@@ -513,7 +513,7 @@ (define_expand "<optab><mode>2"
})
;; -------------------------------------------------------------------------------
-;; - ABS expansion to vmslt and vneg
+;; - [INT] ABS expansion to vmslt and vneg.
;; -------------------------------------------------------------------------------
(define_expand "abs<mode>2"
@@ -532,6 +532,40 @@ (define_expand "abs<mode>2"
DONE;
})
+;; -------------------------------------------------------------------------------
+;; ---- [FP] Unary operations
+;; -------------------------------------------------------------------------------
+;; Includes:
+;; - vfneg.v/vfabs.v
+;; -------------------------------------------------------------------------------
+(define_expand "<optab><mode>2"
+ [(set (match_operand:VF_AUTO 0 "register_operand")
+ (any_float_unop_nofrm:VF_AUTO
+ (match_operand:VF_AUTO 1 "register_operand")))]
+ "TARGET_VECTOR"
+{
+ insn_code icode = code_for_pred (<CODE>, <MODE>mode);
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::RVV_UNOP, operands);
+ DONE;
+})
+
+;; -------------------------------------------------------------------------------
+;; - [FP] Square root
+;; -------------------------------------------------------------------------------
+;; Includes:
+;; - vfsqrt.v
+;; -------------------------------------------------------------------------------
+(define_expand "<optab><mode>2"
+ [(set (match_operand:VF_AUTO 0 "register_operand")
+ (any_float_unop:VF_AUTO
+ (match_operand:VF_AUTO 1 "register_operand")))]
+ "TARGET_VECTOR"
+{
+ insn_code icode = code_for_pred (<CODE>, <MODE>mode);
+ riscv_vector::emit_vlmax_fp_insn (icode, riscv_vector::RVV_UNOP, operands);
+ DONE;
+})
+
;; =========================================================================
;; == Ternary arithmetic
;; =========================================================================
@@ -1,5 +1,5 @@
/* { dg-do run { target { riscv_vector } } } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "abs-template.h"
@@ -7,30 +7,32 @@
#define SZ 128
-#define RUN(TYPE) \
- TYPE a##TYPE[SZ]; \
- for (int i = 0; i < SZ; i++) \
- { \
- if (i & 1) \
- a##TYPE[i] = i - 64; \
- else \
- a##TYPE[i] = i; \
- } \
- vabs_##TYPE (a##TYPE, a##TYPE, SZ); \
- for (int i = 0; i < SZ; i++) \
- { \
- if (i & 1) \
- assert (a##TYPE[i] == abs (i - 64)); \
- else \
- assert (a##TYPE[i] == i); \
+#define RUN(TYPE) \
+ TYPE a##TYPE[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ if (i & 1) \
+ a##TYPE[i] = i - 64; \
+ else \
+ a##TYPE[i] = i; \
+ } \
+ vabs_##TYPE (a##TYPE, a##TYPE, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ if (i & 1) \
+ assert (a##TYPE[i] == __builtin_abs (i - 64)); \
+ else \
+ assert (a##TYPE[i] == i); \
}
-#define RUN_ALL() \
- RUN(int8_t) \
- RUN(int16_t) \
- RUN(int32_t) \
- RUN(int64_t)
+#define RUN_ALL() \
+ RUN(int8_t) \
+ RUN(int16_t) \
+ RUN(int32_t) \
+ RUN(int64_t) \
+ RUN(float) \
+ RUN(double) \
int main ()
{
@@ -1,8 +1,9 @@
/* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "abs-template.h"
/* { dg-final { scan-assembler-times {\tvseti?vli\s+[a-z0-9,]+,ta,mu} 4 } } */
/* { dg-final { scan-assembler-times {\tvmslt\.vi} 4 } } */
/* { dg-final { scan-assembler-times {\tvneg.v\sv[0-9]+,v[0-9]+,v0\.t} 4 } } */
+/* { dg-final { scan-assembler-times {\tvfabs.v} 3 } } */
@@ -1,8 +1,9 @@
/* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "abs-template.h"
/* { dg-final { scan-assembler-times {\tvseti?vli\s+[a-z0-9,]+,ta,mu} 4 } } */
/* { dg-final { scan-assembler-times {\tvmslt\.vi} 4 } } */
