/* { dg-final { scan-assembler-times {vmin.vx} 2 } } */
/* { dg-final { scan-assembler-times {vsadd.vx} 1 } } */
/* { dg-final { scan-assembler-times {vssub.vx} 1 } } */
-/* { dg-final { scan-assembler-times {vaadd.vx} 1 } } */
+/* { dg-final { scan-assembler-times {vaadd.vx} 2 } } */
/* { dg-final { scan-assembler-times {vmin.vx} 2 } } */
/* { dg-final { scan-assembler-times {vsadd.vx} 1 } } */
/* { dg-final { scan-assembler-times {vssub.vx} 1 } } */
-/* { dg-final { scan-assembler-times {vaadd.vx} 1 } } */
+/* { dg-final { scan-assembler-times {vaadd.vx} 2 } } */
/* { dg-final { scan-assembler-times {vmin.vx} 2 } } */
/* { dg-final { scan-assembler-times {vsadd.vx} 1 } } */
/* { dg-final { scan-assembler-times {vssub.vx} 1 } } */
-/* { dg-final { scan-assembler-times {vaadd.vx} 1 { target { no-opts "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m2" "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m4" } } } } */
+/* { dg-final { scan-assembler-times {vaadd.vx} 2 { target { no-opts
+ "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m2"
+ "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m4"
+ } } } } */
/* { dg-final { scan-assembler-times {vmin.vx} 2 } } */
/* { dg-final { scan-assembler-times {vsadd.vx} 1 } } */
/* { dg-final { scan-assembler-times {vssub.vx} 1 } } */
-/* { dg-final { scan-assembler-times {vaadd.vx} 1 } } */
+/* { dg-final { scan-assembler-times {vaadd.vx} 2 } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X4)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X4)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
/* { dg-final { scan-assembler {vmin.vx} } } */
/* { dg-final { scan-assembler-not {vsadd.vx} } } */
/* { dg-final { scan-assembler-not {vssub.vx} } } */
-/* { dg-final { scan-assembler {vaadd.vx} { target { no-opts "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m2" "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m4" } } } } */
+/* { dg-final { scan-assembler {vaadd.vx} { target { no-opts {
+ "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m2"
+ "-O3 -mrvv-vector-bits=zvl -mrvv-max-lmul=m4"
+ } } } } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler-not {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X4)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X4)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler-not {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler-not {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X4)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X4)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X4)
/* { dg-final { scan-assembler {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY)
/* { dg-final { scan-assembler-not {vadd.vx} } } */
/* { dg-final { scan-assembler-not {vsub.vx} } } */
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_ADD_FUNC_WRAP(T), sat_add, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, SAT_S_SUB_FUNC_WRAP(T), sat_sub, VX_BINARY_FUNC_BODY_X8)
DEF_VX_BINARY_CASE_3_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor, VX_BINARY_FUNC_BODY_X8)
+DEF_VX_BINARY_CASE_3_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil, VX_BINARY_FUNC_BODY_X8)
