mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2026-10-28 18:07:12 +08:00
38e21134f7
Add two new planar/row compositing functions for 32-bit ARGB/BGRA
scanlines:
- ARGBOr / ARGBOrRow_*: Applies a bitwise OR with a 32-bit ARGB mask
to every pixel (e.g., 0xFF000000 to force opaque alpha for BGRx to
BGRA conversion). Implemented for C, SSE2, AVX2, AVX-512BW, NEON
(32/64-bit), SVE2, SME, RVV, WASMSIMD, LSX, LASX, HVX, and Windows
intrinsics (row_win.cc).
- ARGBAlphaBlend / ARGBAlphaBlendRow_*: Implements straight-alpha
(unattenuated) Porter-Duff source-over blending matching PDFium's
CompositeRowBgra2Bgra:
dst_a = back_a + src_a - (back_a * src_a) / 255
ratio = (src_a * 255) / dst_a
dst_c = (back_c * (255 - ratio) + src_c * ratio) / 255
Implemented with bit-exact division by 255 across all 65,536
(back_a, src_a) pairs for C, SSSE3, AVX2, AVX-512BW, NEON
(32/64-bit), SVE2, SME, RVV, WASMSIMD, LSX, LASX, HVX, and Windows
intrinsics (row_win.cc).
ARGBOr Performance (2048x64 BGRx->BGRA scanlines, us per call):
uArch Before Highway libyuv vs Hwy
------------------------ ------ ------- ------- ------
Qualcomm Hexagon v81 HVX 259.63 260.97 15.78 16.54x
SiFive X280 (RVV) 1107.57 72.49 50.70 1.43x
SiFive X390 (RVV) 1108.82 70.42 52.26 1.35x
ARGBAlphaBlend Performance (2048x64 BGRA scanlines, us per call):
uArch Before Highway libyuv vs Hwy
------------------------ ------ ------- ------- ------
Intel Lion Cove (AVX2) 429.79 141.97 41.89 3.39x
Intel Cougar Cove (AVX2) 516.83 181.98 51.33 3.55x
Intel Golden Cove (AVX2) 560.31 206.28 67.08 3.08x
AMD Zen 3 (AVX2) 1025.73 251.09 98.99 2.54x
Wasm SIMD128 (V8) 1014.14 964.83 385.80 2.50x
Apple M6 Super (NEON) 221.90 118.42 46.35 2.55x
Apple M5 Super (NEON) 241.90 130.07 49.60 2.62x
ARM C1-Ultra (SME2) 384.63 112.52 63.62 1.77x
Qualcomm Oryon V3-L (SME) 388.13 154.87 87.65 1.77x
Qualcomm Oryon-L (NEON) 318.28 149.36 57.23 2.61x
ARM Cortex-X4 (SVE2) 675.47 213.56 102.53 2.08x
ARM Cortex-A720 (NEON) 795.85 586.55 235.81 2.49x
ARM Cortex-A520 (NEON) 2360.95 2122.99 787.65 2.70x
ARM Cortex-A35 (ARMv7) 4907.31 5895.04 1474.51 4.00x
SiFive X280 (RVV) 4145.06 1096.42 596.82 1.84x
SiFive X390 (RVV) 4144.57 1084.93 597.62 1.82x
Qualcomm Hexagon v81 HVX 4912.03 5786.25 952.19 6.08x
Coral V4 RVV (1024 px) 50276 c 48992 c 4299 c 11.40x
R=wtc@google.com
Bug: none
Change-Id: If661239826a033e89d4b5c1adc0bfdf048a7a726
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8647895
Reviewed-by: richard winterton <rrwinterton@gmail.com>
3306 lines
120 KiB
C++
3306 lines
120 KiB
C++
/*
|
|
* Copyright 2022 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/compare_row.h"
|
|
#include "libyuv/row.h"
|
|
|
|
// Note: every register written by an inline asm block (vector, general
|
|
// purpose, mask/predicate, "cc" for flags and "memory" for stores) must be
|
|
// listed in its clobber list, even when the ABI treats it as caller-saved.
|
|
// With LTO/LTCG the compiler may keep values live in any register across the
|
|
// asm statement, and other OS ABIs (e.g. Windows) differ in which registers
|
|
// are callee-saved. List "memory" first, then "cc", then registers.
|
|
|
|
#if !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)
|
|
|
|
#include "libyuv/basic_types.h"
|
|
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
void CopyRow_LASX(const uint8_t* src, uint8_t* dst, int count) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src), [dst] "+r"(dst), [cnt] "+r"(count)
|
|
:
|
|
: "memory", "$xr0");
|
|
}
|
|
|
|
void SetRow_LASX(uint8_t* dst, uint8_t v8, int count) {
|
|
asm volatile(
|
|
"xvreplgr2vr.b $xr0, %[val]\n\t"
|
|
"1:\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [dst] "+r"(dst), [cnt] "+r"(count)
|
|
: [val] "r"(v8)
|
|
: "memory", "$xr0");
|
|
}
|
|
|
|
void ARGBSetRow_LASX(uint8_t* dst_argb, uint32_t v32, int width) {
|
|
asm volatile(
|
|
"xvreplgr2vr.w $xr0, %[val]\n\t"
|
|
"1:\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [val] "r"(v32)
|
|
: "memory", "$xr0");
|
|
}
|
|
|
|
void MirrorRow_LASX(const uint8_t* src_y, uint8_t* dst_y, int width) {
|
|
static const uint8_t kRev[16] = {15, 14, 13, 12, 11, 10, 9, 8,
|
|
7, 6, 5, 4, 3, 2, 1, 0};
|
|
src_y += width - 32;
|
|
asm volatile(
|
|
"vld $vr1, %[rev], 0\n\t"
|
|
"xvpermi.q $xr1, $xr1, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 0\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], -32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_y), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
: [rev] "r"(kRev)
|
|
: "memory", "$xr0", "$xr1", "$xr31");
|
|
}
|
|
|
|
void MirrorUVRow_LASX(const uint8_t* src_uv, uint8_t* dst_uv, int width) {
|
|
static const uint8_t kRevUV[16] = {14, 15, 12, 13, 10, 11, 8, 9,
|
|
6, 7, 4, 5, 2, 3, 0, 1};
|
|
src_uv += width * 2 - 32;
|
|
asm volatile(
|
|
"vld $vr1, %[rev], 0\n\t"
|
|
"xvpermi.q $xr1, $xr1, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 0\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], -32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_uv), [dst] "+r"(dst_uv), [cnt] "+r"(width)
|
|
: [rev] "r"(kRevUV)
|
|
: "memory", "$xr0", "$xr1", "$xr31");
|
|
}
|
|
|
|
void ARGBMirrorRow_LASX(const uint8_t* src_argb, uint8_t* dst_argb, int width) {
|
|
src_argb += width * 4 - 32;
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 0\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvshuf4i.w $xr0, $xr0, 0x1b\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], -32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr31");
|
|
}
|
|
|
|
void MirrorSplitUVRow_LASX(const uint8_t* src_uv,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kRevUV[16] = {14, 15, 12, 13, 10, 11, 8, 9,
|
|
6, 7, 4, 5, 2, 3, 0, 1};
|
|
src_uv += width * 2 - 64;
|
|
asm volatile(
|
|
"vld $vr4, %[rev], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 48\n\t"
|
|
"vld $vr31, %[src], 16\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 32\n\t"
|
|
"vld $vr31, %[src], 0\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr4\n\t"
|
|
"xvshuf.b $xr1, $xr1, $xr1, $xr4\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvst $xr2, %[du], 0\n\t"
|
|
"xvst $xr3, %[dv], 0\n\t"
|
|
"addi.d %[src], %[src], -64\n\t"
|
|
"addi.d %[du], %[du], 32\n\t"
|
|
"addi.d %[dv], %[dv], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
:
|
|
[src] "+r"(src_uv), [du] "+r"(dst_u), [dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
: [rev] "r"(kRevUV)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void SplitUVRow_LASX(const uint8_t* src_uv,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvst $xr2, %[du], 0\n\t"
|
|
"xvst $xr3, %[dv], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[du], %[du], 32\n\t"
|
|
"addi.d %[dv], %[dv], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
:
|
|
[src] "+r"(src_uv), [du] "+r"(dst_u), [dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr31");
|
|
}
|
|
|
|
void MergeUVRow_LASX(const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_uv,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[su], 0\n\t"
|
|
"xvld $xr1, %[sv], 0\n\t"
|
|
"xvilvl.b $xr2, $xr1, $xr0\n\t"
|
|
"xvilvh.b $xr3, $xr1, $xr0\n\t"
|
|
"vst $vr2, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr3, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[su], %[su], 32\n\t"
|
|
"addi.d %[sv], %[sv], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
:
|
|
[su] "+r"(src_u), [sv] "+r"(src_v), [dst] "+r"(dst_uv), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr31");
|
|
}
|
|
|
|
void ARGBExtractAlphaRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_a,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickod.b $xr0, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr0, $xr0, $xr0\n\t"
|
|
"xvpermi.d $xr31, $xr0, 0xd8\n\t"
|
|
"vst $vr31, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_a), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr31");
|
|
}
|
|
|
|
void I422ToYUY2Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_yuy2,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[sy], 0\n\t"
|
|
"vld $vr1, %[su], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"vld $vr2, %[sv], 0\n\t"
|
|
"xvpermi.d $xr2, $xr2, 0xd8\n\t"
|
|
"xvilvl.b $xr3, $xr2, $xr1\n\t"
|
|
"xvilvl.b $xr4, $xr3, $xr0\n\t"
|
|
"xvilvh.b $xr5, $xr3, $xr0\n\t"
|
|
"vst $vr4, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr4, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr5, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr5, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 32\n\t"
|
|
"addi.d %[su], %[su], 16\n\t"
|
|
"addi.d %[sv], %[sv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_yuy2), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr31");
|
|
}
|
|
|
|
void I422ToUYVYRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_uyvy,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[sy], 0\n\t"
|
|
"vld $vr1, %[su], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"vld $vr2, %[sv], 0\n\t"
|
|
"xvpermi.d $xr2, $xr2, 0xd8\n\t"
|
|
"xvilvl.b $xr3, $xr2, $xr1\n\t"
|
|
"xvilvl.b $xr4, $xr0, $xr3\n\t"
|
|
"xvilvh.b $xr5, $xr0, $xr3\n\t"
|
|
