mirror of
https://github.com/signalwire/freeswitch.git
synced 2026-11-18 07:18:34 +00:00
276 lines
12 KiB
C++
276 lines
12 KiB
C++
/*
|
|
* Copyright 2026 The LibYuv Project Authors. All rights reserved.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license
|
|
* that can be found in the LICENSE file in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#include "libyuv/rotate_row.h"
|
|
#include "libyuv/row.h"
|
|
|
|
// This module is for RVV (RISC-V Vector extension)
|
|
#if !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector)
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
#ifndef RVV_VL_CLOBBER
|
|
#ifdef __clang__
|
|
#define RVV_VL_CLOBBER
|
|
#else
|
|
#define RVV_VL_CLOBBER "vl",
|
|
#endif
|
|
#endif
|
|
|
|
#ifdef HAS_TRANSPOSEWX4_32_RVV
|
|
// Transpose 32 bit values (ARGB) in Wx4 tiles, moving down W rows of source
|
|
// (VL at a time) and across W columns of destination (4 rows).
|
|
// Tuned for 512-bit VL (SiFive X280).
|
|
void TransposeWx4_32_RVV(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width) {
|
|
int vl;
|
|
ptrdiff_t src_step;
|
|
ptrdiff_t dst_step;
|
|
uint8_t* dst1 = dst + (ptrdiff_t)dst_stride;
|
|
uint8_t* dst2 = dst1 + (ptrdiff_t)dst_stride;
|
|
uint8_t* dst3 = dst2 + (ptrdiff_t)dst_stride;
|
|
asm volatile(
|
|
"vsetvli %[vl], %[w], e32, m2, ta, ma \n"
|
|
"slli %[dst_step], %[vl], 2 \n"
|
|
"mul %[src_step], %[vl], %[src_stride] \n"
|
|
"sub %[w], %[w], %[vl] \n"
|
|
"bltz %[w], 2f \n"
|
|
"1: \n"
|
|
"vlsseg4e32.v v8, (%[src]), %[src_stride] \n"
|
|
"add %[src], %[src], %[src_step] \n"
|
|
"vse32.v v8, (%[dst0]) \n"
|
|
"vse32.v v10, (%[dst1]) \n"
|
|
"vse32.v v12, (%[dst2]) \n"
|
|
"vse32.v v14, (%[dst3]) \n"
|
|
"add %[dst0], %[dst0], %[dst_step] \n"
|
|
"add %[dst1], %[dst1], %[dst_step] \n"
|
|
"add %[dst2], %[dst2], %[dst_step] \n"
|
|
"add %[dst3], %[dst3], %[dst_step] \n"
|
|
"sub %[w], %[w], %[vl] \n"
|
|
"bgez %[w], 1b \n"
|
|
"2: \n"
|
|
"add %[w], %[w], %[vl] \n"
|
|
"beqz %[w], 3f \n"
|
|
"vsetvli %[vl], %[w], e32, m2, ta, ma \n"
|
|
"vlsseg4e32.v v8, (%[src]), %[src_stride] \n"
|
|
"vse32.v v8, (%[dst0]) \n"
|
|
"vse32.v v10, (%[dst1]) \n"
|
|
"vse32.v v12, (%[dst2]) \n"
|
|
"vse32.v v14, (%[dst3]) \n"
|
|
"3: \n"
|
|
: [w] "+r"(width), // %[w]
|
|
[src] "+r"(src), // %[src]
|
|
[dst0] "+r"(dst), // %[dst0]
|
|
[dst1] "+r"(dst1), // %[dst1]
|
|
[dst2] "+r"(dst2), // %[dst2]
|
|
[dst3] "+r"(dst3), // %[dst3]
|
|
[vl] "=&r"(vl), // %[vl]
|
|
[src_step] "=&r"(src_step), // %[src_step]
|
|
[dst_step] "=&r"(dst_step) // %[dst_step]
|
|
: [src_stride] "r"((ptrdiff_t)src_stride) // %[src_stride]
|
|
: RVV_VL_CLOBBER "vtype", "memory", "v8", "v9", "v10", "v11", "v12",
|
|
"v13", "v14", "v15");
|
|
}
|
|
#endif // HAS_TRANSPOSEWX4_32_RVV
|
|
|
|
#ifdef HAS_TRANSPOSEWXH_32_RVV
|
|
// Transpose 32 bit plane of any width x height using RVV without scalar
|
|
// fallback.