/* { dg-final { scan-assembler-not {vadd.vx} } } */
/* { dg-final { scan-assembler {vsub.vx} } } */
DEF_AVG_CEIL(uint16_t, uint32_t)
DEF_AVG_CEIL(uint32_t, uint64_t)
+DEF_AVG_CEIL(int8_t, int16_t)
+DEF_AVG_CEIL(int16_t, int32_t)
+DEF_AVG_CEIL(int32_t, int64_t)
+
#ifdef HAS_INT128
DEF_AVG_FLOOR(uint64_t, uint128_t)
DEF_AVG_FLOOR(int64_t, int128_t)
DEF_AVG_CEIL(uint64_t, uint128_t)
+ DEF_AVG_CEIL(int64_t, int128_t)
#endif
#define AVG_FLOOR_FUNC(T) test_##T##_avg_floor
DEF_VX_BINARY_CASE_2_WRAP(T, SAT_S_ADD_FUNC(T), sat_add) \
DEF_VX_BINARY_CASE_2_WRAP(T, SAT_S_SUB_FUNC(T), sat_sub) \
DEF_VX_BINARY_CASE_2_WRAP(T, AVG_FLOOR_FUNC_WRAP(T), avg_floor) \
+ DEF_VX_BINARY_CASE_2_WRAP(T, AVG_CEIL_FUNC_WRAP(T), avg_ceil) \
#define TEST_BINARY_VX_UNSIGNED_0(T) \
DEF_VX_BINARY_CASE_0_WRAP(T, +, add) \
},
};
+int8_t TEST_BINARY_DATA(int8_t, avg_ceil)[][3][N] =
+{
+ {
+ { 0 },
+ {
+ 2, 2, 2, 2,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 4, 4, 4, 4,
+ },
+ {
+ 1, 1, 1, 1,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 2, 2, 2, 2,
+ },
+ },
+ {
+ { 127 },
+ {
+ 127, 127, 127, 127,
+ -128, -128, -128, -128,
+ -127, -127, -127, -127,
+ 1, 1, 1, 1,
+ },
+ {
+ 127, 127, 127, 127,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 64, 64, 64, 64,
+ },
+ },
+ {
+ {-128 },
+ {
+ 0, 0, 0, 0,
+ -128, -128, -128, -128,
+ 126, 126, 126, 126,
+ 127, 127, 127, 127,
+ },
+ {
+ -64, -64, -64, -64,
+ -128, -128, -128, -128,
+ -1, -1, -1, -1,
+ 0, 0, 0, 0,
+ },
+ },
+};
+
+int16_t TEST_BINARY_DATA(int16_t, avg_ceil)[][3][N] =
+{
+ {
+ { 0 },
+ {
+ 2, 2, 2, 2,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 4, 4, 4, 4,
+ },
+ {
+ 1, 1, 1, 1,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 2, 2, 2, 2,
+ },
+ },
+ {
+ { 32767 },
+ {
+ 32767, 32767, 32767, 32767,
+ -32768, -32768, -32768, -32768,
+ -32767, -32767, -32767, -32767,
+ 1, 1, 1, 1,
+ },
+ {
+ 32767, 32767, 32767, 32767,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 16384, 16384, 16384, 16384,
+ },
+ },
+ {
+ {-32768 },
+ {
+ 0, 0, 0, 0,
+ -32768, -32768, -32768, -32768,
+ 32766, 32766, 32766, 32766,
+ 32767, 32767, 32767, 32767,
+ },
+ {
+ -16384, -16384, -16384, -16384,
+ -32768, -32768, -32768, -32768,
+ -1, -1, -1, -1,
+ 0, 0, 0, 0,
+ },
+ },
+};
+
+int32_t TEST_BINARY_DATA(int32_t, avg_ceil)[][3][N] =
+{
+ {
+ { 0 },
+ {
+ 2, 2, 2, 2,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 4, 4, 4, 4,
+ },
+ {
+ 1, 1, 1, 1,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 2, 2, 2, 2,
+ },
+ },
+ {
+ { 2147483647 },
+ {
+ 2147483647, 2147483647, 2147483647, 2147483647,
+ -2147483648, -2147483648, -2147483648, -2147483648,
+ -2147483647, -2147483647, -2147483647, -2147483647,
+ 1, 1, 1, 1,
+ },
+ {
+ 2147483647, 2147483647, 2147483647, 2147483647,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 1073741824, 1073741824, 1073741824, 1073741824,
+ },
+ },
+ {
+ {-2147483648 },
+ {
+ 0, 0, 0, 0,
+ -2147483648, -2147483648, -2147483648, -2147483648,
+ 2147483646, 2147483646, 2147483646, 2147483646,
+ 2147483647, 2147483647, 2147483647, 2147483647,
+ },
+ {
+ -1073741824, -1073741824, -1073741824, -1073741824,
+ -2147483648, -2147483648, -2147483648, -2147483648,
+ -1, -1, -1, -1,
+ 0, 0, 0, 0,