"vst $vr4, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr4, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr5, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr5, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 32\n\t"
|
|
"addi.d %[su], %[su], 16\n\t"
|
|
"addi.d %[sv], %[sv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_uyvy), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr31");
|
|
}
|
|
|
|
void YUY2ToYRow_LASX(const uint8_t* src_yuy2, uint8_t* dst_y, int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_yuy2), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr31");
|
|
}
|
|
|
|
void UYVYToYRow_LASX(const uint8_t* src_uyvy, uint8_t* dst_y, int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickod.b $xr2, $xr1, $xr0\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_uyvy), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr31");
|
|
}
|
|
|
|
void YUY2ToUV422Row_LASX(const uint8_t* src_yuy2,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickod.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickev.b $xr3, $xr2, $xr2\n\t"
|
|
"xvpickod.b $xr4, $xr2, $xr2\n\t"
|
|
"xvpermi.d $xr31, $xr3, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"xvpermi.d $xr31, $xr4, 0xd8\n\t"
|
|
"vst $vr31, %[dv], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_yuy2), [du] "+r"(dst_u), [dv] "+r"(dst_v),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void UYVYToUV422Row_LASX(const uint8_t* src_uyvy,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickev.b $xr3, $xr2, $xr2\n\t"
|
|
"xvpickod.b $xr4, $xr2, $xr2\n\t"
|
|
"xvpermi.d $xr31, $xr3, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"xvpermi.d $xr31, $xr4, 0xd8\n\t"
|
|
"vst $vr31, %[dv], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_uyvy), [du] "+r"(dst_u), [dv] "+r"(dst_v),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void YUY2ToUVRow_LASX(const uint8_t* src_yuy2,
|
|
int src_stride_yuy2,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
const uint8_t* next_yuy2 = src_yuy2 + src_stride_yuy2;
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[s0], 16\n\t"
|
|
"vld $vr31, %[s0], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"vld $vr2, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 32\n\t"
|
|
"xvpermi.q $xr2, $xr31, 0x02\n\t"
|
|
"vld $vr3, %[s1], 16\n\t"
|
|
"vld $vr31, %[s1], 48\n\t"
|
|
"xvpermi.q $xr3, $xr31, 0x02\n\t"
|
|
"xvpickod.b $xr0, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr2, $xr3, $xr2\n\t"
|
|
"xvavgr.bu $xr0, $xr0, $xr2\n\t"
|
|
"xvpickev.b $xr1, $xr0, $xr0\n\t"
|
|
"xvpickod.b $xr2, $xr0, $xr0\n\t"
|
|
"xvpermi.d $xr31, $xr1, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"xvpermi.d $xr31, $xr2, 0xd8\n\t"
|
|
"vst $vr31, %[dv], 0\n\t"
|
|
"addi.d %[s0], %[s0], 64\n\t"
|
|
"addi.d %[s1], %[s1], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_yuy2), [s1] "+r"(next_yuy2), [du] "+r"(dst_u),
|
|
[dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr31");
|
|
}
|
|
|
|
void UYVYToUVRow_LASX(const uint8_t* src_uyvy,
|
|
int src_stride_uyvy,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
const uint8_t* next_uyvy = src_uyvy + src_stride_uyvy;
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[s0], 16\n\t"
|
|
"vld $vr31, %[s0], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"vld $vr2, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 32\n\t"
|
|
"xvpermi.q $xr2, $xr31, 0x02\n\t"
|
|
"vld $vr3, %[s1], 16\n\t"
|
|
"vld $vr31, %[s1], 48\n\t"
|
|
"xvpermi.q $xr3, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr0, $xr1, $xr0\n\t"
|
|
"xvpickev.b $xr2, $xr3, $xr2\n\t"
|
|
"xvavgr.bu $xr0, $xr0, $xr2\n\t"
|
|
"xvpickev.b $xr1, $xr0, $xr0\n\t"
|
|
"xvpickod.b $xr2, $xr0, $xr0\n\t"
|
|
"xvpermi.d $xr31, $xr1, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"xvpermi.d $xr31, $xr2, 0xd8\n\t"
|
|
"vst $vr31, %[dv], 0\n\t"
|
|
"addi.d %[s0], %[s0], 64\n\t"
|
|
"addi.d %[s1], %[s1], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_uyvy), [s1] "+r"(next_uyvy), [du] "+r"(dst_u),
|
|
[dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr31");
|
|
}
|
|
|
|
void I422ToARGBRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr20", "$xr21", "$xr22",
|
|
"$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422ToRGBARow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr2, $xr5\n\t"
|
|
"xvilvl.b $xr7, $xr4, $xr3\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr20", "$xr21", "$xr22",
|
|
"$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422AlphaToARGBRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
const uint8_t* src_a,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"vld $vr5, %[sa], 0\n\t"
|
|
"xvpermi.d $xr5, $xr5, 0xd8\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[sa], %[sa], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width), [sa] "+r"(src_a)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr20", "$xr21", "$xr22",
|
|
"$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422ToRGB24Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
static const uint8_t kShuf0[16] = {0, 1, 2, 4, 5, 6, 8, 9,
|
|
10, 12, 13, 14, 16, 17, 18, 20};
|
|
static const uint8_t kShuf1[16] = {21, 22, 24, 25, 26, 28, 29, 30,
|
|
0, 0, 0, 0, 0, 0, 0, 0};
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"vld $vr12, %[s0], 0\n\t"
|
|
"xvpermi.q $xr12, $xr12, 0x00\n\t"
|
|
"vld $vr13, %[s1], 0\n\t"
|
|
"xvpermi.q $xr13, $xr13, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr4, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"xvshuf.b $xr10, $xr9, $xr8, $xr12\n\t"
|
|
"xvshuf.b $xr11, $xr9, $xr8, $xr13\n\t"
|
|
"vst $vr10, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr10, 0x01\n\t"
|
|
"vst $vr31, %[dst], 24\n\t"
|
|
"vstelm.d $vr11, %[dst], 16, 0\n\t"
|
|
"xvpermi.q $xr31, $xr11, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dst], 40, 0\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 48\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants), [s0] "r"(kShuf0), [s1] "r"(kShuf1)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr6", "$xr7",
|
|
"$xr8", "$xr9", "$xr10", "$xr11", "$xr12", "$xr13", "$xr20", "$xr21",
|
|
"$xr22", "$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422ToRGB565Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr14, 31\n\t"
|
|
"xvsrli.h $xr14, $xr14, 8\n\t"
|
|
"xvldi $xr15, 63\n\t"
|
|
"xvsrli.h $xr15, $xr15, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvsrli.b $xr3, $xr3, 2\n\t"
|
|
"xvsrli.b $xr4, $xr4, 3\n\t"
|
|
"xvilvl.b $xr2, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr4, $xr4, $xr4\n\t"
|
|
"xvand.v $xr2, $xr2, $xr14\n\t"
|
|
"xvand.v $xr3, $xr3, $xr15\n\t"
|
|
"xvand.v $xr4, $xr4, $xr14\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr4, $xr4, 11\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvor.v $xr2, $xr2, $xr4\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr10", "$xr11",
|
|
"$xr14", "$xr15", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422ToARGB4444Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr14, 15\n\t"
|
|
"xvsrli.h $xr14, $xr14, 8\n\t"
|
|
"xvldi $xr15, 240\n\t"
|
|
"xvsrli.h $xr15, $xr15, 8\n\t"
|
|
"xvslli.h $xr15, $xr15, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvsrli.b $xr2, $xr2, 4\n\t"
|
|
"xvsrli.b $xr3, $xr3, 4\n\t"
|
|
"xvsrli.b $xr4, $xr4, 4\n\t"
|
|
"xvilvl.b $xr2, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr4, $xr4, $xr4\n\t"
|
|
"xvand.v $xr2, $xr2, $xr14\n\t"
|
|
"xvand.v $xr3, $xr3, $xr14\n\t"
|
|
"xvand.v $xr4, $xr4, $xr14\n\t"
|
|
"xvslli.h $xr3, $xr3, 4\n\t"
|
|
"xvslli.h $xr4, $xr4, 8\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvor.v $xr2, $xr2, $xr4\n\t"
|
|
"xvor.v $xr2, $xr2, $xr15\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr10", "$xr11",
|
|
"$xr14", "$xr15", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I422ToARGB1555Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr14, 31\n\t"
|
|
"xvsrli.h $xr14, $xr14, 8\n\t"
|
|
"xvldi $xr15, 128\n\t"
|
|
"xvsrli.h $xr15, $xr15, 8\n\t"
|
|
"xvslli.h $xr15, $xr15, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvldrepl.d $xr10, %[su], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr10, 0x01\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"xvldrepl.d $xr11, %[sv], 0\n\t"
|
|
"xvshuf4i.w $xr31, $xr11, 0x01\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvilvl.b $xr10, $xr10, $xr10\n\t"
|
|
"xvilvl.b $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvsrli.b $xr3, $xr3, 3\n\t"
|
|
"xvsrli.b $xr4, $xr4, 3\n\t"
|
|
"xvilvl.b $xr2, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr4, $xr4, $xr4\n\t"
|
|
"xvand.v $xr2, $xr2, $xr14\n\t"
|
|
"xvand.v $xr3, $xr3, $xr14\n\t"
|
|
"xvand.v $xr4, $xr4, $xr14\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr4, $xr4, 10\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvor.v $xr2, $xr2, $xr4\n\t"
|
|
"xvor.v $xr2, $xr2, $xr15\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 8\n\t"
|
|
"addi.d %[sv], %[sv], 8\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr10", "$xr11",
|
|
