Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: Add vmla[q]_lane[q]_[s16|s32|u16|u32|f32] #483

Merged
merged 1 commit into from
Jul 30, 2024
Merged

feat: Add vmla[q]_lane[q]_[s16|s32|u16|u32|f32] #483

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
50 changes: 40 additions & 10 deletions neon2rvv.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9036,25 +9036,55 @@ FORCE_INLINE float32x4_t vmlaq_lane_f32(float32x4_t a, float32x4_t b, float32x2_
return __riscv_vfmacc_vv_f32m1(a, b, c_dup_lane, 4);
}

// FORCE_INLINE int16x4_t vmla_laneq_s16(int16x4_t a, int16x4_t b, int16x8_t v, const int lane);
FORCE_INLINE int16x4_t vmla_laneq_s16(int16x4_t a, int16x4_t b, int16x8_t c, const int lane) {
vint16m1_t c_dup_lane = __riscv_vrgather_vx_i16m1(c, lane, 8);
return __riscv_vmacc_vv_i16m1(a, b, c_dup_lane, 4);
}

// FORCE_INLINE int16x8_t vmlaq_laneq_s16(int16x8_t a, int16x8_t b, int16x8_t v, const int lane);
FORCE_INLINE int16x8_t vmlaq_laneq_s16(int16x8_t a, int16x8_t b, int16x8_t c, const int lane) {
vint16m1_t c_dup_lane = __riscv_vrgather_vx_i16m1(c, lane, 8);
return __riscv_vmacc_vv_i16m1(a, b, c_dup_lane, 8);
}

// FORCE_INLINE int32x2_t vmla_laneq_s32(int32x2_t a, int32x2_t b, int32x4_t v, const int lane);
FORCE_INLINE int32x2_t vmla_laneq_s32(int32x2_t a, int32x2_t b, int32x4_t c, const int lane) {
vint32m1_t c_dup_lane = __riscv_vrgather_vx_i32m1(c, lane, 4);
return __riscv_vmacc_vv_i32m1(a, b, c_dup_lane, 2);
}

// FORCE_INLINE int32x4_t vmlaq_laneq_s32(int32x4_t a, int32x4_t b, int32x4_t v, const int lane);
FORCE_INLINE int32x4_t vmlaq_laneq_s32(int32x4_t a, int32x4_t b, int32x4_t c, const int lane) {
vint32m1_t c_dup_lane = __riscv_vrgather_vx_i32m1(c, lane, 4);
return __riscv_vmacc_vv_i32m1(a, b, c_dup_lane, 4);
}

// FORCE_INLINE uint16x4_t vmla_laneq_u16(uint16x4_t a, uint16x4_t b, uint16x8_t v, const int lane);
FORCE_INLINE uint16x4_t vmla_laneq_u16(uint16x4_t a, uint16x4_t b, uint16x8_t c, const int lane) {
vuint16m1_t c_dup_lane = __riscv_vrgather_vx_u16m1(c, lane, 8);
return __riscv_vmacc_vv_u16m1(a, b, c_dup_lane, 4);
}

// FORCE_INLINE uint16x8_t vmlaq_laneq_u16(uint16x8_t a, uint16x8_t b, uint16x8_t v, const int lane);
FORCE_INLINE uint16x8_t vmlaq_laneq_u16(uint16x8_t a, uint16x8_t b, uint16x8_t c, const int lane) {
vuint16m1_t c_dup_lane = __riscv_vrgather_vx_u16m1(c, lane, 8);
return __riscv_vmacc_vv_u16m1(a, b, c_dup_lane, 8);
}

// FORCE_INLINE uint32x2_t vmla_laneq_u32(uint32x2_t a, uint32x2_t b, uint32x4_t v, const int lane);
FORCE_INLINE uint32x2_t vmla_laneq_u32(uint32x2_t a, uint32x2_t b, uint32x4_t c, const int lane) {
vuint32m1_t c_dup_lane = __riscv_vrgather_vx_u32m1(c, lane, 4);
return __riscv_vmacc_vv_u32m1(a, b, c_dup_lane, 2);
}

// FORCE_INLINE uint32x4_t vmlaq_laneq_u32(uint32x4_t a, uint32x4_t b, uint32x4_t v, const int lane);
FORCE_INLINE uint32x4_t vmlaq_laneq_u32(uint32x4_t a, uint32x4_t b, uint32x4_t c, const int lane) {
vuint32m1_t c_dup_lane = __riscv_vrgather_vx_u32m1(c, lane, 4);
return __riscv_vmacc_vv_u32m1(a, b, c_dup_lane, 4);
}

// FORCE_INLINE float32x2_t vmla_laneq_f32(float32x2_t a, float32x2_t b, float32x4_t v, const int lane);
FORCE_INLINE float32x2_t vmla_laneq_f32(float32x2_t a, float32x2_t b, float32x4_t c, const int lane) {
vfloat32m1_t c_dup_lane = __riscv_vrgather_vx_f32m1(c, lane, 4);
return __riscv_vfmacc_vv_f32m1(a, b, c_dup_lane, 2);
}