/* { dg-final { scan-assembler-times {\tvneg.v\sv[0-9]+,v[0-9]+,v0\.t} 4 } } */
+/* { dg-final { scan-assembler-times {\tvfabs.v} 3 } } */
@@ -6,7 +6,7 @@
void vabs_##TYPE (TYPE *dst, TYPE *a, int n) \
{ \
for (int i = 0; i < n; i++) \
- dst[i] = abs (a[i]); \
+ dst[i] = __builtin_abs (a[i]); \
}
#define TEST_TYPE2(TYPE) \
@@ -14,13 +14,24 @@
void vabs_##TYPE (TYPE *dst, TYPE *a, int n) \
{ \
for (int i = 0; i < n; i++) \
- dst[i] = llabs (a[i]); \
+ dst[i] = __builtin_llabs (a[i]); \
+ }
+
+#define TEST_TYPE3(TYPE) \
+ __attribute__((noipa)) \
+ void vabs_##TYPE (TYPE *dst, TYPE *a, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ dst[i] = __builtin_fabs (a[i]); \
}
#define TEST_ALL() \
TEST_TYPE(int8_t) \
TEST_TYPE(int16_t) \
TEST_TYPE(int32_t) \
- TEST_TYPE2(int64_t)
+ TEST_TYPE2(int64_t) \
+ TEST_TYPE3(_Float16) \
+ TEST_TYPE3(float) \
+ TEST_TYPE3(double) \
TEST_ALL()
new file mode 100644
@@ -0,0 +1,35 @@
+/* { dg-do run { target { riscv_zvfh_hw } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "abs-template.h"
+
+#include <assert.h>
+
+#define SZ 128
+
+#define RUN(TYPE) \
+ TYPE a##TYPE[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ if (i & 1) \
+ a##TYPE[i] = i - 64; \
+ else \
+ a##TYPE[i] = i; \
+ } \
+ vabs_##TYPE (a##TYPE, a##TYPE, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ if (i & 1) \
+ assert (a##TYPE[i] == __builtin_abs (i - 64)); \
+ else \
+ assert (a##TYPE[i] == i); \
+ }
+
+
+#define RUN_ALL() \
+ RUN(_Float16) \
+
+int main ()
+{
+ RUN_ALL()
+}
new file mode 100644
@@ -0,0 +1,30 @@
+/* { dg-do run { target { riscv_vector } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "vfsqrt-template.h"
+
+#include <assert.h>
+
+#define SZ 255
+
+#define EPS 1e-5
+
+#define RUN(TYPE) \
+ TYPE a##TYPE[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ a##TYPE[i] = (TYPE)i; \
+ } \
+ vsqrt_##TYPE (a##TYPE, a##TYPE, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ assert (__builtin_fabs \
+ (a##TYPE[i] - __builtin_sqrtf ((TYPE)i)) < EPS); \
+
+#define RUN_ALL() \
+ RUN(float) \
+ RUN(double) \
+
+int main ()
+{
+ RUN_ALL()
+}
new file mode 100644
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "vfsqrt-template.h"
+
+/* We cannot link this without the Zvfh extension so define
+ it here instead of in the template directly (which is also
+ included by the run test that might need to be linked without
+ the extension). */
+TEST_TYPE3(_Float16)
+
+/* { dg-final { scan-assembler-times {\tvfsqrt\.v} 3 } } */
new file mode 100644
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "vfsqrt-template.h"
+
+/* We cannot link this without the Zvfh extension so define
+ it here instead of in the template directly (which is also
+ included by the run test that might need to be linked without
+ the extension). */
+TEST_TYPE3(_Float16)
+
+/* { dg-final { scan-assembler-times {\tvfsqrt\.v} 3 } } */
new file mode 100644
@@ -0,0 +1,31 @@
+#include <stdint-gcc.h>
+
+#define TEST_TYPE(TYPE) \
+ __attribute__((noipa)) \
+ void vsqrt_##TYPE (TYPE *dst, TYPE *a, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ dst[i] = __builtin_sqrtf (a[i]); \
+ }
+
+#define TEST_TYPE2(TYPE) \
+ __attribute__((noipa)) \
+ void vsqrt_##TYPE (TYPE *dst, TYPE *a, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ dst[i] = __builtin_sqrt (a[i]); \
+ }
+
+#define TEST_TYPE3(TYPE) \
+ __attribute__((noipa)) \
+ void vsqrt_##TYPE (TYPE *dst, TYPE *a, int n) \
+ { \
+ for (int i = 0; i < n; i++) \