"$xr14", "$xr15", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void NV12ToARGBRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_uv,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr1, %[suv], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"xvilvl.h $xr1, $xr1, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[suv], %[suv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [suv] "+r"(src_uv), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24",
|
|
"$xr31");
|
|
}
|
|
|
|
void NV21ToARGBRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_vu,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr1, %[suv], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"xvshuf4i.b $xr1, $xr1, 0xb1\n\t"
|
|
"xvilvl.h $xr1, $xr1, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[suv], %[suv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [suv] "+r"(src_vu), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24",
|
|
"$xr31");
|
|
}
|
|
|
|
void NV12ToRGB565Row_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_uv,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr14, 31\n\t"
|
|
"xvsrli.h $xr14, $xr14, 8\n\t"
|
|
"xvldi $xr15, 63\n\t"
|
|
"xvsrli.h $xr15, $xr15, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr1, %[suv], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"xvilvl.h $xr1, $xr1, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvsrli.b $xr3, $xr3, 2\n\t"
|
|
"xvsrli.b $xr4, $xr4, 3\n\t"
|
|
"xvilvl.b $xr2, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr4, $xr4, $xr4\n\t"
|
|
"xvand.v $xr2, $xr2, $xr14\n\t"
|
|
"xvand.v $xr3, $xr3, $xr15\n\t"
|
|
"xvand.v $xr4, $xr4, $xr14\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr4, $xr4, 11\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvor.v $xr2, $xr2, $xr4\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[suv], %[suv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [suv] "+r"(src_uv), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr14", "$xr15",
|
|
"$xr20", "$xr21", "$xr22", "$xr23", "$xr24");
|
|
}
|
|
|
|
void I444ToARGBRow_LASX(const uint8_t* src_y,
|
|
const uint8_t* src_u,
|
|
const uint8_t* src_v,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr10, %[su], 0\n\t"
|
|
"xvpermi.d $xr10, $xr10, 0xd8\n\t"
|
|
"vld $vr11, %[sv], 0\n\t"
|
|
"xvpermi.d $xr11, $xr11, 0xd8\n\t"
|
|
"xvilvl.b $xr1, $xr11, $xr10\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[su], %[su], 16\n\t"
|
|
"addi.d %[sv], %[sv], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [su] "+r"(src_u), [sv] "+r"(src_v),
|
|
[dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr20", "$xr21", "$xr22",
|
|
"$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void I400ToARGBRow_LASX(const uint8_t* src_y,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"vld $vr23, %[c], 96\n\t"
|
|
"xvpermi.q $xr23, $xr23, 0x00\n\t"
|
|
"vld $vr24, %[c], 128\n\t"
|
|
"xvpermi.q $xr24, $xr24, 0x00\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sy], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvssrani.bu.h $xr2, $xr0, 6\n\t"
|
|
"xvilvl.b $xr6, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr2\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr2", "$xr5", "$xr6", "$xr7", "$xr8", "$xr9",
|
|
"$xr23", "$xr24", "$xr31");
|
|
}
|
|
|
|
void J400ToARGBRow_LASX(const uint8_t* src_y, uint8_t* dst_argb, int width) {
|
|
asm volatile(
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr2, %[sy], 0\n\t"
|
|
"xvpermi.d $xr2, $xr2, 0xd8\n\t"
|
|
"xvilvl.b $xr6, $xr2, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr2\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sy] "+r"(src_y), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr2", "$xr5", "$xr6", "$xr7", "$xr8", "$xr9", "$xr31");
|
|
}
|
|
|
|
void YUY2ToARGBRow_LASX(const uint8_t* src_yuy2,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr10, %[src], 0\n\t"
|
|
"xvpickev.b $xr0, $xr10, $xr10\n\t"
|
|
"xvpickod.b $xr1, $xr10, $xr10\n\t"
|
|
"xvilvl.h $xr1, $xr1, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_yuy2), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr20", "$xr21", "$xr22", "$xr23",
|
|
"$xr24", "$xr31");
|
|
}
|
|
|
|
void UYVYToARGBRow_LASX(const uint8_t* src_uyvy,
|
|
uint8_t* dst_argb,
|
|
const struct YuvConstants* yuvconstants,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldrepl.w $xr20, %[c], 0\n\t"
|
|
"xvldrepl.w $xr21, %[c], 32\n\t"
|
|
"xvldrepl.w $xr22, %[c], 64\n\t"
|
|
"xvldrepl.h $xr23, %[c], 96\n\t"
|
|
"xvldrepl.h $xr24, %[c], 128\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr10, %[src], 0\n\t"
|
|
"xvpickod.b $xr0, $xr10, $xr10\n\t"
|
|
"xvpickev.b $xr1, $xr10, $xr10\n\t"
|
|
"xvilvl.h $xr1, $xr1, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr0, $xr0\n\t"
|
|
"xvxori.b $xr1, $xr1, 0x80\n\t"
|
|
"xvmuh.hu $xr0, $xr0, $xr23\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr24\n\t"
|
|
"xvmulwev.h.bu.b $xr2, $xr20, $xr1\n\t"
|
|
"xvmulwod.h.bu.b $xr4, $xr22, $xr1\n\t"
|
|
"xvmulwev.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvmaddwod.h.bu.b $xr3, $xr21, $xr1\n\t"
|
|
"xvsadd.h $xr2, $xr0, $xr2\n\t"
|
|
"xvsadd.h $xr4, $xr0, $xr4\n\t"
|
|
"xvssub.h $xr3, $xr0, $xr3\n\t"
|
|
"xvssrani.bu.h $xr2, $xr2, 6\n\t"
|
|
"xvssrani.bu.h $xr3, $xr3, 6\n\t"
|
|
"xvssrani.bu.h $xr4, $xr4, 6\n\t"
|
|
"xvilvl.b $xr6, $xr3, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr5, $xr4\n\t"
|
|
"xvilvl.h $xr8, $xr7, $xr6\n\t"
|
|
"xvilvh.h $xr9, $xr7, $xr6\n\t"
|
|
"vst $vr8, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr8, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr9, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr9, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_uyvy), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [c] "r"(yuvconstants)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr20", "$xr21", "$xr22", "$xr23",
|
|
"$xr24", "$xr31");
|
|
}
|
|
|
|
static LIBYUV_NOINLINE void ARGBTo24Helper_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_24,
|
|
const uint8_t* shuf,
|
|
int width) {
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 64\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 80\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"vld $vr2, %[src], 32\n\t"
|
|
"vld $vr31, %[src], 96\n\t"
|
|
"xvpermi.q $xr2, $xr31, 0x02\n\t"
|
|
"vld $vr3, %[src], 48\n\t"
|
|
"vld $vr31, %[src], 112\n\t"
|
|
"xvpermi.q $xr3, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr4\n\t"
|
|
"xvshuf.b $xr1, $xr1, $xr1, $xr4\n\t"
|
|
"xvshuf.b $xr2, $xr2, $xr2, $xr4\n\t"
|
|
"xvshuf.b $xr3, $xr3, $xr3, $xr4\n\t"
|
|
"vstelm.d $vr0, %[dst], 0, 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dst], 48, 0\n\t"
|
|
"vstelm.w $vr0, %[dst], 8, 2\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 56, 2\n\t"
|
|
"vstelm.w $vr1, %[dst], 12, 0\n\t"
|
|
"vstelm.w $vr1, %[dst], 16, 1\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 60, 0\n\t"
|
|
"vstelm.w $vr31, %[dst], 64, 1\n\t"
|
|
"vstelm.w $vr1, %[dst], 20, 2\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 68, 2\n\t"
|
|
"vstelm.d $vr2, %[dst], 24, 0\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dst], 72, 0\n\t"
|
|
"vstelm.w $vr2, %[dst], 32, 2\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 80, 2\n\t"
|
|
"vstelm.w $vr3, %[dst], 36, 0\n\t"
|
|
"vstelm.w $vr3, %[dst], 40, 1\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 84, 0\n\t"
|
|
"vstelm.w $vr31, %[dst], 88, 1\n\t"
|
|
"vstelm.w $vr3, %[dst], 44, 2\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 92, 2\n\t"
|
|
"addi.d %[src], %[src], 128\n\t"
|
|
"addi.d %[dst], %[dst], 96\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_24), [cnt] "+r"(width)
|
|
: [shuf] "r"(shuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void ARGBToRGB24Row_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_rgb24,
|
|
int width) {
|
|
static const uint8_t kShuf[16] = {0, 1, 2, 4, 5, 6, 8, 9,
|
|
10, 12, 13, 14, 0, 0, 0, 0};
|
|
ARGBTo24Helper_LASX(src_argb, dst_rgb24, kShuf, width);
|
|
}
|
|
|
|
void ARGBToRAWRow_LASX(const uint8_t* src_argb, uint8_t* dst_raw, int width) {
|
|
static const uint8_t kShuf[16] = {2, 1, 0, 6, 5, 4, 10, 9,
|
|
8, 14, 13, 12, 0, 0, 0, 0};
|
|
ARGBTo24Helper_LASX(src_argb, dst_raw, kShuf, width);
|
|
}
|
|
|
|
static LIBYUV_NOINLINE void RGB24ToARGBHelper_LASX(const uint8_t* src_24,
|
|
uint8_t* dst_argb,
|
|
const uint8_t* shuf,
|
|
int width) {
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"vld $vr5, %[shuf], 16\n\t"
|
|
"xvpermi.q $xr5, $xr5, 0x00\n\t"
|
|
"xvldi $xr6, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 12\n\t"
|
|
"vld $vr31, %[src], 60\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"vld $vr2, %[src], 24\n\t"
|
|
"vld $vr31, %[src], 72\n\t"
|
|
"xvpermi.q $xr2, $xr31, 0x02\n\t"
|
|
"vld $vr3, %[src], 32\n\t"
|
|
"vld $vr31, %[src], 80\n\t"
|
|
"xvpermi.q $xr3, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr6, $xr0, $xr4\n\t"
|
|
"xvshuf.b $xr1, $xr6, $xr1, $xr4\n\t"
|
|
"xvshuf.b $xr2, $xr6, $xr2, $xr4\n\t"
|
|
"xvshuf.b $xr3, $xr6, $xr3, $xr5\n\t"
|
|
"vst $vr0, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vst $vr31, %[dst], 64\n\t"
|
|
