305 lines
16 KiB
C
305 lines
16 KiB
C
/* SPDX-License-Identifier: MIT
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person
|
|
* obtaining a copy of this software and associated documentation
|
|
* files (the "Software"), to deal in the Software without
|
|
* restriction, including without limitation the rights to use, copy,
|
|
* modify, merge, publish, distribute, sublicense, and/or sell copies
|
|
* of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be
|
|
* included in all copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
|
|
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
|
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
|
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
* SOFTWARE.
|
|
*
|
|
* Copyright:
|
|
* 2023 Chi-Wei Chu <wewe5215@gapp.nthu.edu.tw>
|
|
*/
|
|
|
|
#if !defined(SIMDE_ARM_NEON_CMLA_LANE_H)
|
|
#define SIMDE_ARM_NEON_CMLA_LANE_H
|
|
|
|
#include "add.h"
|
|
#include "combine.h"
|
|
#include "cvt.h"
|
|
#include "dup_lane.h"
|
|
#include "get_high.h"
|
|
#include "get_low.h"
|
|
#include "mul.h"
|
|
#include "types.h"
|
|
HEDLEY_DIAGNOSTIC_PUSH
|
|
SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
|
|
SIMDE_BEGIN_DECLS_
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float16x4_t simde_vcmla_lane_f16(simde_float16x4_t r, simde_float16x4_t a, simde_float16x4_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1)
|
|
{
|
|
simde_float32x4_private r_ = simde_float32x4_to_private(simde_vcvt_f32_f16(r)),
|
|
a_ = simde_float32x4_to_private(simde_vcvt_f32_f16(a)),
|
|
b_ = simde_float32x4_to_private(
|
|
simde_vcvt_f32_f16(simde_vdup_n_f16(simde_float16x4_to_private(b).values[lane])));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_) && \
|
|
((SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16) || (SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FLOAT16))
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_.values, a_.values, 0, 0, 2, 2);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmla_lane_f16
|
|
#define vcmla_lane_f16(r, a, b, lane) simde_vcmla_lane_f16(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmla_lane_f16(r, a, b, lane) vcmla_lane_f16(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float32x2_t simde_vcmla_lane_f32(simde_float32x2_t r, simde_float32x2_t a, simde_float32x2_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 0)
|
|
{
|
|
simde_float32x2_private r_ = simde_float32x2_to_private(r), a_ = simde_float32x2_to_private(a),
|
|
b_ = simde_float32x2_to_private(simde_vdup_n_f32(simde_float32x2_to_private(b).values[lane]));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_)
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(32, 8, a_.values, a_.values, 0, 0);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_float32x2_from_private(r_);
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmla_lane_f32
|
|
#define vcmla_lane_f32(r, a, b, lane) simde_vcmla_lane_f32(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmla_lane_f32(r, a, b, lane) vcmla_lane_f32(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float16x4_t simde_vcmla_laneq_f16(simde_float16x4_t r, simde_float16x4_t a, simde_float16x8_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1)
|
|
{
|
|
simde_float32x4_private r_ = simde_float32x4_to_private(simde_vcvt_f32_f16(r)),
|
|
a_ = simde_float32x4_to_private(simde_vcvt_f32_f16(a)),
|
|
b_ = simde_float32x4_to_private(
|
|
simde_vcvt_f32_f16(simde_vdup_n_f16(simde_float16x8_to_private(b).values[lane])));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_) && \
|
|
((SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16) || (SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FLOAT16))
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_.values, a_.values, 0, 0, 2, 2);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmla_laneq_f16
|
|
#define vcmla_laneq_f16(r, a, b, lane) simde_vcmla_laneq_f16(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmla_laneq_f16(r, a, b, lane) vcmla_laneq_f16(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float32x2_t simde_vcmla_laneq_f32(simde_float32x2_t r, simde_float32x2_t a, simde_float32x4_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1)
|
|
{
|
|
simde_float32x2_private r_ = simde_float32x2_to_private(r), a_ = simde_float32x2_to_private(a),
|
|
b_ = simde_float32x2_to_private(simde_vdup_n_f32(simde_float32x4_to_private(b).values[lane]));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_)
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(32, 8, a_.values, a_.values, 0, 0);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_float32x2_from_private(r_);
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmla_laneq_f32
|
|
#define vcmla_laneq_f32(r, a, b, lane) simde_vcmla_laneq_f32(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmla_laneq_f32(r, a, b, lane) vcmla_laneq_f32(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float16x8_t simde_vcmlaq_lane_f16(simde_float16x8_t r, simde_float16x8_t a, simde_float16x4_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1)
|
|
{
|
|
simde_float32x4_private r_low = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_low_f16(r))),
|
|
a_low = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_low_f16(a))),
|
|
r_high = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_high_f16(r))),
|
|
a_high = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_high_f16(a))),
|
|
b_ = simde_float32x4_to_private(
|
|
