aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp113
1 files changed, 61 insertions, 52 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp
index 5434599f03..b6565baa23 100644
--- a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp
+++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_24way_16bit.hpp
@@ -31,91 +31,100 @@
template <>
template <typename T>
inline void TransformImpl<12, 1, true, 4, 4>::Transform(
- T *out, const T *const in, const int stride,
- const int x0, const int xmax, const int k0, const int kmax)
-{
- // Redirect to a 24 x uint16_t specialisation
- TransformImpl<24, 1, true, 2, 2>::Transform(
- reinterpret_cast<uint16_t *>(out),
- reinterpret_cast<const uint16_t *const>(in),
- stride * 2, x0 * 2, xmax * 2, k0, kmax);
+ T* out, const T* const in, const int stride,
+ const int x0, const int xmax, const int k0, const int kmax
+) {
+ // Redirect to a 24 x uint16_t specialisation
+ TransformImpl<24, 1, true, 2, 2>::Transform(
+ reinterpret_cast<uint16_t *>(out),
+ reinterpret_cast<const uint16_t * const>(in),
+ stride*2, x0*2, xmax*2, k0, kmax
+ );
}
// Generic 24x16-bit sized specialisation
template <>
template <typename T>
inline void TransformImpl<24, 1, true, 2, 2>::Transform(
- T *out, const T *const in, const int stride,
- const int x0, const int xmax, const int k0, const int kmax)
-{
- // Redirect to a uint16_t specialisation
- Transform(
- reinterpret_cast<uint16_t *>(out),
- reinterpret_cast<const uint16_t *const>(in),
- stride, x0, xmax, k0, kmax);
+ T* out, const T* const in, const int stride,
+ const int x0, const int xmax, const int k0, const int kmax
+) {
+ // Redirect to a uint16_t specialisation
+ Transform(
+ reinterpret_cast<uint16_t *>(out),
+ reinterpret_cast<const uint16_t * const>(in),
+ stride, x0, xmax, k0, kmax
+ );
}
// Specialised 24 x uint16_t version
template <>
-inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x1(const uint16_t *&in0, uint16_t *out)
-{
- __asm __volatile(
+inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x1(const uint16_t *&in0, uint16_t *out) {
+ __asm __volatile (
"LDP q0, q1, [%[in0]], #32\n"
- "STP q0, q1, [%[out]]\n" ASM_PREFETCH("[%[in0], #192]")
+ "STP q0, q1, [%[out]]\n"
+ ASM_PREFETCH("[%[in0], #192]")
"LDR q2, [%[in0]], #16\n"
"STR q2, [%[out], #32]\n"
- : [in0] "+r"(in0), [out] "+r"(out)
- :
- : "v0", "v1", "v2", "memory");
+ : [in0] "+r" (in0), [out] "+r" (out)
+ :
+ : "v0", "v1", "v2", "memory"
+ );
}
template <>
-inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x2(const uint16_t *&in0, const uint16_t *&in1, uint16_t *out)
-{
- __asm __volatile(
+inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x2(const uint16_t *&in0, const uint16_t *&in1,uint16_t *out) {
+ __asm __volatile (
"LDP q0, q1, [%[in0]], #32\n"
- "STP q0, q1, [%[out]]\n" ASM_PREFETCH("[%[in0], #192]")
+ "STP q0, q1, [%[out]]\n"
+ ASM_PREFETCH("[%[in0], #192]")
"LDR q2, [%[in0]], #16\n"
- "LDP q3, q4, [%[in1]], #32\n"
- "STP q2, q3, [%[out], #32]\n" ASM_PREFETCH("[%[in1], #192]")
- "LDR q5, [%[in1]], #16\n"
+ "LDP q3, q4, [%[in1]], #32\n"
+ "STP q2, q3, [%[out], #32]\n"
+ ASM_PREFETCH("[%[in1], #192]")
+ "LDR q5, [%[in1]], #16\n"
"STP q4, q5, [%[out], #64]\n"
- : [in0] "+r"(in0), [in1] "+r"(in1), [out] "+r"(out)
- :
- : "v0", "v1", "v2", "v3", "v4", "v5", "memory");
+ : [in0] "+r" (in0), [in1] "+r" (in1), [out] "+r" (out)
+ :
+ : "v0", "v1", "v2", "v3", "v4", "v5", "memory"
+ );
}
template <>
-inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x4(const uint16_t *&in0, const uint16_t *&in1, const uint16_t *&in2, const uint16_t *&in3, uint16_t *out)
-{
- __asm __volatile(
+inline void TransposeInterleaveCommon<24, uint16_t, uint16_t>::moveblock_1x4(const uint16_t *&in0, const uint16_t *&in1, const uint16_t *&in2, const uint16_t *&in3, uint16_t *out) {
+ __asm __volatile (
"LDP q0, q1, [%[in0]], #32\n"
"STP q0, q1, [%[out]]\n"
- "LDR q2, [%[in0]], #16\n" ASM_PREFETCH("[%[in0], #192]")
- "LDP q3, q4, [%[in1]], #32\n"
+ "LDR q2, [%[in0]], #16\n"
+ ASM_PREFETCH("[%[in0], #192]")
+ "LDP q3, q4, [%[in1]], #32\n"
"STP q2, q3, [%[out], #32]\n"
- "LDR q5, [%[in1]], #16\n" ASM_PREFETCH("[%[in1], #192]")
+ "LDR q5, [%[in1]], #16\n"
+ ASM_PREFETCH("[%[in1], #192]")
"STP q4, q5, [%[out], #64]\n"
- "LDP q6, q7, [%[in2]], #32\n"
+ "LDP q6, q7, [%[in2]], #32\n"
"STP q6, q7, [%[out], #96]\n"
- "LDR q8, [%[in2]], #16\n" ASM_PREFETCH("[%[in2], #192]")
- "LDP q9, q10, [%[in3]], #32\n"
+ "LDR q8, [%[in2]], #16\n"
+ ASM_PREFETCH("[%[in2], #192]")
+ "LDP q9, q10, [%[in3]], #32\n"
"STP q8, q9, [%[out], #128]\n"
- "LDR q11, [%[in3]], #16\n"
- "STP q10, q11, [%[out], #160]\n" ASM_PREFETCH("[%[in3], #192]")
+ "LDR q11, [%[in3]], #16\n"
+ "STP q10, q11, [%[out], #160]\n"
+ ASM_PREFETCH("[%[in3], #192]")
- : [in0] "+r"(in0), [in1] "+r"(in1), [in2] "+r"(in2), [in3] "+r"(in3), [out] "+r"(out)
- :
- : "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "memory");
+ : [in0] "+r" (in0), [in1] "+r" (in1), [in2] "+r" (in2), [in3] "+r" (in3), [out] "+r" (out)
+ :
+ : "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "memory"
+ );
}
template <>
template <>
inline void TransformImpl<24, 1, true, 2, 2>::Transform(
- uint16_t *out, const uint16_t *const in, const int stride,
- const int x0, const int xmax, const int k0, const int kmax)
-{
- TransposeInterleaveCommon<24, uint16_t, uint16_t>::Transform(out, in, stride, x0, xmax, k0, kmax);
+ uint16_t* out, const uint16_t* const in, const int stride,
+ const int x0, const int xmax, const int k0, const int kmax
+) {
+ TransposeInterleaveCommon<24, uint16_t, uint16_t>::Transform(out, in, stride, x0, xmax, k0, kmax);
}
-#endif // __arch64__
+#endif // __arch64__
generated by cgit v1.2.1 (git 2.23.0) at 2024-06-02 10:36:35 +0000