"vst $vr1, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vst $vr31, %[dst], 80\n\t"
|
|
"vst $vr2, %[dst], 32\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vst $vr31, %[dst], 96\n\t"
|
|
"vst $vr3, %[dst], 48\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vst $vr31, %[dst], 112\n\t"
|
|
"addi.d %[src], %[src], 96\n\t"
|
|
"addi.d %[dst], %[dst], 128\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_24), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [shuf] "r"(shuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr31");
|
|
}
|
|
|
|
void RGB24ToARGBRow_LASX(const uint8_t* src_rgb24,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
static const uint8_t kShuf[32] = {0, 1, 2, 16, 3, 4, 5, 16, 6, 7, 8,
|
|
16, 9, 10, 11, 16, 4, 5, 6, 16, 7, 8,
|
|
9, 16, 10, 11, 12, 16, 13, 14, 15, 16};
|
|
RGB24ToARGBHelper_LASX(src_rgb24, dst_argb, kShuf, width);
|
|
}
|
|
|
|
void RAWToARGBRow_LASX(const uint8_t* src_raw, uint8_t* dst_argb, int width) {
|
|
static const uint8_t kShuf[32] = {2, 1, 0, 16, 5, 4, 3, 16, 8, 7, 6,
|
|
16, 11, 10, 9, 16, 6, 5, 4, 16, 9, 8,
|
|
7, 16, 12, 11, 10, 16, 15, 14, 13, 16};
|
|
RGB24ToARGBHelper_LASX(src_raw, dst_argb, kShuf, width);
|
|
}
|
|
|
|
void RAWToRGB24Row_LASX(const uint8_t* src_raw, uint8_t* dst_rgb24, int width) {
|
|
static const uint8_t kShuf[32] = {2, 1, 0, 5, 4, 3, 8, 7, 6, 11, 10,
|
|
9, 0, 0, 0, 0, 6, 5, 4, 9, 8, 7,
|
|
12, 11, 10, 15, 14, 13, 0, 0, 0, 0};
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"vld $vr5, %[shuf], 16\n\t"
|
|
"xvpermi.q $xr5, $xr5, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 12\n\t"
|
|
"vld $vr31, %[src], 60\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"vld $vr2, %[src], 24\n\t"
|
|
"vld $vr31, %[src], 72\n\t"
|
|
"xvpermi.q $xr2, $xr31, 0x02\n\t"
|
|
"vld $vr3, %[src], 32\n\t"
|
|
"vld $vr31, %[src], 80\n\t"
|
|
"xvpermi.q $xr3, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr4\n\t"
|
|
"xvshuf.b $xr1, $xr1, $xr1, $xr4\n\t"
|
|
"xvshuf.b $xr2, $xr2, $xr2, $xr4\n\t"
|
|
"xvshuf.b $xr3, $xr3, $xr3, $xr5\n\t"
|
|
"vstelm.d $vr0, %[dst], 0, 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dst], 48, 0\n\t"
|
|
"vstelm.w $vr0, %[dst], 8, 2\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 56, 2\n\t"
|
|
"vstelm.w $vr1, %[dst], 12, 0\n\t"
|
|
"vstelm.w $vr1, %[dst], 16, 1\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 60, 0\n\t"
|
|
"vstelm.w $vr31, %[dst], 64, 1\n\t"
|
|
"vstelm.w $vr1, %[dst], 20, 2\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 68, 2\n\t"
|
|
"vstelm.d $vr2, %[dst], 24, 0\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dst], 72, 0\n\t"
|
|
"vstelm.w $vr2, %[dst], 32, 2\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 80, 2\n\t"
|
|
"vstelm.w $vr3, %[dst], 36, 0\n\t"
|
|
"vstelm.w $vr3, %[dst], 40, 1\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 84, 0\n\t"
|
|
"vstelm.w $vr31, %[dst], 88, 1\n\t"
|
|
"vstelm.w $vr3, %[dst], 44, 2\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vstelm.w $vr31, %[dst], 92, 2\n\t"
|
|
"addi.d %[src], %[src], 96\n\t"
|
|
"addi.d %[dst], %[dst], 96\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_raw), [dst] "+r"(dst_rgb24), [cnt] "+r"(width)
|
|
: [shuf] "r"(kShuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr31");
|
|
}
|
|
|
|
void ARGBToRGB565Row_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_rgb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvsrli.b $xr3, $xr3, 2\n\t"
|
|
"xvand.v $xr4, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr2, 8\n\t"
|
|
"xvand.v $xr3, $xr3, $xr7\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr2, $xr2, 11\n\t"
|
|
"xvor.v $xr4, $xr4, $xr3\n\t"
|
|
"xvor.v $xr4, $xr4, $xr2\n\t"
|
|
"xvst $xr4, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_rgb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr6", "$xr7",
|
|
"$xr31");
|
|
}
|
|
|
|
void ARGBToRGB565DitherRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_rgb,
|
|
uint32_t dither4,
|
|
int width) {
|
|
asm volatile(
|
|
"xvreplgr2vr.w $xr5, %[dith]\n\t"
|
|
"xvilvl.b $xr5, $xr5, $xr5\n\t"
|
|
"xvilvl.h $xr5, $xr5, $xr5\n\t"
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvsadd.bu $xr0, $xr0, $xr5\n\t"
|
|
"xvpickev.b $xr2, $xr0, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr0, $xr0\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvsrli.b $xr3, $xr3, 2\n\t"
|
|
"xvand.v $xr4, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr2, 8\n\t"
|
|
"xvand.v $xr3, $xr3, $xr7\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr2, $xr2, 11\n\t"
|
|
"xvor.v $xr4, $xr4, $xr3\n\t"
|
|
"xvor.v $xr4, $xr4, $xr2\n\t"
|
|
"xvpermi.d $xr31, $xr4, 0xd8\n\t"
|
|
"vst $vr31, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_rgb), [cnt] "+r"(width)
|
|
: [dith] "r"(dither4)
|
|
: "memory", "$xr0", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6", "$xr7",
|
|
"$xr31");
|
|
}
|
|
|
|
void ARGBToARGB1555Row_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_rgb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvsrli.b $xr2, $xr2, 3\n\t"
|
|
"xvand.v $xr4, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr2, 8\n\t"
|
|
"xvsrli.h $xr5, $xr3, 15\n\t"
|
|
"xvsrli.b $xr3, $xr3, 3\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvslli.h $xr3, $xr3, 5\n\t"
|
|
"xvslli.h $xr2, $xr2, 10\n\t"
|
|
"xvslli.h $xr5, $xr5, 15\n\t"
|
|
"xvor.v $xr4, $xr4, $xr3\n\t"
|
|
"xvor.v $xr4, $xr4, $xr2\n\t"
|
|
"xvor.v $xr4, $xr4, $xr5\n\t"
|
|
"xvst $xr4, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_rgb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr31");
|
|
}
|
|
|
|
void ARGBToARGB4444Row_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_rgb,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvsrli.b $xr2, $xr2, 4\n\t"
|
|
"xvsrli.b $xr3, $xr3, 4\n\t"
|
|
"xvslli.b $xr3, $xr3, 4\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_rgb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr31");
|
|
}
|
|
|
|
void RGB565ToARGBRow_LASX(const uint8_t* src_rgb565,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr1, $xr1, $xr1\n\t"
|
|
"xvpickev.b $xr2, $xr2, $xr2\n\t"
|
|
"xvpickev.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr8, $xr2, $xr1\n\t"
|
|
"xvilvl.b $xr9, $xr5, $xr3\n\t"
|
|
"xvilvl.h $xr10, $xr9, $xr8\n\t"
|
|
"xvilvh.h $xr11, $xr9, $xr8\n\t"
|
|
"vst $vr10, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr10, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr11, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr11, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_rgb565), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr31");
|
|
}
|
|
|
|
void ARGB1555ToARGBRow_LASX(const uint8_t* src_argb1555,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr1, $xr1, $xr1\n\t"
|
|
"xvpickev.b $xr2, $xr2, $xr2\n\t"
|
|
"xvpickev.b $xr3, $xr3, $xr3\n\t"
|
|
"xvpickev.b $xr5, $xr5, $xr5\n\t"
|
|
"xvilvl.b $xr8, $xr2, $xr1\n\t"
|
|
"xvilvl.b $xr9, $xr5, $xr3\n\t"
|
|
"xvilvl.h $xr10, $xr9, $xr8\n\t"
|
|
"xvilvh.h $xr11, $xr9, $xr8\n\t"
|
|
"vst $vr10, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr10, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr11, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr11, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb1555), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr8", "$xr9", "$xr10", "$xr11", "$xr31");
|
|
}
|
|
|
|
void ARGB4444ToARGBRow_LASX(const uint8_t* src_argb4444,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 0x0f\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.b $xr2, $xr0, 4\n\t"
|
|
"xvslli.b $xr3, $xr1, 4\n\t"
|
|
"xvor.v $xr1, $xr1, $xr3\n\t"
|
|
"xvslli.b $xr3, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvilvl.b $xr4, $xr2, $xr1\n\t"
|
|
"xvilvh.b $xr5, $xr2, $xr1\n\t"
|
|
"vst $vr4, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr4, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr5, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr5, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb4444), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr31");
|
|
}
|
|
|
|
void ARGBMultiplyRow_LASX(const uint8_t* src_argb,
|
|
const uint8_t* src_argb1,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvmulwev.h.bu $xr2, $xr0, $xr1\n\t"
|
|
"xvmulwod.h.bu $xr3, $xr0, $xr1\n\t"
|
|
"xvsrari.h $xr2, $xr2, 8\n\t"
|
|
"xvsrari.h $xr3, $xr3, 8\n\t"
|
|
"xvpackev.b $xr2, $xr3, $xr2\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb), [s1] "+r"(src_argb1), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3");
|
|
}
|
|
|
|
void ARGBAddRow_LASX(const uint8_t* src_argb,
|
|
const uint8_t* src_argb1,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvsadd.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb), [s1] "+r"(src_argb1), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1");
|
|
}
|
|
|
|
void ARGBSubtractRow_LASX(const uint8_t* src_argb,
|
|
const uint8_t* src_argb1,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvssub.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb), [s1] "+r"(src_argb1), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1");
|
|
}
|
|
|
|
void ARGBAttenuateRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
static const uint8_t kAlpha[16] = {3, 3, 3, 16, 7, 7, 7, 16,
|
|
11, 11, 11, 16, 15, 15, 15, 16};
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"xvldi $xr6, 255\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvshuf.b $xr1, $xr5, $xr0, $xr4\n\t"
|
|
"xvmulwev.h.bu $xr2, $xr0, $xr1\n\t"
|
|
"xvmulwod.h.bu $xr3, $xr0, $xr1\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr6\n\t"
|
|
"xvadd.h $xr3, $xr3, $xr6\n\t"
|
|
"xvpackod.b $xr2, $xr3, $xr2\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [shuf] "r"(kAlpha)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6");
|
|
}
|
|
|
|
void ARGBBlendRow_LASX(const uint8_t* src_argb0,
|
|
const uint8_t* src_argb1,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
static const uint8_t kAlpha[16] = {3, 3, 3, 3, 7, 7, 7, 7,
|
|
11, 11, 11, 11, 15, 15, 15, 15};
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"xvldi $xr5, 0xff\n\t"
|
|
"xvslli.w $xr5, $xr5, 24\n\t"
|
|
"xvldi $xr6, 1\n\t"
|
|
"xvslli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr8, 0\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvshuf.b $xr2, $xr0, $xr0, $xr4\n\t"
|
|
"xvilvl.b $xr3, $xr8, $xr2\n\t"
|
|
"xvilvh.b $xr2, $xr8, $xr2\n\t"
|
|
"xvsub.h $xr3, $xr6, $xr3\n\t"
|
|
"xvsub.h $xr2, $xr6, $xr2\n\t"
|
|
"xvilvl.b $xr7, $xr8, $xr1\n\t"
|
|
"xvilvh.b $xr1, $xr8, $xr1\n\t"
|
|
"xvmul.h $xr7, $xr7, $xr3\n\t"
|
|
"xvmul.h $xr1, $xr1, $xr2\n\t"
|
|
"xvsrlni.b.h $xr1, $xr7, 8\n\t"
|
|
"xvsadd.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvor.v $xr0, $xr0, $xr5\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb0), [s1] "+r"(src_argb1), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
: [shuf] "r"(kAlpha)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8");
|
|
}
|
|
|
|
void ARGBOrRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_argb,
|
|
uint32_t value,
|
|
int width) {
|
|
asm volatile(
|
|
"xvreplgr2vr.w $xr1, %[val]\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvor.v $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [val] "r"(value)
|
|
: "memory", "$xr0", "$xr1");
|
|
}
|
|
|
|
void ARGBAlphaBlendRow_LASX(const uint8_t* src_argb0,
|
|
const uint8_t* src_argb1,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
static const uint8_t kShufRatio[16] = {0, 0, 0, 0, 4, 4, 4, 4,
|
|
8, 8, 8, 8, 12, 12, 12, 12};
|
|
asm volatile(
|
|
"vld $vr8, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr8, $xr8, 0x00\n\t"
|
|
"xvldi $xr9, 1\n\t"
|
|
"xvsrli.h $xr10, $xr9, 8\n\t"
|
|
"xvldi $xr11, 0xff\n\t"
|
|
"xvsrli.w $xr11, $xr11, 8\n\t"
|
|
"xvldi $xr12, 0\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvsrli.w $xr2, $xr0, 24\n\t"
|
|
"xvsrli.w $xr3, $xr1, 24\n\t"
|
|
"xvmul.h $xr4, $xr3, $xr2\n\t"
|
|
"xvadd.h $xr4, $xr4, $xr10\n\t"
|
|
"xvmuh.hu $xr4, $xr4, $xr9\n\t"
|
|
"xvsub.w $xr4, $xr2, $xr4\n\t"
|
|
"xvadd.w $xr4, $xr3, $xr4\n\t"
|
|
"xvseq.w $xr5, $xr3, $xr12\n\t"
|
|
"xvsrli.w $xr5, $xr5, 24\n\t"
|
|
"xvor.v $xr6, $xr2, $xr5\n\t"
|
|
"xvor.v $xr7, $xr4, $xr5\n\t"
|
|
"xvslli.w $xr5, $xr6, 8\n\t"
|
|
"xvsub.w $xr5, $xr5, $xr6\n\t"
|
|
"xvffint.s.wu $xr5, $xr5\n\t"
|
|
"xvffint.s.wu $xr7, $xr7\n\t"
|
|
"xvfdiv.s $xr5, $xr5, $xr7\n\t"
|
|
"xvftintrz.wu.s $xr5, $xr5\n\t"
|
|
"xvshuf.b $xr5, $xr5, $xr5, $xr8\n\t"
|
|
"xvnor.v $xr6, $xr5, $xr5\n\t"
|
|
"xvmulwev.h.bu $xr2, $xr1, $xr6\n\t"
|
|
"xvmulwod.h.bu $xr3, $xr1, $xr6\n\t"
|
|
"xvmulwev.h.bu $xr7, $xr0, $xr5\n\t"
|
|
"xvmulwod.h.bu $xr0, $xr0, $xr5\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr7\n\t"
|
|
"xvadd.h $xr3, $xr3, $xr0\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr10\n\t"
|
|
"xvadd.h $xr3, $xr3, $xr10\n\t"
|
|
"xvmuh.hu $xr2, $xr2, $xr9\n\t"
|
|
"xvmuh.hu $xr3, $xr3, $xr9\n\t"
|
|
"xvslli.h $xr3, $xr3, 8\n\t"
|
|
"xvor.v $xr2, $xr2, $xr3\n\t"
|
|
"xvand.v $xr2, $xr2, $xr11\n\t"
|
|
"xvslli.w $xr4, $xr4, 24\n\t"
|
|
"xvor.v $xr2, $xr2, $xr4\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb0), [s1] "+r"(src_argb1), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
: [shuf] "r"(kShufRatio)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr12");
|
|
}
|
|
|
|
void ARGBShadeRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_argb,
|
|
int width,
|
|
uint32_t value) {
|
|
asm volatile(
|
|
"xvreplgr2vr.w $xr4, %[val]\n\t"
|
|
"xvilvl.b $xr5, $xr4, $xr4\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvilvl.b $xr1, $xr0, $xr0\n\t"
|
|
"xvilvh.b $xr2, $xr0, $xr0\n\t"
|
|
"xvmuh.hu $xr1, $xr1, $xr5\n\t"
|
|
"xvmuh.hu $xr2, $xr2, $xr5\n\t"
|
|
"xvsrlni.b.h $xr2, $xr1, 8\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [val] "r"(value)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr4", "$xr5");
|
|
}
|
|
|
|
void ARGBShuffleRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_argb,
|
|
const uint8_t* shuffler,
|
|
int width) {
|
|
asm volatile(
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvshuf.b $xr0, $xr0, $xr0, $xr4\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [shuf] "r"(shuffler)
|
|
: "memory", "$xr0", "$xr4");
|
|
}
|
|
|
|
void ARGBGrayRow_LASX(const uint8_t* src_argb, uint8_t* dst_argb, int width) {
|
|
asm volatile(
|
|
"xvldi $xr4, 29\n\t"
|
|
"xvldi $xr5, 150\n\t"
|
|
"xvldi $xr6, 77\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvmulwev.h.bu $xr7, $xr2, $xr4\n\t"
|
|
"xvmaddwod.h.bu $xr7, $xr2, $xr6\n\t"
|
|
"xvmaddwev.h.bu $xr7, $xr3, $xr5\n\t"
|
|
"xvsrlri.h $xr7, $xr7, 8\n\t"
|
|
"xvsrli.h $xr3, $xr3, 8\n\t"
|
|
"xvslli.h $xr3, $xr3, 8\n\t"
|
|
"xvor.v $xr3, $xr7, $xr3\n\t"
|
|
"xvslli.h $xr8, $xr7, 8\n\t"
|
|
"xvor.v $xr7, $xr7, $xr8\n\t"
|
|
"xvilvl.h $xr0, $xr3, $xr7\n\t"
|
|
"xvilvh.h $xr1, $xr3, $xr7\n\t"
|
|
"vst $vr0, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr1, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr31");
|
|
}
|
|
|
|
void ARGBSepiaRow_LASX(uint8_t* dst_argb, int width) {
|
|
asm volatile(
|
|
"xvldi $xr4, 17\n\t"
|
|
"xvldi $xr5, 68\n\t"
|
|
"xvldi $xr6, 35\n\t"
|
|
"xvldi $xr7, 22\n\t"
|
|
"xvldi $xr8, 88\n\t"
|
|
"xvldi $xr9, 45\n\t"
|
|
"xvldi $xr10, 24\n\t"
|
|
"xvldi $xr11, 98\n\t"
|
|
"xvldi $xr12, 50\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[dst], 0\n\t"
|
|
"vld $vr31, %[dst], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"vld $vr1, %[dst], 16\n\t"
|
|
"vld $vr31, %[dst], 48\n\t"
|
|
"xvpermi.q $xr1, $xr31, 0x02\n\t"
|
|
"xvpickev.b $xr2, $xr1, $xr0\n\t"
|
|
"xvpickod.b $xr3, $xr1, $xr0\n\t"
|
|
"xvmulwev.h.bu $xr13, $xr2, $xr4\n\t"
|
|
"xvmaddwod.h.bu $xr13, $xr2, $xr6\n\t"
|
|
"xvmaddwev.h.bu $xr13, $xr3, $xr5\n\t"
|
|
"xvmulwev.h.bu $xr14, $xr2, $xr7\n\t"
|
|
"xvmaddwod.h.bu $xr14, $xr2, $xr9\n\t"
|
|
"xvmaddwev.h.bu $xr14, $xr3, $xr8\n\t"
|
|
"xvmulwev.h.bu $xr15, $xr2, $xr10\n\t"
|
|
"xvmaddwod.h.bu $xr15, $xr2, $xr12\n\t"
|
|
"xvmaddwev.h.bu $xr15, $xr3, $xr11\n\t"
|
|
"xvsrli.h $xr13, $xr13, 7\n\t"
|
|
"xvsrli.h $xr14, $xr14, 7\n\t"
|
|
"xvsrli.h $xr15, $xr15, 7\n\t"
|
|
"xvsat.hu $xr13, $xr13, 7\n\t"
|
|
"xvsat.hu $xr14, $xr14, 7\n\t"
|
|
"xvsat.hu $xr15, $xr15, 7\n\t"
|
|
"xvpickev.b $xr13, $xr13, $xr13\n\t"
|
|
"xvpickev.b $xr14, $xr14, $xr14\n\t"
|
|
"xvpickev.b $xr15, $xr15, $xr15\n\t"
|
|
"xvpickod.b $xr3, $xr3, $xr3\n\t"
|
|
"xvilvl.b $xr2, $xr14, $xr13\n\t"
|
|
"xvilvl.b $xr3, $xr3, $xr15\n\t"
|
|
"xvilvl.h $xr0, $xr3, $xr2\n\t"
|
|
"xvilvh.h $xr1, $xr3, $xr2\n\t"
|
|
"vst $vr0, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr1, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr12", "$xr13", "$xr14",
|
|
"$xr15", "$xr31");
|
|
}
|
|
|
|
void ARGBColorMatrixRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_argb,
|
|
const int8_t* matrix_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"vldrepl.w $vr4, %[mat], 0\n\t"
|
|
"vldrepl.w $vr31, %[mat], 0\n\t"
|
|
"xvpermi.q $xr4, $xr31, 0x02\n\t"
|
|
"vldrepl.w $vr5, %[mat], 4\n\t"
|
|
"vldrepl.w $vr31, %[mat], 4\n\t"
|
|
"xvpermi.q $xr5, $xr31, 0x02\n\t"
|
|
"vldrepl.w $vr6, %[mat], 8\n\t"
|
|
"vldrepl.w $vr31, %[mat], 8\n\t"
|
|
"xvpermi.q $xr6, $xr31, 0x02\n\t"
|
|
"vldrepl.w $vr7, %[mat], 12\n\t"
|
|
"vldrepl.w $vr31, %[mat], 12\n\t"
|
|
"xvpermi.q $xr7, $xr31, 0x02\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvmulwev.h.bu.b $xr8, $xr0, $xr4\n\t"
|
|
"xvmaddwod.h.bu.b $xr8, $xr0, $xr4\n\t"
|
|
"xvmulwev.h.bu.b $xr9, $xr0, $xr5\n\t"
|
|
"xvmaddwod.h.bu.b $xr9, $xr0, $xr5\n\t"
|
|
"xvmulwev.h.bu.b $xr10, $xr0, $xr6\n\t"
|
|
"xvmaddwod.h.bu.b $xr10, $xr0, $xr6\n\t"
|
|
"xvmulwev.h.bu.b $xr11, $xr0, $xr7\n\t"
|
|
"xvmaddwod.h.bu.b $xr11, $xr0, $xr7\n\t"
|
|
"xvhaddw.w.h $xr8, $xr8, $xr8\n\t"
|
|
"xvhaddw.w.h $xr9, $xr9, $xr9\n\t"
|
|
"xvhaddw.w.h $xr10, $xr10, $xr10\n\t"
|
|
"xvhaddw.w.h $xr11, $xr11, $xr11\n\t"
|
|
"xvilvl.w $xr12, $xr9, $xr8\n\t"
|
|
"xvilvh.w $xr13, $xr9, $xr8\n\t"
|
|
"xvilvl.w $xr14, $xr11, $xr10\n\t"
|
|
"xvilvh.w $xr15, $xr11, $xr10\n\t"
|
|
"xvilvl.d $xr8, $xr14, $xr12\n\t"
|
|
"xvilvh.d $xr9, $xr14, $xr12\n\t"
|
|
"xvilvl.d $xr10, $xr15, $xr13\n\t"
|
|
"xvilvh.d $xr11, $xr15, $xr13\n\t"
|
|
"xvssrani.h.w $xr9, $xr8, 6\n\t"
|
|
"xvssrani.h.w $xr11, $xr10, 6\n\t"
|
|
"xvssrani.bu.h $xr11, $xr9, 0\n\t"
|
|