simde_vcvt_f32_f16(simde_vdup_n_f16(simde_float16x4_to_private(b).values[lane])));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_) && \
|
|
((SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16) || (SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FLOAT16))
|
|
a_low.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_low.values, a_low.values, 0, 0, 2, 2);
|
|
a_high.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_high.values, a_high.values, 0, 0, 2, 2);
|
|
r_low.values += b_.values * a_low.values;
|
|
r_high.values += b_.values * a_high.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_low.values) / (2 * sizeof(r_low.values[0]))); i++)
|
|
{
|
|
r_low.values[2 * i] += b_.values[lane] * a_low.values[2 * i];
|
|
r_low.values[2 * i + 1] += b_.values[lane] * a_low.values[2 * i];
|
|
r_high.values[2 * i] += b_.values[lane] * a_high.values[2 * i];
|
|
r_high.values[2 * i + 1] += b_.values[lane] * a_high.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_vcombine_f16(simde_vcvt_f16_f32(simde_float32x4_from_private(r_low)),
|
|
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmlaq_lane_f16
|
|
#define vcmlaq_lane_f16(r, a, b, lane) simde_vcmlaq_lane_f16(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmlaq_lane_f16(r, a, b, lane) vcmlaq_lane_f16(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float32x4_t simde_vcmlaq_lane_f32(simde_float32x4_t r, simde_float32x4_t a, simde_float32x2_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 0)
|
|
{
|
|
simde_float32x4_private r_ = simde_float32x4_to_private(r), a_ = simde_float32x4_to_private(a),
|
|
b_ = simde_float32x4_to_private(simde_vdupq_n_f32(simde_float32x2_to_private(b).values[lane]));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_)
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(32, 16, a_.values, a_.values, 0, 0, 2, 2);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_float32x4_from_private(r_);
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmlaq_lane_f32
|
|
#define vcmlaq_lane_f32(r, a, b, lane) simde_vcmlaq_lane_f32(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmlaq_lane_f32(r, a, b, lane) vcmlaq_lane_f32(r, a, b, 0);
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float16x8_t simde_vcmlaq_laneq_f16(simde_float16x8_t r, simde_float16x8_t a, simde_float16x8_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 3)
|
|
{
|
|
simde_float32x4_private r_low = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_low_f16(r))),
|
|
a_low = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_low_f16(a))),
|
|
r_high = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_high_f16(r))),
|
|
a_high = simde_float32x4_to_private(simde_vcvt_f32_f16(simde_vget_high_f16(a))),
|
|
b_ = simde_float32x4_to_private(
|
|
simde_vcvt_f32_f16(simde_vdup_n_f16(simde_float16x8_to_private(b).values[lane])));
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_) && \
|
|
((SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16) || (SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FLOAT16))
|
|
a_low.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_low.values, a_low.values, 0, 0, 2, 2);
|
|
r_low.values += b_.values * a_low.values;
|
|
a_high.values = SIMDE_SHUFFLE_VECTOR_(16, 4, a_high.values, a_high.values, 0, 0, 2, 2);
|
|
r_high.values += b_.values * a_high.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_low.values) / (2 * sizeof(r_low.values[0]))); i++)
|
|
{
|
|
r_low.values[2 * i] += b_.values[lane] * a_low.values[2 * i];
|
|
r_low.values[2 * i + 1] += b_.values[lane] * a_low.values[2 * i];
|
|
r_high.values[2 * i] += b_.values[lane] * a_high.values[2 * i];
|
|
r_high.values[2 * i + 1] += b_.values[lane] * a_high.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_vcombine_f16(simde_vcvt_f16_f32(simde_float32x4_from_private(r_low)),
|
|
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmlaq_laneq_f16
|
|
#define vcmlaq_laneq_f16(r, a, b, lane) simde_vcmlaq_laneq_f16(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmlaq_laneq_f16(r, a, b, lane) vcmlaq_laneq_f16(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_FUNCTION_ATTRIBUTES
|
|
simde_float32x4_t simde_vcmlaq_laneq_f32(simde_float32x4_t r, simde_float32x4_t a, simde_float32x4_t b, const int lane)
|
|
SIMDE_REQUIRE_CONSTANT_RANGE(lane, 0, 1)
|
|
{
|
|
simde_float32x4_private r_ = simde_float32x4_to_private(r), a_ = simde_float32x4_to_private(a),
|
|
b_ = simde_float32x4_to_private(simde_vdupq_n_f32(simde_float32x4_to_private(b).values[lane]));
|
|
|
|
#if defined(SIMDE_SHUFFLE_VECTOR_)
|
|
a_.values = SIMDE_SHUFFLE_VECTOR_(32, 16, a_.values, a_.values, 0, 0, 2, 2);
|
|
r_.values += b_.values * a_.values;
|
|
#else
|
|
SIMDE_VECTORIZE
|
|
for (size_t i = 0; i < (sizeof(r_.values) / (2 * sizeof(r_.values[0]))); i++)
|
|
{
|
|
r_.values[2 * i] += b_.values[lane] * a_.values[2 * i];
|
|
r_.values[2 * i + 1] += b_.values[lane] * a_.values[2 * i];
|
|
}
|
|
#endif
|
|
return simde_float32x4_from_private(r_);
|
|
}
|
|
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES)
|
|
#undef vcmlaq_laneq_f32
|
|
#define vcmlaq_laneq_f32(r, a, b, lane) simde_vcmlaq_laneq_f32(r, a, b, lane)
|
|
#endif
|
|
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
|
|
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
|
|
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
|
|
#define simde_vcmlaq_laneq_f32(r, a, b, lane) vcmlaq_laneq_f32(r, a, b, lane)
|
|
#endif
|
|
|
|
SIMDE_END_DECLS_
|
|
HEDLEY_DIAGNOSTIC_POP
|
|
|
|
#endif /* !defined(SIMDE_ARM_NEON_CMLA_LANE_H) */
|