+ },
+ },
+};
+
+int64_t TEST_BINARY_DATA(int64_t, avg_ceil)[][3][N] =
+{
+ {
+ { 0 },
+ {
+ 2, 2, 2, 2,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 4, 4, 4, 4,
+ },
+ {
+ 1, 1, 1, 1,
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 2, 2, 2, 2,
+ },
+ },
+ {
+ { 9223372036854775807ull },
+ {
+ 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull,
+ -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull,
+ -9223372036854775807ull, -9223372036854775807ull, -9223372036854775807ull, -9223372036854775807ull,
+ 1, 1, 1, 1,
+ },
+ {
+ 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull,
+ 0, 0, 0, 0,
+ 0, 0, 0, 0,
+ 4611686018427387904ull, 4611686018427387904ull, 4611686018427387904ull, 4611686018427387904ull,
+ },
+ },
+ {
+ {-9223372036854775808ull },
+ {
+ 0, 0, 0, 0,
+ -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull,
+ 9223372036854775806ull, 9223372036854775806ull, 9223372036854775806ull, 9223372036854775806ull,
+ 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull, 9223372036854775807ull,
+ },
+ {
+ -4611686018427387904ull, -4611686018427387904ull, -4611686018427387904ull, -4611686018427387904ull,
+ -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull, -9223372036854775808ull,
+ -1, -1, -1, -1,
+ 0, 0, 0, 0,
+ },
+ },
+};
+
#endif
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "-std=c99 --param=gpr2vr-cost=0" } */
+
+#include "vx_binary.h"
+#include "vx_binary_data.h"
+
+#define T int16_t
+#define NAME avg_ceil
+#define FUNC AVG_CEIL_FUNC_WRAP(T)
+#define TEST_DATA TEST_BINARY_DATA_WRAP(T, NAME)
+
+DEF_VX_BINARY_CASE_2_WRAP(T, FUNC, NAME)
+
+#define TEST_RUN(T, NAME, out, in, x, n) \
+ RUN_VX_BINARY_CASE_2_WRAP(T, NAME, FUNC, out, in, x, n)
+
+#include "vx_binary_run.h"
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "-std=c99 --param=gpr2vr-cost=0" } */
+
+#include "vx_binary.h"
+#include "vx_binary_data.h"
+
+#define T int32_t
+#define NAME avg_ceil
+#define FUNC AVG_CEIL_FUNC_WRAP(T)
+#define TEST_DATA TEST_BINARY_DATA_WRAP(T, NAME)
+
+DEF_VX_BINARY_CASE_2_WRAP(T, FUNC, NAME)
+
+#define TEST_RUN(T, NAME, out, in, x, n) \
+ RUN_VX_BINARY_CASE_2_WRAP(T, NAME, FUNC, out, in, x, n)
+
+#include "vx_binary_run.h"
--- /dev/null
+/* { dg-do run { target { riscv_v && rv64 } } } */
+/* { dg-additional-options "-std=c99 --param=gpr2vr-cost=0" } */
+
+#include "vx_binary.h"
+#include "vx_binary_data.h"
+
+#define T int64_t
+#define NAME avg_ceil
+#define FUNC AVG_CEIL_FUNC_WRAP(T)
+#define TEST_DATA TEST_BINARY_DATA_WRAP(T, NAME)
+
+DEF_VX_BINARY_CASE_2_WRAP(T, FUNC, NAME)
+
+#define TEST_RUN(T, NAME, out, in, x, n) \
+ RUN_VX_BINARY_CASE_2_WRAP(T, NAME, FUNC, out, in, x, n)
+
+#include "vx_binary_run.h"
--- /dev/null
+/* { dg-do run { target { riscv_v } } } */
+/* { dg-additional-options "-std=c99 --param=gpr2vr-cost=0" } */
+
+#include "vx_binary.h"
+#include "vx_binary_data.h"
+
+#define T int8_t
+#define NAME avg_ceil
+#define FUNC AVG_CEIL_FUNC_WRAP(T)
+#define TEST_DATA TEST_BINARY_DATA_WRAP(T, NAME)
+
+DEF_VX_BINARY_CASE_2_WRAP(T, FUNC, NAME)
+
+#define TEST_RUN(T, NAME, out, in, x, n) \
+ RUN_VX_BINARY_CASE_2_WRAP(T, NAME, FUNC, out, in, x, n)
+
+#include "vx_binary_run.h"