// FORCE_INLINE float32x4_t vmlaq_laneq_f32(float32x4_t a, float32x4_t b, float32x4_t v, const int lane);
FORCE_INLINE float32x4_t vmlaq_laneq_f32(float32x4_t a, float32x4_t b, float32x4_t c, const int lane) {
vfloat32m1_t c_dup_lane = __riscv_vrgather_vx_f32m1(c, lane, 4);
return __riscv_vfmacc_vv_f32m1(a, b, c_dup_lane, 4);
}

FORCE_INLINE uint16x8_t vmlaq_lane_u16(uint16x8_t a, uint16x8_t b, uint16x4_t c, const int lane) {
vuint16m1_t c_dup_lane = __riscv_vrgather_vx_u16m1(c, lane, 8);
Expand Down
270 changes: 260 additions & 10 deletions tests/impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32555,25 +32555,275 @@ result_t test_vmlaq_lane_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#endif // ENABLE_TEST_ALL
}

result_t test_vmla_laneq_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
result_t test_vmla_laneq_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const int16_t *_a = (const int16_t *)impl.test_cases_int_pointer1;
const int16_t *_b = (const int16_t *)impl.test_cases_int_pointer2;
const int16_t *_c = (const int16_t *)impl.test_cases_int_pointer3;
int16_t _d[4];
int16x4_t a = vld1_s16(_a);
int16x4_t b = vld1_s16(_b);
int16x8_t c = vld1q_s16(_c);
int16x4_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 4; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmla_laneq_s16(a, b, c, IDX); \
CHECK_RESULT(validate_int16(d, _d[0], _d[1], _d[2], _d[3]))

IMM_8_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const int16_t *_a = (const int16_t *)impl.test_cases_int_pointer1;
const int16_t *_b = (const int16_t *)impl.test_cases_int_pointer2;
const int16_t *_c = (const int16_t *)impl.test_cases_int_pointer3;
int16_t _d[8];
int16x8_t a = vld1q_s16(_a);
int16x8_t b = vld1q_s16(_b);
int16x8_t c = vld1q_s16(_c);
int16x8_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 8; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmlaq_laneq_s16(a, b, c, IDX); \
CHECK_RESULT(validate_int16(d, _d[0], _d[1], _d[2], _d[3], _d[4], _d[5], _d[6], _d[7]))

IMM_8_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmla_laneq_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const int32_t *_a = (const int32_t *)impl.test_cases_int_pointer1;
const int32_t *_b = (const int32_t *)impl.test_cases_int_pointer2;
const int32_t *_c = (const int32_t *)impl.test_cases_int_pointer3;
int32_t _d[2];
int32x2_t a = vld1_s32(_a);
int32x2_t b = vld1_s32(_b);
int32x4_t c = vld1q_s32(_c);
int32x2_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 2; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmla_laneq_s32(a, b, c, IDX); \
CHECK_RESULT(validate_int32(d, _d[0], _d[1]))

IMM_4_ITER
#undef TEST_IMPL

result_t test_vmlaq_laneq_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const int32_t *_a = (const int32_t *)impl.test_cases_int_pointer1;
const int32_t *_b = (const int32_t *)impl.test_cases_int_pointer2;
const int32_t *_c = (const int32_t *)impl.test_cases_int_pointer3;
int32_t _d[4];
int32x4_t a = vld1q_s32(_a);
int32x4_t b = vld1q_s32(_b);
int32x4_t c = vld1q_s32(_c);
int32x4_t d;

result_t test_vmla_laneq_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
#define TEST_IMPL(IDX) \
for (int i = 0; i < 4; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmlaq_laneq_s32(a, b, c, IDX); \
CHECK_RESULT(validate_int32(d, _d[0], _d[1], _d[2], _d[3]))

result_t test_vmlaq_laneq_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
IMM_4_ITER
#undef TEST_IMPL

result_t test_vmla_laneq_u16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_u16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
result_t test_vmla_laneq_u16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const uint16_t *_a = (const uint16_t *)impl.test_cases_int_pointer1;
const uint16_t *_b = (const uint16_t *)impl.test_cases_int_pointer2;
const uint16_t *_c = (const uint16_t *)impl.test_cases_int_pointer3;
uint16_t _d[4];
uint16x4_t a = vld1_u16(_a);
uint16x4_t b = vld1_u16(_b);
uint16x8_t c = vld1q_u16(_c);
uint16x4_t d;

result_t test_vmla_laneq_u32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
#define TEST_IMPL(IDX) \
for (int i = 0; i < 4; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmla_laneq_u16(a, b, c, IDX); \
CHECK_RESULT(validate_uint16(d, _d[0], _d[1], _d[2], _d[3]))

result_t test_vmlaq_laneq_u32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
IMM_8_ITER
#undef TEST_IMPL

result_t test_vmla_laneq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
result_t test_vmlaq_laneq_u16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const uint16_t *_a = (const uint16_t *)impl.test_cases_int_pointer1;
const uint16_t *_b = (const uint16_t *)impl.test_cases_int_pointer2;
const uint16_t *_c = (const uint16_t *)impl.test_cases_int_pointer3;
uint16_t _d[8];
uint16x8_t a = vld1q_u16(_a);
uint16x8_t b = vld1q_u16(_b);
uint16x8_t c = vld1q_u16(_c);
uint16x8_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 8; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmlaq_laneq_u16(a, b, c, IDX); \
CHECK_RESULT(validate_uint16(d, _d[0], _d[1], _d[2], _d[3], _d[4], _d[5], _d[6], _d[7]))