|
|
void TransposeWxH_32_RVV(const uint8_t* src,
|
|
int src_stride,
|
|
uint8_t* dst,
|
|
int dst_stride,
|
|
int width,
|
|
int height) {
|
|
while (width >= 4) {
|
|
TransposeWx4_32_RVV(src, src_stride, dst, dst_stride, height);
|
|
src += 16;
|
|
dst += 4 * (ptrdiff_t)dst_stride;
|
|
width -= 4;
|
|
}
|
|
while (width > 0) {
|
|
const uint8_t* s_col = src;
|
|
uint8_t* d_row = dst;
|
|
int h = height;
|
|
int vl;
|
|
ptrdiff_t src_step;
|
|
ptrdiff_t dst_step;
|
|
asm volatile(
|
|
"vsetvli %[vl], %[h], e32, m8, ta, ma \n"
|
|
"slli %[d_step], %[vl], 2 \n"
|
|
"mul %[s_step], %[vl], %[src_stride] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bltz %[h], 2f \n"
|
|
"1: \n"
|
|
"vlse32.v v8, (%[s]), %[src_stride] \n"
|
|
"add %[s], %[s], %[s_step] \n"
|
|
"vse32.v v8, (%[d]) \n"
|
|
"add %[d], %[d], %[d_step] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bgez %[h], 1b \n"
|
|
"2: \n"
|
|
"add %[h], %[h], %[vl] \n"
|
|
"beqz %[h], 3f \n"
|
|
"vsetvli %[vl], %[h], e32, m8, ta, ma \n"
|
|
"vlse32.v v8, (%[s]), %[src_stride] \n"
|
|
"vse32.v v8, (%[d]) \n"
|
|
"3: \n"
|
|
: [h] "+r"(h), // %[h]
|
|
[s] "+r"(s_col), // %[s]
|
|
[d] "+r"(d_row), // %[d]
|
|
[vl] "=&r"(vl), // %[vl]
|
|
[s_step] "=&r"(src_step), // %[s_step]
|
|
[d_step] "=&r"(dst_step) // %[d_step]
|
|
: [src_stride] "r"((ptrdiff_t)src_stride) // %[src_stride]
|
|
: RVV_VL_CLOBBER "vtype", "memory", "v8", "v9", "v10", "v11", "v12",
|
|
"v13", "v14", "v15");
|
|
src += 4;
|
|
dst += (ptrdiff_t)dst_stride;
|
|
width -= 1;
|
|
}
|
|
}
|
|
#endif // HAS_TRANSPOSEWXH_32_RVV
|
|
|
|
#ifdef HAS_TRANSPOSEWXH_16_RVV
|
|
void TransposeWxH_16_RVV(const uint16_t* src,
|
|
int src_stride,
|
|
uint16_t* dst,
|
|
int dst_stride,
|
|
int width,
|
|
int height) {
|
|
const ptrdiff_t s_stride = (ptrdiff_t)src_stride * 2;
|
|
const ptrdiff_t d_stride = (ptrdiff_t)dst_stride * 2;
|
|
while (width >= 4) {
|
|
const uint16_t* s0 = src;
|
|
const uint16_t* s1 = src + 1;
|
|
const uint16_t* s2 = src + 2;
|
|
const uint16_t* s3 = src + 3;
|
|
uint8_t* d0 = (uint8_t*)dst;
|
|
uint8_t* d1 = d0 + d_stride;
|
|
uint8_t* d2 = d1 + d_stride;
|
|
uint8_t* d3 = d2 + d_stride;
|
|
int h = height;
|
|
int vl;
|
|
ptrdiff_t src_step;
|
|
ptrdiff_t dst_step;
|
|
asm volatile(
|
|
"vsetvli %[vl], %[h], e16, m8, ta, ma \n"
|
|
"slli %[d_step], %[vl], 1 \n"
|
|
"mul %[s_step], %[vl], %[s_stride] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bltz %[h], 2f \n"
|
|
"1: \n"
|
|
"vlse16.v v0, (%[s0]), %[s_stride] \n"
|
|
"vlse16.v v8, (%[s1]), %[s_stride] \n"
|
|
"vlse16.v v16, (%[s2]), %[s_stride] \n"
|
|
"vlse16.v v24, (%[s3]), %[s_stride] \n"
|
|
"add %[s0], %[s0], %[s_step] \n"
|
|
"add %[s1], %[s1], %[s_step] \n"
|
|
"add %[s2], %[s2], %[s_step] \n"
|
|
"add %[s3], %[s3], %[s_step] \n"
|
|
"vse16.v v0, (%[d0]) \n"