"xvst $xr11, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [mat] "r"(matrix_argb)
|
|
: "memory", "$xr0", "$xr4", "$xr5", "$xr6", "$xr7", "$xr8", "$xr9",
|
|
"$xr10", "$xr11", "$xr12", "$xr13", "$xr14", "$xr15", "$xr31");
|
|
}
|
|
|
|
void ARGBQuantizeRow_LASX(uint8_t* dst_argb,
|
|
int scale,
|
|
int interval_size,
|
|
int interval_offset,
|
|
int width) {
|
|
asm volatile(
|
|
"xvreplgr2vr.h $xr4, %[scl]\n\t"
|
|
"xvreplgr2vr.b $xr5, %[isz]\n\t"
|
|
"xvreplgr2vr.b $xr6, %[ioff]\n\t"
|
|
"xvldi $xr7, 0xff\n\t"
|
|
"xvslli.w $xr7, $xr7, 24\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[dst], 0\n\t"
|
|
"xvilvl.b $xr1, $xr0, $xr0\n\t"
|
|
"xvilvh.b $xr2, $xr0, $xr0\n\t"
|
|
"xvsrli.h $xr1, $xr1, 8\n\t"
|
|
"xvsrli.h $xr2, $xr2, 8\n\t"
|
|
"xvmuh.hu $xr1, $xr1, $xr4\n\t"
|
|
"xvmuh.hu $xr2, $xr2, $xr4\n\t"
|
|
"xvsrlni.b.h $xr2, $xr1, 0\n\t"
|
|
"xvmul.b $xr2, $xr2, $xr5\n\t"
|
|
"xvadd.b $xr2, $xr2, $xr6\n\t"
|
|
"xvand.v $xr0, $xr0, $xr7\n\t"
|
|
"xvandn.v $xr2, $xr7, $xr2\n\t"
|
|
"xvor.v $xr2, $xr2, $xr0\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -8\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [dst] "+r"(dst_argb), [cnt] "+r"(width)
|
|
: [scl] "r"(scale), [isz] "r"(interval_size), [ioff] "r"(interval_offset)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr4", "$xr5", "$xr6", "$xr7");
|
|
}
|
|
|
|
void SobelRow_LASX(const uint8_t* src_sobelx,
|
|
const uint8_t* src_sobely,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr4, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sx], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr1, %[sy], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"xvsadd.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvilvl.b $xr2, $xr0, $xr0\n\t"
|
|
"xvilvl.b $xr3, $xr4, $xr0\n\t"
|
|
"xvilvl.h $xr0, $xr3, $xr2\n\t"
|
|
"xvilvh.h $xr1, $xr3, $xr2\n\t"
|
|
"vst $vr0, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr0, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr1, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr1, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sx], %[sx], 16\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sx] "+r"(src_sobelx), [sy] "+r"(src_sobely), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void SobelToPlaneRow_LASX(const uint8_t* src_sobelx,
|
|
const uint8_t* src_sobely,
|
|
uint8_t* dst_y,
|
|
int width) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[sx], 0\n\t"
|
|
"xvld $xr1, %[sy], 0\n\t"
|
|
"xvsadd.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[sx], %[sx], 32\n\t"
|
|
"addi.d %[sy], %[sy], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sx] "+r"(src_sobelx), [sy] "+r"(src_sobely), [dst] "+r"(dst_y),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1");
|
|
}
|
|
|
|
void SobelXYRow_LASX(const uint8_t* src_sobelx,
|
|
const uint8_t* src_sobely,
|
|
uint8_t* dst_argb,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr4, 0xff\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[sx], 0\n\t"
|
|
"xvpermi.d $xr0, $xr0, 0xd8\n\t"
|
|
"vld $vr1, %[sy], 0\n\t"
|
|
"xvpermi.d $xr1, $xr1, 0xd8\n\t"
|
|
"xvsadd.bu $xr2, $xr0, $xr1\n\t"
|
|
"xvilvl.b $xr1, $xr2, $xr1\n\t"
|
|
"xvilvl.b $xr0, $xr4, $xr0\n\t"
|
|
"xvilvl.h $xr2, $xr0, $xr1\n\t"
|
|
"xvilvh.h $xr3, $xr0, $xr1\n\t"
|
|
"vst $vr2, %[dst], 0\n\t"
|
|
"xvpermi.q $xr31, $xr2, 0x01\n\t"
|
|
"vst $vr31, %[dst], 32\n\t"
|
|
"vst $vr3, %[dst], 16\n\t"
|
|
"xvpermi.q $xr31, $xr3, 0x01\n\t"
|
|
"vst $vr31, %[dst], 48\n\t"
|
|
"addi.d %[sx], %[sx], 16\n\t"
|
|
"addi.d %[sy], %[sy], 16\n\t"
|
|
"addi.d %[dst], %[dst], 64\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sx] "+r"(src_sobelx), [sy] "+r"(src_sobely), [dst] "+r"(dst_argb),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr31");
|
|
}
|
|
|
|
void InterpolateRow_LASX(uint8_t* dst_ptr,
|
|
const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
int width,
|
|
int32_t source_y_fraction) {
|
|
int y1_fraction = source_y_fraction;
|
|
const uint8_t* src_ptr1 = src_ptr + src_stride;
|
|
if (y1_fraction == 0) {
|
|
CopyRow_LASX(src_ptr, dst_ptr, width);
|
|
return;
|
|
}
|
|
if (y1_fraction == 128) {
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvavgr.bu $xr0, $xr0, $xr1\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_ptr), [s1] "+r"(src_ptr1), [dst] "+r"(dst_ptr),
|
|
[cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1");
|
|
return;
|
|
}
|
|
uint16_t y_frac = (uint16_t)((256 - y1_fraction) + (y1_fraction << 8));
|
|
asm volatile(
|
|
"xvreplgr2vr.h $xr3, %[yf]\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[s0], 0\n\t"
|
|
"xvld $xr1, %[s1], 0\n\t"
|
|
"xvilvl.b $xr2, $xr1, $xr0\n\t"
|
|
"xvilvh.b $xr4, $xr1, $xr0\n\t"
|
|
"xvmulwev.h.bu $xr0, $xr2, $xr3\n\t"
|
|
"xvmaddwod.h.bu $xr0, $xr2, $xr3\n\t"
|
|
"xvmulwev.h.bu $xr1, $xr4, $xr3\n\t"
|
|
"xvmaddwod.h.bu $xr1, $xr4, $xr3\n\t"
|
|
"xvsrarni.b.h $xr1, $xr0, 8\n\t"
|
|
"xvst $xr1, %[dst], 0\n\t"
|
|
"addi.d %[s0], %[s0], 32\n\t"
|
|
"addi.d %[s1], %[s1], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_ptr), [s1] "+r"(src_ptr1), [dst] "+r"(dst_ptr),
|
|
[cnt] "+r"(width)
|
|
: [yf] "r"(y_frac)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4");
|
|
}
|
|
|
|
void HalfFloatRow_LASX(const uint16_t* src,
|
|
uint16_t* dst,
|
|
float scale,
|
|
int width) {
|
|
const float bias = 1.9259299444e-34f; // 2^-112, see HalfFloatRow_C
|
|
asm volatile(
|
|
"xvldrepl.w $xr3, %[scale], 0\n\t"
|
|
"xvldrepl.w $xr5, %[bias], 0\n\t"
|
|
"xvldi $xr4, 0\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvilvl.h $xr1, $xr4, $xr0\n\t"
|
|
"xvilvh.h $xr2, $xr4, $xr0\n\t"
|
|
"xvffint.s.wu $xr1, $xr1\n\t"
|
|
"xvffint.s.wu $xr2, $xr2\n\t"
|
|
"xvfmul.s $xr1, $xr1, $xr3\n\t"
|
|
"xvfmul.s $xr2, $xr2, $xr3\n\t"
|
|
"xvfmul.s $xr1, $xr1, $xr5\n\t"
|
|
"xvfmul.s $xr2, $xr2, $xr5\n\t"
|
|
"xvsrlni.h.w $xr2, $xr1, 13\n\t"
|
|
"xvst $xr2, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src), [dst] "+r"(dst), [cnt] "+r"(width)
|
|
: [scale] "r"(&scale), [bias] "r"(&bias)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5");
|
|
}
|
|
|
|
void ARGBToYMatrixRow_LASX(const uint8_t* src_argb,
|
|
uint8_t* dst_y,
|
|
int width,
|
|
const struct ArgbConstants* c) {
|
|
asm volatile(
|
|
"xvldrepl.b $xr0, %[c], 0\n\t"
|
|
"xvldrepl.b $xr1, %[c], 1\n\t"
|
|
"xvldrepl.b $xr2, %[c], 2\n\t"
|
|
"xvldrepl.b $xr3, %[c], 3\n\t"
|
|
"xvldrepl.h $xr4, %[c], 96\n\t"
|
|
"1:\n\t"
|
|
"vld $vr5, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 32\n\t"
|
|
"xvpermi.q $xr5, $xr31, 0x02\n\t"
|
|
"vld $vr6, %[src], 16\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr6, $xr31, 0x02\n\t"
|
|
"xvor.v $xr9, $xr4, $xr4\n\t"
|
|
"xvpickev.b $xr7, $xr6, $xr5\n\t"
|
|
"xvpickod.b $xr8, $xr6, $xr5\n\t"
|
|
"xvmaddwev.h.bu $xr9, $xr7, $xr0\n\t"
|
|
"xvmaddwev.h.bu $xr9, $xr8, $xr1\n\t"
|
|
"xvmaddwod.h.bu $xr9, $xr7, $xr2\n\t"
|
|
"xvmaddwod.h.bu $xr9, $xr8, $xr3\n\t"
|
|
"xvpickod.b $xr9, $xr9, $xr9\n\t"
|
|
"xvpermi.d $xr31, $xr9, 0xd8\n\t"
|
|
"vst $vr31, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 64\n\t"
|
|
"addi.d %[dst], %[dst], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
: [c] "r"(c)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr31");
|
|
}
|
|
|
|
void RGBToYMatrixRow_LASX(const uint8_t* src_rgba,
|
|
uint8_t* dst_y,
|
|
int width,
|
|
const struct ArgbConstants* c) {
|
|
static const uint8_t kShuf[32] = {0, 1, 2, 0, 3, 4, 5, 0, 6, 7, 8,
|
|
0, 9, 10, 11, 0, 4, 5, 6, 0, 7, 8,
|
|
9, 0, 10, 11, 12, 0, 13, 14, 15, 0};
|
|
asm volatile(
|
|
"xvldrepl.b $xr0, %[c], 0\n\t"
|
|
"xvldrepl.b $xr1, %[c], 1\n\t"
|
|
"xvldrepl.b $xr2, %[c], 2\n\t"
|
|
"xvldrepl.h $xr3, %[c], 96\n\t"
|
|
"vld $vr4, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr4, $xr4, 0x00\n\t"
|
|
"vld $vr5, %[shuf], 16\n\t"
|
|
"xvpermi.q $xr5, $xr5, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr6, %[src], 0\n\t"
|
|
"vld $vr31, %[src], 48\n\t"
|
|
"xvpermi.q $xr6, $xr31, 0x02\n\t"
|
|
"vld $vr7, %[src], 12\n\t"
|
|
"vld $vr31, %[src], 60\n\t"
|
|
"xvpermi.q $xr7, $xr31, 0x02\n\t"
|
|
"vld $vr8, %[src], 24\n\t"
|
|
"vld $vr31, %[src], 72\n\t"
|
|
"xvpermi.q $xr8, $xr31, 0x02\n\t"
|
|
"vld $vr9, %[src], 32\n\t"
|
|
"vld $vr31, %[src], 80\n\t"
|
|
"xvpermi.q $xr9, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr6, $xr6, $xr6, $xr4\n\t"
|
|
"xvshuf.b $xr7, $xr7, $xr7, $xr4\n\t"
|
|
"xvshuf.b $xr8, $xr8, $xr8, $xr4\n\t"
|
|
"xvshuf.b $xr9, $xr9, $xr9, $xr5\n\t"
|
|
"xvor.v $xr12, $xr3, $xr3\n\t"
|
|
"xvor.v $xr13, $xr3, $xr3\n\t"
|
|
"xvpickev.b $xr10, $xr7, $xr6\n\t"
|
|
"xvpickod.b $xr11, $xr7, $xr6\n\t"
|
|
"xvmaddwev.h.bu $xr12, $xr10, $xr0\n\t"
|
|
"xvmaddwev.h.bu $xr12, $xr11, $xr1\n\t"
|
|
"xvmaddwod.h.bu $xr12, $xr10, $xr2\n\t"
|
|
"xvpickev.b $xr10, $xr9, $xr8\n\t"
|
|
"xvpickod.b $xr11, $xr9, $xr8\n\t"
|
|
"xvmaddwev.h.bu $xr13, $xr10, $xr0\n\t"
|
|
"xvmaddwev.h.bu $xr13, $xr11, $xr1\n\t"
|
|
"xvmaddwod.h.bu $xr13, $xr10, $xr2\n\t"
|
|
"xvpickod.b $xr12, $xr13, $xr12\n\t"
|
|
"xvst $xr12, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 96\n\t"
|
|
"addi.d %[dst], %[dst], 32\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_rgba), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