IMM_8_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmla_laneq_u32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const uint32_t *_a = (const uint32_t *)impl.test_cases_int_pointer1;
const uint32_t *_b = (const uint32_t *)impl.test_cases_int_pointer2;
const uint32_t *_c = (const uint32_t *)impl.test_cases_int_pointer3;
uint32_t _d[2];
uint32x2_t a = vld1_u32(_a);
uint32x2_t b = vld1_u32(_b);
uint32x4_t c = vld1q_u32(_c);
uint32x2_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 2; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmla_laneq_u32(a, b, c, IDX); \
CHECK_RESULT(validate_uint32(d, _d[0], _d[1]))

IMM_4_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_u32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const uint32_t *_a = (const uint32_t *)impl.test_cases_int_pointer1;
const uint32_t *_b = (const uint32_t *)impl.test_cases_int_pointer2;
const uint32_t *_c = (const uint32_t *)impl.test_cases_int_pointer3;
uint32_t _d[4];
uint32x4_t a = vld1q_u32(_a);
uint32x4_t b = vld1q_u32(_b);
uint32x4_t c = vld1q_u32(_c);
uint32x4_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 4; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmlaq_laneq_u32(a, b, c, IDX); \
CHECK_RESULT(validate_uint32(d, _d[0], _d[1], _d[2], _d[3]))

IMM_4_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmla_laneq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const float *_a = (const float *)impl.test_cases_float_pointer1;
const float *_b = (const float *)impl.test_cases_float_pointer2;
const float *_c = (const float *)impl.test_cases_float_pointer3;
float _d[4];
float32x2_t a = vld1_f32(_a);
float32x2_t b = vld1_f32(_b);
float32x4_t c = vld1q_f32(_c);
float32x2_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 2; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmla_laneq_f32(a, b, c, IDX); \
CHECK_RESULT(validate_float_error(d, _d[0], _d[1], 0.0001f))

IMM_4_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_laneq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
const float *_a = (const float *)impl.test_cases_float_pointer1;
const float *_b = (const float *)impl.test_cases_float_pointer2;
const float *_c = (const float *)impl.test_cases_float_pointer3;
float _d[4];
float32x4_t a = vld1q_f32(_a);
float32x4_t b = vld1q_f32(_b);
float32x4_t c = vld1q_f32(_c);
float32x4_t d;

#define TEST_IMPL(IDX) \
for (int i = 0; i < 4; i++) { \
_d[i] = _a[i] + _b[i] * _c[IDX]; \
} \
d = vmlaq_laneq_f32(a, b, c, IDX); \
CHECK_RESULT(validate_float_error(d, _d[0], _d[1], _d[2], _d[3], 0.0001f))

IMM_4_ITER
#undef TEST_IMPL

return TEST_SUCCESS;
#else
return TEST_UNIMPL;
#endif // ENABLE_TEST_ALL
}

result_t test_vmlaq_lane_u16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
#ifdef ENABLE_TEST_ALL
Expand Down
20 changes: 10 additions & 10 deletions tests/impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1918,16 +1918,16 @@
_(vmlaq_lane_s16) \
_(vmlaq_lane_s32) \
_(vmlaq_lane_f32) \
/*_(vmla_laneq_s16) */ \
/*_(vmlaq_laneq_s16) */ \
/*_(vmla_laneq_s32) */ \
/*_(vmlaq_laneq_s32) */ \
/*_(vmla_laneq_u16) */ \
/*_(vmlaq_laneq_u16) */ \
/*_(vmla_laneq_u32) */ \
/*_(vmlaq_laneq_u32) */ \
/*_(vmla_laneq_f32) */ \
/*_(vmlaq_laneq_f32) */ \
_(vmla_laneq_s16) \
_(vmlaq_laneq_s16) \
_(vmla_laneq_s32) \
_(vmlaq_laneq_s32) \
_(vmla_laneq_u16) \
_(vmlaq_laneq_u16) \
_(vmla_laneq_u32) \
_(vmlaq_laneq_u32) \
_(vmla_laneq_f32) \
_(vmlaq_laneq_f32) \
_(vmlaq_lane_u16) \
_(vmlaq_lane_u32) \
_(vmlal_lane_s16) \
Expand Down
Loading