|
|
"vse16.v v8, (%[d1]) \n"
|
|
"vse16.v v16, (%[d2]) \n"
|
|
"vse16.v v24, (%[d3]) \n"
|
|
"add %[d0], %[d0], %[d_step] \n"
|
|
"add %[d1], %[d1], %[d_step] \n"
|
|
"add %[d2], %[d2], %[d_step] \n"
|
|
"add %[d3], %[d3], %[d_step] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bgez %[h], 1b \n"
|
|
"2: \n"
|
|
"add %[h], %[h], %[vl] \n"
|
|
"beqz %[h], 3f \n"
|
|
"vsetvli %[vl], %[h], e16, m8, ta, ma \n"
|
|
"vlse16.v v0, (%[s0]), %[s_stride] \n"
|
|
"vlse16.v v8, (%[s1]), %[s_stride] \n"
|
|
"vlse16.v v16, (%[s2]), %[s_stride] \n"
|
|
"vlse16.v v24, (%[s3]), %[s_stride] \n"
|
|
"vse16.v v0, (%[d0]) \n"
|
|
"vse16.v v8, (%[d1]) \n"
|
|
"vse16.v v16, (%[d2]) \n"
|
|
"vse16.v v24, (%[d3]) \n"
|
|
"3: \n"
|
|
: [h] "+r"(h), // %[h]
|
|
[s0] "+r"(s0), // %[s0]
|
|
[s1] "+r"(s1), // %[s1]
|
|
[s2] "+r"(s2), // %[s2]
|
|
[s3] "+r"(s3), // %[s3]
|
|
[d0] "+r"(d0), // %[d0]
|
|
[d1] "+r"(d1), // %[d1]
|
|
[d2] "+r"(d2), // %[d2]
|
|
[d3] "+r"(d3), // %[d3]
|
|
[vl] "=&r"(vl), // %[vl]
|
|
[s_step] "=&r"(src_step), // %[s_step]
|
|
[d_step] "=&r"(dst_step) // %[d_step]
|
|
: [s_stride] "r"(s_stride) // %[s_stride]
|
|
: RVV_VL_CLOBBER "vtype", "memory", "v0", "v1", "v2", "v3", "v4", "v5",
|
|
"v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15",
|
|
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
|
|
"v26", "v27", "v28", "v29", "v30", "v31");
|
|
src += 4;
|
|
dst += 4 * (ptrdiff_t)dst_stride;
|
|
width -= 4;
|
|
}
|
|
while (width > 0) {
|
|
const uint16_t* s_col = src;
|
|
uint16_t* d_row = dst;
|
|
int h = height;
|
|
int vl;
|
|
ptrdiff_t src_step;
|
|
ptrdiff_t dst_step;
|
|
asm volatile(
|
|
"vsetvli %[vl], %[h], e16, m8, ta, ma \n"
|
|
"slli %[d_step], %[vl], 1 \n"
|
|
"mul %[s_step], %[vl], %[s_stride] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bltz %[h], 2f \n"
|
|
"1: \n"
|
|
"vlse16.v v8, (%[s]), %[s_stride] \n"
|
|
"add %[s], %[s], %[s_step] \n"
|
|
"vse16.v v8, (%[d]) \n"
|
|
"add %[d], %[d], %[d_step] \n"
|
|
"sub %[h], %[h], %[vl] \n"
|
|
"bgez %[h], 1b \n"
|
|
"2: \n"
|
|
"add %[h], %[h], %[vl] \n"
|
|
"beqz %[h], 3f \n"
|
|
"vsetvli %[vl], %[h], e16, m8, ta, ma \n"
|
|
"vlse16.v v8, (%[s]), %[s_stride] \n"
|
|
"vse16.v v8, (%[d]) \n"
|
|
"3: \n"
|
|
: [h] "+r"(h), // %[h]
|
|
[s] "+r"(s_col), // %[s]
|
|
[d] "+r"(d_row), // %[d]
|
|
[vl] "=&r"(vl), // %[vl]
|
|
[s_step] "=&r"(src_step), // %[s_step]
|
|
[d_step] "=&r"(dst_step) // %[d_step]
|
|
: [s_stride] "r"(s_stride) // %[s_stride]
|
|
: RVV_VL_CLOBBER "vtype", "memory", "v8", "v9", "v10", "v11", "v12",
|
|
"v13", "v14", "v15");
|
|
src += 1;
|
|
dst += (ptrdiff_t)dst_stride;
|
|
width -= 1;
|
|
}
|
|
}
|
|
#endif // HAS_TRANSPOSEWXH_16_RVV
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|
|
|
|
#endif // !defined(LIBYUV_DISABLE_RVV) && defined(__riscv_vector)
|