: [c] "r"(c), [shuf] "r"(kShuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr12", "$xr13", "$xr31");
|
|
}
|
|
|
|
void RGB565ToYRow_LASX(const uint8_t* src_rgb565, uint8_t* dst_y, int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"xvldi $xr8, 25\n\t"
|
|
"xvsrli.h $xr8, $xr8, 8\n\t"
|
|
"xvldi $xr9, 129\n\t"
|
|
"xvsrli.h $xr9, $xr9, 8\n\t"
|
|
"xvldi $xr10, 66\n\t"
|
|
"xvsrli.h $xr10, $xr10, 8\n\t"
|
|
"xvldi $xr11, 33\n\t"
|
|
"xvsrli.h $xr11, $xr11, 8\n\t"
|
|
"xvslli.h $xr11, $xr11, 7\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvmul.h $xr1, $xr1, $xr8\n\t"
|
|
"xvmadd.h $xr1, $xr2, $xr9\n\t"
|
|
"xvmadd.h $xr1, $xr3, $xr10\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr11\n\t"
|
|
"xvpickod.b $xr1, $xr1, $xr1\n\t"
|
|
"xvpermi.d $xr31, $xr1, 0xd8\n\t"
|
|
"vst $vr31, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_rgb565), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr6", "$xr7",
|
|
"$xr8", "$xr9", "$xr10", "$xr11", "$xr31");
|
|
}
|
|
|
|
void ARGB1555ToYRow_LASX(const uint8_t* src_argb1555,
|
|
uint8_t* dst_y,
|
|
int width) {
|
|
asm volatile(
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr8, 25\n\t"
|
|
"xvsrli.h $xr8, $xr8, 8\n\t"
|
|
"xvldi $xr9, 129\n\t"
|
|
"xvsrli.h $xr9, $xr9, 8\n\t"
|
|
"xvldi $xr10, 66\n\t"
|
|
"xvsrli.h $xr10, $xr10, 8\n\t"
|
|
"xvldi $xr11, 33\n\t"
|
|
"xvsrli.h $xr11, $xr11, 8\n\t"
|
|
"xvslli.h $xr11, $xr11, 7\n\t"
|
|
"1:\n\t"
|
|
"xvld $xr0, %[src], 0\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvmul.h $xr1, $xr1, $xr8\n\t"
|
|
"xvmadd.h $xr1, $xr2, $xr9\n\t"
|
|
"xvmadd.h $xr1, $xr3, $xr10\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr11\n\t"
|
|
"xvpickod.b $xr1, $xr1, $xr1\n\t"
|
|
"xvpermi.d $xr31, $xr1, 0xd8\n\t"
|
|
"vst $vr31, %[dst], 0\n\t"
|
|
"addi.d %[src], %[src], 32\n\t"
|
|
"addi.d %[dst], %[dst], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -16\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [src] "+r"(src_argb1555), [dst] "+r"(dst_y), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr8", "$xr9", "$xr10", "$xr11", "$xr31");
|
|
}
|
|
|
|
void RGB565ToUVRow_LASX(const uint8_t* src_rgb565,
|
|
int src_stride_rgb565,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
const uint8_t* next_rgb565 = src_rgb565 + src_stride_rgb565;
|
|
asm volatile(
|
|
"xvldi $xr20, 112\n\t"
|
|
"xvsrli.h $xr20, $xr20, 8\n\t"
|
|
"xvldi $xr21, 74\n\t"
|
|
"xvsrli.h $xr21, $xr21, 8\n\t"
|
|
"xvldi $xr22, 38\n\t"
|
|
"xvsrli.h $xr22, $xr22, 8\n\t"
|
|
"xvldi $xr23, 94\n\t"
|
|
"xvsrli.h $xr23, $xr23, 8\n\t"
|
|
"xvldi $xr24, 18\n\t"
|
|
"xvsrli.h $xr24, $xr24, 8\n\t"
|
|
"xvldi $xr25, 128\n\t"
|
|
"xvsrli.h $xr25, $xr25, 8\n\t"
|
|
"xvslli.h $xr25, $xr25, 8\n\t"
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvor.v $xr10, $xr1, $xr1\n\t"
|
|
"xvor.v $xr11, $xr2, $xr2\n\t"
|
|
"xvor.v $xr12, $xr3, $xr3\n\t"
|
|
"vld $vr0, %[s0], 16\n\t"
|
|
"vld $vr31, %[s0], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr10, $xr1, $xr10\n\t"
|
|
"xvpickev.b $xr11, $xr2, $xr11\n\t"
|
|
"xvpickev.b $xr12, $xr3, $xr12\n\t"
|
|
"vld $vr0, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvor.v $xr13, $xr1, $xr1\n\t"
|
|
"xvor.v $xr14, $xr2, $xr2\n\t"
|
|
"xvor.v $xr15, $xr3, $xr3\n\t"
|
|
"vld $vr0, %[s1], 16\n\t"
|
|
"vld $vr31, %[s1], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr7\n\t"
|
|
"xvsrli.h $xr3, $xr0, 11\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 2\n\t"
|
|
"xvsrli.h $xr2, $xr2, 4\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr5, $xr3, $xr15\n\t"
|
|
"xvpickev.b $xr4, $xr2, $xr14\n\t"
|
|
"xvpickev.b $xr3, $xr1, $xr13\n\t"
|
|
"xvor.v $xr0, $xr10, $xr10\n\t"
|
|
"xvor.v $xr1, $xr11, $xr11\n\t"
|
|
"xvor.v $xr2, $xr12, $xr12\n\t"
|
|
"xvhaddw.hu.bu $xr0, $xr0, $xr0\n\t"
|
|
"xvhaddw.hu.bu $xr3, $xr3, $xr3\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr3\n\t"
|
|
"xvsrari.h $xr0, $xr0, 2\n\t"
|
|
"xvhaddw.hu.bu $xr1, $xr1, $xr1\n\t"
|
|
"xvhaddw.hu.bu $xr4, $xr4, $xr4\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr4\n\t"
|
|
"xvsrari.h $xr1, $xr1, 2\n\t"
|
|
"xvhaddw.hu.bu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.hu.bu $xr5, $xr5, $xr5\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr5\n\t"
|
|
"xvsrari.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr6, $xr25, $xr25\n\t"
|
|
"xvor.v $xr7, $xr25, $xr25\n\t"
|
|
"xvmadd.h $xr6, $xr0, $xr20\n\t"
|
|
"xvmsub.h $xr6, $xr1, $xr21\n\t"
|
|
"xvmsub.h $xr6, $xr2, $xr22\n\t"
|
|
"xvmadd.h $xr7, $xr2, $xr20\n\t"
|
|
"xvmsub.h $xr7, $xr1, $xr23\n\t"
|
|
"xvmsub.h $xr7, $xr0, $xr24\n\t"
|
|
"xvpickod.b $xr6, $xr7, $xr6\n\t"
|
|
"xvpermi.d $xr31, $xr6, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"vstelm.d $vr6, %[dv], 0, 1\n\t"
|
|
"xvpermi.q $xr31, $xr6, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dv], 8, 1\n\t"
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"xvldi $xr7, 63\n\t"
|
|
"xvsrli.h $xr7, $xr7, 8\n\t"
|
|
"addi.d %[s0], %[s0], 64\n\t"
|
|
"addi.d %[s1], %[s1], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_rgb565), [s1] "+r"(next_rgb565), [du] "+r"(dst_u),
|
|
[dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr10", "$xr11", "$xr12", "$xr13", "$xr14", "$xr15", "$xr20",
|
|
"$xr21", "$xr22", "$xr23", "$xr24", "$xr25", "$xr31");
|
|
}
|
|
|
|
void ARGB1555ToUVRow_LASX(const uint8_t* src_argb1555,
|
|
int src_stride_argb1555,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
const uint8_t* next_argb1555 = src_argb1555 + src_stride_argb1555;
|
|
asm volatile(
|
|
"xvldi $xr20, 112\n\t"
|
|
"xvsrli.h $xr20, $xr20, 8\n\t"
|
|
"xvldi $xr21, 74\n\t"
|
|
"xvsrli.h $xr21, $xr21, 8\n\t"
|
|
"xvldi $xr22, 38\n\t"
|
|
"xvsrli.h $xr22, $xr22, 8\n\t"
|
|
"xvldi $xr23, 94\n\t"
|
|
"xvsrli.h $xr23, $xr23, 8\n\t"
|
|
"xvldi $xr24, 18\n\t"
|
|
"xvsrli.h $xr24, $xr24, 8\n\t"
|
|
"xvldi $xr25, 128\n\t"
|
|
"xvsrli.h $xr25, $xr25, 8\n\t"
|
|
"xvslli.h $xr25, $xr25, 8\n\t"
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"1:\n\t"
|
|
"vld $vr0, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvor.v $xr10, $xr1, $xr1\n\t"
|
|
"xvor.v $xr11, $xr2, $xr2\n\t"
|
|
"xvor.v $xr12, $xr3, $xr3\n\t"
|
|
"vld $vr0, %[s0], 16\n\t"
|
|
"vld $vr31, %[s0], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr10, $xr1, $xr10\n\t"
|
|
"xvpickev.b $xr11, $xr2, $xr11\n\t"
|
|
"xvpickev.b $xr12, $xr3, $xr12\n\t"
|
|
"vld $vr0, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 32\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvor.v $xr13, $xr1, $xr1\n\t"
|
|
"xvor.v $xr14, $xr2, $xr2\n\t"
|
|
"xvor.v $xr15, $xr3, $xr3\n\t"
|
|
"vld $vr0, %[s1], 16\n\t"
|
|
"vld $vr31, %[s1], 48\n\t"
|
|
"xvpermi.q $xr0, $xr31, 0x02\n\t"
|
|
"xvand.v $xr1, $xr0, $xr6\n\t"
|
|
"xvsrli.h $xr2, $xr0, 5\n\t"
|
|
"xvand.v $xr2, $xr2, $xr6\n\t"
|
|
"xvsrli.h $xr3, $xr0, 10\n\t"
|
|
"xvand.v $xr3, $xr3, $xr6\n\t"
|
|
"xvsrai.h $xr5, $xr0, 15\n\t"
|
|
"xvslli.h $xr4, $xr1, 3\n\t"
|
|
"xvsrli.h $xr1, $xr1, 2\n\t"
|
|
"xvor.v $xr1, $xr4, $xr1\n\t"
|
|
"xvslli.h $xr4, $xr2, 3\n\t"
|
|
"xvsrli.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr2, $xr4, $xr2\n\t"
|
|
"xvslli.h $xr4, $xr3, 3\n\t"
|
|
"xvsrli.h $xr3, $xr3, 2\n\t"
|
|
"xvor.v $xr3, $xr4, $xr3\n\t"
|
|
"xvpickev.b $xr5, $xr3, $xr15\n\t"
|
|
"xvpickev.b $xr4, $xr2, $xr14\n\t"
|
|
"xvpickev.b $xr3, $xr1, $xr13\n\t"
|
|
"xvor.v $xr0, $xr10, $xr10\n\t"
|
|
"xvor.v $xr1, $xr11, $xr11\n\t"
|
|
"xvor.v $xr2, $xr12, $xr12\n\t"
|
|
"xvhaddw.hu.bu $xr0, $xr0, $xr0\n\t"
|
|
"xvhaddw.hu.bu $xr3, $xr3, $xr3\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr3\n\t"
|
|
"xvsrari.h $xr0, $xr0, 2\n\t"
|
|
"xvhaddw.hu.bu $xr1, $xr1, $xr1\n\t"
|
|
"xvhaddw.hu.bu $xr4, $xr4, $xr4\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr4\n\t"
|
|
"xvsrari.h $xr1, $xr1, 2\n\t"
|
|
"xvhaddw.hu.bu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.hu.bu $xr5, $xr5, $xr5\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr5\n\t"
|
|
"xvsrari.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr6, $xr25, $xr25\n\t"
|
|
"xvor.v $xr7, $xr25, $xr25\n\t"
|
|
"xvmadd.h $xr6, $xr0, $xr20\n\t"
|
|
"xvmsub.h $xr6, $xr1, $xr21\n\t"
|
|
"xvmsub.h $xr6, $xr2, $xr22\n\t"
|
|
"xvmadd.h $xr7, $xr2, $xr20\n\t"
|
|
"xvmsub.h $xr7, $xr1, $xr23\n\t"
|
|
"xvmsub.h $xr7, $xr0, $xr24\n\t"
|
|
"xvpickod.b $xr6, $xr7, $xr6\n\t"
|
|
"xvpermi.d $xr31, $xr6, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"vstelm.d $vr6, %[dv], 0, 1\n\t"
|
|
"xvpermi.q $xr31, $xr6, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dv], 8, 1\n\t"
|
|
"xvldi $xr6, 31\n\t"
|
|
"xvsrli.h $xr6, $xr6, 8\n\t"
|
|
"addi.d %[s0], %[s0], 64\n\t"
|
|
"addi.d %[s1], %[s1], 64\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src_argb1555), [s1] "+r"(next_argb1555), [du] "+r"(dst_u),
|
|
[dv] "+r"(dst_v), [cnt] "+r"(width)
|
|
:
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr10", "$xr11", "$xr12", "$xr13", "$xr14", "$xr15", "$xr20",
|
|
"$xr21", "$xr22", "$xr23", "$xr24", "$xr25", "$xr31");
|
|
}
|
|
|
|
static LIBYUV_NOINLINE void XRGBToUVHelper_LASX(const uint8_t* src,
|
|
