBLE_TRUNC> 4 "register_operand" " vr, vr"))
+ (sign_extend:VWEXTI
+ (match_operand:<V_DOUBLE_TRUNC> 3 "register_operand" " vr, vr")))
+ (match_operand:VWEXTI 2 "vector_merge_operand" " vu, 0")))]
+ "TARGET_VECTOR"
+ "vwmulsu.vv\t%0,%3,%4%p1"
+ [(set_attr "type" "viwmul")
+ (set_attr "mode" "<V_DOUBLE_TRUNC>")])
;;
;; - Intrinsics (https://github.com/riscv/rvv-intrinsic-doc)
;; - Auto-vectorization (autovec.md)
-;; - Combine optimization (TBD)
+;; - Optimization (autovec-opt.md)
(include "vector-iterators.md")
)
(include "autovec.md")
+(include "autovec-opt.md")
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=scalable" } */
+
+#include <stdint-gcc.h>
+
+#define TEST_TYPE(TYPE1, TYPE2) \
+ __attribute__ ((noipa)) void vwmul_##TYPE1_##TYPE2 (TYPE1 *__restrict dst, \
+ TYPE2 *__restrict a, \
+ TYPE1 *__restrict b, \
+ int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ dst[i] = ((TYPE1) a[i]) * b[i]; \
+ }
+
+#define TEST_ALL() \
+ TEST_TYPE (int16_t, int8_t) \
+ TEST_TYPE (uint16_t, uint8_t) \
+ TEST_TYPE (int32_t, int16_t) \
+ TEST_TYPE (uint32_t, uint16_t) \
+ TEST_TYPE (int64_t, int32_t) \
+ TEST_TYPE (uint64_t, uint32_t)
+
+TEST_ALL ()
+
+/* { dg-final { scan-assembler-times {\tvsext\.vf2} 3 } } */
+/* { dg-final { scan-assembler-times {\tvzext\.vf2} 3 } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=scalable" } */
+
+#include <stdint-gcc.h>
+
+#define TEST_TYPE(TYPE1, TYPE2) \
+ __attribute__ ((noipa)) void vwadd_##TYPE1_##TYPE2 ( \
+ TYPE1 *__restrict dst, TYPE1 *__restrict dst2, TYPE1 *__restrict dst3, \
+ TYPE1 *__restrict dst4, TYPE2 *__restrict a, TYPE2 *__restrict b, \
+ TYPE2 *__restrict a2, TYPE2 *__restrict b2, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ { \
+ dst[i] = (TYPE1) a[i] * (TYPE1) b[i]; \
+ dst2[i] = (TYPE1) a2[i] * (TYPE1) b[i]; \
+ dst3[i] = (TYPE1) a2[i] * (TYPE1) a[i]; \
+ dst4[i] = (TYPE1) a[i] * (TYPE1) b2[i]; \
+ } \
+ }
+
+#define TEST_ALL() \
+ TEST_TYPE (int16_t, int8_t) \
+ TEST_TYPE (uint16_t, uint8_t) \
+ TEST_TYPE (int32_t, int16_t) \
+ TEST_TYPE (uint32_t, uint16_t) \
+ TEST_TYPE (int64_t, int32_t) \
+ TEST_TYPE (uint64_t, uint32_t)
+
+TEST_ALL ()
+
+/* { dg-final { scan-assembler-times {\tvwmul\.vv} 12 } } */
+/* { dg-final { scan-assembler-times {\tvwmulu\.vv} 12 } } */
--- /dev/null
+/* { dg-do compile } */
+/* { dg-additional-options "-march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=scalable" } */
+
+#include <stdint-gcc.h>
+
+#define TEST_TYPE(TYPE1, TYPE2, TYPE3) \
+ __attribute__ ((noipa)) void vwadd_##TYPE1_##TYPE2 ( \
+ TYPE1 *__restrict dst, TYPE1 *__restrict dst2, TYPE1 *__restrict dst3, \
+ TYPE1 *__restrict dst4, TYPE2 *__restrict a, TYPE3 *__restrict b, \
+ TYPE3 *__restrict a2, TYPE2 *__restrict b2, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ { \
+ dst[i] = (TYPE1) a[i] * (TYPE1) b[i]; \
+ dst2[i] = (TYPE1) a2[i] * (TYPE1) b[i]; \
+ dst3[i] = (TYPE1) a2[i] * (TYPE1) a[i]; \
+ dst4[i] = (TYPE1) a[i] * (TYPE1) b2[i]; \
+ } \
+ }
+
+#define TEST_ALL() \
+ TEST_TYPE (int16_t, int8_t, uint8_t) \
+ TEST_TYPE (int32_t, int16_t, uint16_t) \
+ TEST_TYPE (int64_t, int32_t, uint32_t)
+
+TEST_ALL ()
+
+/* { dg-final { scan-assembler-times {\tvwmulsu\.vv} 6 } } */
+/* { dg-final { scan-assembler-times {\tvwmul\.vv} 3 } } */
+/* { dg-final { scan-assembler-times {\tvwmulu\.vv} 3 } } */
+/* { dg-final { scan-assembler-not {\tvmul} } } */
--- /dev/null
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "--param=riscv-autovec-preference=scalable" } */
+
+#include <assert.h>
+#include "widen-7.c"
+
+#define SZ 512
+
+#define RUN(TYPE1, TYPE2, LIMIT) \
+ TYPE2 a##TYPE2[SZ]; \
+ TYPE1 b##TYPE1[SZ]; \
+ TYPE1 dst##TYPE1[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ a##TYPE2[i] = LIMIT + i % LIMIT; \
+ b##TYPE1[i] = LIMIT + i & LIMIT; \
+ } \
+ vwmul_##TYPE1_##TYPE2 (dst##TYPE1, a##TYPE2, b##TYPE1, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ assert (dst##TYPE1[i] == (((TYPE1) a##TYPE2[i]) * b##TYPE1[i]));
+
+#define RUN_ALL() \
+ RUN (int16_t, int8_t, -128) \
+ RUN (uint16_t, uint8_t, 255) \
+ RUN (int32_t, int16_t, -32768) \
+ RUN (uint32_t, uint16_t, 65535) \
+ RUN (int64_t, int32_t, -2147483648) \
+ RUN (uint64_t, uint32_t, 4294967295)
+
+int
+main ()
+{
+ RUN_ALL ()
+}
projects / thirdparty / gcc.git / commitdiff