+ dst[i] = __builtin_sqrtf16 (a[i]); \
+ }
+
+#define TEST_ALL() \
+ TEST_TYPE(float) \
+ TEST_TYPE2(double) \
+
+TEST_ALL()
new file mode 100644
@@ -0,0 +1,33 @@
+/* { dg-do run { target { riscv_zvfh_hw } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "vfsqrt-template.h"
+
+/* We cannot link this without the Zvfh extension so define
+ it here instead of in the template directly. */
+TEST_TYPE3(_Float16)
+
+#include <assert.h>
+
+#define SZ 255
+
+#define EPS 1e-5
+
+#define RUN(TYPE) \
+ TYPE a##TYPE[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ a##TYPE[i] = (TYPE)i; \
+ } \
+ vsqrt_##TYPE (a##TYPE, a##TYPE, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ assert (__builtin_fabs \
+ (a##TYPE[i] - __builtin_sqrtf ((TYPE)i)) < EPS); \
+
+#define RUN_ALL() \
+ RUN(_Float16) \
+
+int main ()
+{
+ RUN_ALL()
+}
@@ -1,5 +1,5 @@
/* { dg-do run { target { riscv_vector } } } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "vneg-template.h"
@@ -13,7 +13,7 @@
{ \
a##TYPE[i] = i - 127; \
} \
- vneg_##TYPE (a##TYPE, a##TYPE, SZ); \
+ vneg_##TYPE (a##TYPE, a##TYPE, SZ); \
for (int i = 0; i < SZ; i++) \
assert (a##TYPE[i] == -(i - 127));
@@ -21,7 +21,9 @@
RUN(int8_t) \
RUN(int16_t) \
RUN(int32_t) \
- RUN(int64_t)
+ RUN(int64_t) \
+ RUN(float) \
+ RUN(double) \
int main ()
{
@@ -1,6 +1,7 @@
/* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv32gcv_zvfh -mabi=ilp32d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "vneg-template.h"
/* { dg-final { scan-assembler-times {\tvneg\.v} 4 } } */
+/* { dg-final { scan-assembler-times {\tvfneg\.v} 3 } } */
@@ -1,6 +1,7 @@
/* { dg-do compile } */
-/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax" } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model -march=rv64gcv_zvfh -mabi=lp64d --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
#include "vneg-template.h"
/* { dg-final { scan-assembler-times {\tvneg\.v} 4 } } */
+/* { dg-final { scan-assembler-times {\tvfneg\.v} 3 } } */
@@ -13,6 +13,9 @@
TEST_TYPE(int8_t) \
TEST_TYPE(int16_t) \
TEST_TYPE(int32_t) \
- TEST_TYPE(int64_t)
+ TEST_TYPE(int64_t) \
+ TEST_TYPE(_Float16) \
+ TEST_TYPE(float) \
+ TEST_TYPE(double) \
TEST_ALL()
new file mode 100644
@@ -0,0 +1,26 @@
+/* { dg-do run { target { riscv_zvfh_hw } } } */
+/* { dg-additional-options "-std=c99 -fno-vect-cost-model --param=riscv-autovec-preference=fixed-vlmax -ffast-math" } */
+
+#include "vneg-template.h"
+
+#include <assert.h>
+
+#define SZ 255
+
+#define RUN(TYPE) \
+ TYPE a##TYPE[SZ]; \
+ for (int i = 0; i < SZ; i++) \
+ { \
+ a##TYPE[i] = i - 127; \
+ } \
+ vneg_##TYPE (a##TYPE, a##TYPE, SZ); \
+ for (int i = 0; i < SZ; i++) \
+ assert (a##TYPE[i] == -(i - 127));
+
+#define RUN_ALL() \
+ RUN(_Float16) \
+
+int main ()
+{
+ RUN_ALL()
+}
@@ -31,5 +31,19 @@ void f4 (_Float16 * __restrict a, _Float16 * __restrict b, int n)
a[i] = a[i]/b[i];
}
-/* We can't enable FP16 NEG/PLUS/MINUS/MULT/DIV auto-vectorization when -march="*zvfhmin*". */
+void f6 (_Float16 * __restrict a, _Float16 * __restrict b, int n)
+{
+ for (int i = 0; i < n; i++)
+ a[i] = __builtin_fabs (b[i]);
+}
+
+void f7 (_Float16 * __restrict a, _Float16 * __restrict b, int n)
+{
+ for (int i = 0; i < n; i++)
+ a[i] = __builtin_sqrtf (b[i]);
+}
+
+/* We can't enable FP16 NEG/PLUS/MINUS/MULT/DIV/ABS/SQRTF auto-vectorization
+ when -march="*zvfhmin*" because the min variant of the extension only
+ provides loads, stores and conversions. */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 0 "vect" } } */