int stride,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
const uint8_t* shuf,
|
|
int width) {
|
|
const uint8_t* next = src + stride;
|
|
asm volatile(
|
|
"xvldi $xr20, 112\n\t"
|
|
"xvsrli.h $xr20, $xr20, 8\n\t"
|
|
"xvldi $xr21, 74\n\t"
|
|
"xvsrli.h $xr21, $xr21, 8\n\t"
|
|
"xvldi $xr22, 38\n\t"
|
|
"xvsrli.h $xr22, $xr22, 8\n\t"
|
|
"xvldi $xr23, 94\n\t"
|
|
"xvsrli.h $xr23, $xr23, 8\n\t"
|
|
"xvldi $xr24, 18\n\t"
|
|
"xvsrli.h $xr24, $xr24, 8\n\t"
|
|
"xvldi $xr25, 128\n\t"
|
|
"xvsrli.h $xr25, $xr25, 8\n\t"
|
|
"xvslli.h $xr25, $xr25, 8\n\t"
|
|
"vld $vr16, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr16, $xr16, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr8, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 64\n\t"
|
|
"xvpermi.q $xr8, $xr31, 0x02\n\t"
|
|
"vld $vr9, %[s0], 16\n\t"
|
|
"vld $vr31, %[s0], 80\n\t"
|
|
"xvpermi.q $xr9, $xr31, 0x02\n\t"
|
|
"vld $vr10, %[s0], 32\n\t"
|
|
"vld $vr31, %[s0], 96\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"vld $vr11, %[s0], 48\n\t"
|
|
"vld $vr31, %[s0], 112\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr8, $xr8, $xr8, $xr16\n\t"
|
|
"xvshuf.b $xr9, $xr9, $xr9, $xr16\n\t"
|
|
"xvshuf.b $xr10, $xr10, $xr10, $xr16\n\t"
|
|
"xvshuf.b $xr11, $xr11, $xr11, $xr16\n\t"
|
|
"xvpickev.b $xr12, $xr9, $xr8\n\t"
|
|
"xvpickod.b $xr13, $xr9, $xr8\n\t"
|
|
"xvpickev.b $xr14, $xr11, $xr10\n\t"
|
|
"xvpickod.b $xr15, $xr11, $xr10\n\t"
|
|
"xvpickev.b $xr0, $xr14, $xr12\n\t"
|
|
"xvpickod.b $xr2, $xr14, $xr12\n\t"
|
|
"xvpickev.b $xr1, $xr15, $xr13\n\t"
|
|
"vld $vr8, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 64\n\t"
|
|
"xvpermi.q $xr8, $xr31, 0x02\n\t"
|
|
"vld $vr9, %[s1], 16\n\t"
|
|
"vld $vr31, %[s1], 80\n\t"
|
|
"xvpermi.q $xr9, $xr31, 0x02\n\t"
|
|
"vld $vr10, %[s1], 32\n\t"
|
|
"vld $vr31, %[s1], 96\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"vld $vr11, %[s1], 48\n\t"
|
|
"vld $vr31, %[s1], 112\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr8, $xr8, $xr8, $xr16\n\t"
|
|
"xvshuf.b $xr9, $xr9, $xr9, $xr16\n\t"
|
|
"xvshuf.b $xr10, $xr10, $xr10, $xr16\n\t"
|
|
"xvshuf.b $xr11, $xr11, $xr11, $xr16\n\t"
|
|
"xvpickev.b $xr12, $xr9, $xr8\n\t"
|
|
"xvpickod.b $xr13, $xr9, $xr8\n\t"
|
|
"xvpickev.b $xr14, $xr11, $xr10\n\t"
|
|
"xvpickod.b $xr15, $xr11, $xr10\n\t"
|
|
"xvpickev.b $xr3, $xr14, $xr12\n\t"
|
|
"xvpickod.b $xr5, $xr14, $xr12\n\t"
|
|
"xvpickev.b $xr4, $xr15, $xr13\n\t"
|
|
"xvhaddw.hu.bu $xr0, $xr0, $xr0\n\t"
|
|
"xvhaddw.hu.bu $xr3, $xr3, $xr3\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr3\n\t"
|
|
"xvsrari.h $xr0, $xr0, 2\n\t"
|
|
"xvhaddw.hu.bu $xr1, $xr1, $xr1\n\t"
|
|
"xvhaddw.hu.bu $xr4, $xr4, $xr4\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr4\n\t"
|
|
"xvsrari.h $xr1, $xr1, 2\n\t"
|
|
"xvhaddw.hu.bu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.hu.bu $xr5, $xr5, $xr5\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr5\n\t"
|
|
"xvsrari.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr6, $xr25, $xr25\n\t"
|
|
"xvor.v $xr7, $xr25, $xr25\n\t"
|
|
"xvmadd.h $xr6, $xr0, $xr20\n\t"
|
|
"xvmsub.h $xr6, $xr1, $xr21\n\t"
|
|
"xvmsub.h $xr6, $xr2, $xr22\n\t"
|
|
"xvmadd.h $xr7, $xr2, $xr20\n\t"
|
|
"xvmsub.h $xr7, $xr1, $xr23\n\t"
|
|
"xvmsub.h $xr7, $xr0, $xr24\n\t"
|
|
"xvpickod.b $xr6, $xr7, $xr6\n\t"
|
|
"xvpermi.d $xr31, $xr6, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"vstelm.d $vr6, %[dv], 0, 1\n\t"
|
|
"xvpermi.q $xr31, $xr6, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dv], 8, 1\n\t"
|
|
"addi.d %[s0], %[s0], 128\n\t"
|
|
"addi.d %[s1], %[s1], 128\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src), [s1] "+r"(next), [du] "+r"(dst_u), [dv] "+r"(dst_v),
|
|
[cnt] "+r"(width)
|
|
: [shuf] "r"(shuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr12", "$xr13", "$xr14",
|
|
"$xr15", "$xr16", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24", "$xr25",
|
|
"$xr31");
|
|
}
|
|
|
|
void BGRAToUVRow_LASX(const uint8_t* src_bgra,
|
|
int src_stride_bgra,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kShuf[16] = {3, 2, 1, 0, 7, 6, 5, 4,
|
|
11, 10, 9, 8, 15, 14, 13, 12};
|
|
XRGBToUVHelper_LASX(src_bgra, src_stride_bgra, dst_u, dst_v, kShuf, width);
|
|
}
|
|
|
|
void ABGRToUVRow_LASX(const uint8_t* src_abgr,
|
|
int src_stride_abgr,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kShuf[16] = {2, 1, 0, 3, 6, 5, 4, 7,
|
|
10, 9, 8, 11, 14, 13, 12, 15};
|
|
XRGBToUVHelper_LASX(src_abgr, src_stride_abgr, dst_u, dst_v, kShuf, width);
|
|
}
|
|
|
|
void RGBAToUVRow_LASX(const uint8_t* src_rgba,
|
|
int src_stride_rgba,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kShuf[16] = {1, 2, 3, 0, 5, 6, 7, 4,
|
|
9, 10, 11, 8, 13, 14, 15, 12};
|
|
XRGBToUVHelper_LASX(src_rgba, src_stride_rgba, dst_u, dst_v, kShuf, width);
|
|
}
|
|
|
|
static LIBYUV_NOINLINE void RGB24ToUVHelper_LASX(const uint8_t* src,
|
|
int stride,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
const uint8_t* shuf,
|
|
int width) {
|
|
const uint8_t* next = src + stride;
|
|
asm volatile(
|
|
"xvldi $xr20, 112\n\t"
|
|
"xvsrli.h $xr20, $xr20, 8\n\t"
|
|
"xvldi $xr21, 74\n\t"
|
|
"xvsrli.h $xr21, $xr21, 8\n\t"
|
|
"xvldi $xr22, 38\n\t"
|
|
"xvsrli.h $xr22, $xr22, 8\n\t"
|
|
"xvldi $xr23, 94\n\t"
|
|
"xvsrli.h $xr23, $xr23, 8\n\t"
|
|
"xvldi $xr24, 18\n\t"
|
|
"xvsrli.h $xr24, $xr24, 8\n\t"
|
|
"xvldi $xr25, 128\n\t"
|
|
"xvsrli.h $xr25, $xr25, 8\n\t"
|
|
"xvslli.h $xr25, $xr25, 8\n\t"
|
|
"vld $vr16, %[shuf], 0\n\t"
|
|
"xvpermi.q $xr16, $xr16, 0x00\n\t"
|
|
"vld $vr17, %[shuf], 16\n\t"
|
|
"xvpermi.q $xr17, $xr17, 0x00\n\t"
|
|
"1:\n\t"
|
|
"vld $vr8, %[s0], 0\n\t"
|
|
"vld $vr31, %[s0], 48\n\t"
|
|
"xvpermi.q $xr8, $xr31, 0x02\n\t"
|
|
"vld $vr9, %[s0], 12\n\t"
|
|
"vld $vr31, %[s0], 60\n\t"
|
|
"xvpermi.q $xr9, $xr31, 0x02\n\t"
|
|
"vld $vr10, %[s0], 24\n\t"
|
|
"vld $vr31, %[s0], 72\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"vld $vr11, %[s0], 32\n\t"
|
|
"vld $vr31, %[s0], 80\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr8, $xr8, $xr8, $xr16\n\t"
|
|
"xvshuf.b $xr9, $xr9, $xr9, $xr16\n\t"
|
|
"xvshuf.b $xr10, $xr10, $xr10, $xr16\n\t"
|
|
"xvshuf.b $xr11, $xr11, $xr11, $xr17\n\t"
|
|
"xvpickev.b $xr12, $xr9, $xr8\n\t"
|
|
"xvpickod.b $xr13, $xr9, $xr8\n\t"
|
|
"xvpickev.b $xr14, $xr11, $xr10\n\t"
|
|
"xvpickod.b $xr15, $xr11, $xr10\n\t"
|
|
"xvpickev.b $xr0, $xr14, $xr12\n\t"
|
|
"xvpickod.b $xr2, $xr14, $xr12\n\t"
|
|
"xvpickev.b $xr1, $xr15, $xr13\n\t"
|
|
"vld $vr8, %[s1], 0\n\t"
|
|
"vld $vr31, %[s1], 48\n\t"
|
|
"xvpermi.q $xr8, $xr31, 0x02\n\t"
|
|
"vld $vr9, %[s1], 12\n\t"
|
|
"vld $vr31, %[s1], 60\n\t"
|
|
"xvpermi.q $xr9, $xr31, 0x02\n\t"
|
|
"vld $vr10, %[s1], 24\n\t"
|
|
"vld $vr31, %[s1], 72\n\t"
|
|
"xvpermi.q $xr10, $xr31, 0x02\n\t"
|
|
"vld $vr11, %[s1], 32\n\t"
|
|
"vld $vr31, %[s1], 80\n\t"
|
|
"xvpermi.q $xr11, $xr31, 0x02\n\t"
|
|
"xvshuf.b $xr8, $xr8, $xr8, $xr16\n\t"
|
|
"xvshuf.b $xr9, $xr9, $xr9, $xr16\n\t"
|
|
"xvshuf.b $xr10, $xr10, $xr10, $xr16\n\t"
|
|
"xvshuf.b $xr11, $xr11, $xr11, $xr17\n\t"
|
|
"xvpickev.b $xr12, $xr9, $xr8\n\t"
|
|
"xvpickod.b $xr13, $xr9, $xr8\n\t"
|
|
"xvpickev.b $xr14, $xr11, $xr10\n\t"
|
|
"xvpickod.b $xr15, $xr11, $xr10\n\t"
|
|
"xvpickev.b $xr3, $xr14, $xr12\n\t"
|
|
"xvpickod.b $xr5, $xr14, $xr12\n\t"
|
|
"xvpickev.b $xr4, $xr15, $xr13\n\t"
|
|
"xvhaddw.hu.bu $xr0, $xr0, $xr0\n\t"
|
|
"xvhaddw.hu.bu $xr3, $xr3, $xr3\n\t"
|
|
"xvadd.h $xr0, $xr0, $xr3\n\t"
|
|
"xvsrari.h $xr0, $xr0, 2\n\t"
|
|
"xvhaddw.hu.bu $xr1, $xr1, $xr1\n\t"
|
|
"xvhaddw.hu.bu $xr4, $xr4, $xr4\n\t"
|
|
"xvadd.h $xr1, $xr1, $xr4\n\t"
|
|
"xvsrari.h $xr1, $xr1, 2\n\t"
|
|
"xvhaddw.hu.bu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.hu.bu $xr5, $xr5, $xr5\n\t"
|
|
"xvadd.h $xr2, $xr2, $xr5\n\t"
|
|
"xvsrari.h $xr2, $xr2, 2\n\t"
|
|
"xvor.v $xr6, $xr25, $xr25\n\t"
|
|
"xvor.v $xr7, $xr25, $xr25\n\t"
|
|
"xvmadd.h $xr6, $xr0, $xr20\n\t"
|
|
"xvmsub.h $xr6, $xr1, $xr21\n\t"
|
|
"xvmsub.h $xr6, $xr2, $xr22\n\t"
|
|
"xvmadd.h $xr7, $xr2, $xr20\n\t"
|
|
"xvmsub.h $xr7, $xr1, $xr23\n\t"
|
|
"xvmsub.h $xr7, $xr0, $xr24\n\t"
|
|
"xvpickod.b $xr6, $xr7, $xr6\n\t"
|
|
"xvpermi.d $xr31, $xr6, 0xd8\n\t"
|
|
"vst $vr31, %[du], 0\n\t"
|
|
"vstelm.d $vr6, %[dv], 0, 1\n\t"
|
|
"xvpermi.q $xr31, $xr6, 0x01\n\t"
|
|
"vstelm.d $vr31, %[dv], 8, 1\n\t"
|
|
"addi.d %[s0], %[s0], 96\n\t"
|
|
"addi.d %[s1], %[s1], 96\n\t"
|
|
"addi.d %[du], %[du], 16\n\t"
|
|
"addi.d %[dv], %[dv], 16\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [s0] "+r"(src), [s1] "+r"(next), [du] "+r"(dst_u), [dv] "+r"(dst_v),
|
|
[cnt] "+r"(width)
|
|
: [shuf] "r"(shuf)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11", "$xr12", "$xr13", "$xr14",
|
|
"$xr15", "$xr16", "$xr17", "$xr20", "$xr21", "$xr22", "$xr23", "$xr24",
|
|
"$xr25", "$xr31");
|
|
}
|
|
|
|
void RGB24ToUVRow_LASX(const uint8_t* src_rgb24,
|
|
int src_stride_rgb24,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kShuf[32] = {0, 1, 2, 0, 3, 4, 5, 0, 6, 7, 8,
|
|
0, 9, 10, 11, 0, 4, 5, 6, 0, 7, 8,
|
|
9, 0, 10, 11, 12, 0, 13, 14, 15, 0};
|
|
RGB24ToUVHelper_LASX(src_rgb24, src_stride_rgb24, dst_u, dst_v, kShuf, width);
|
|
}
|
|
|
|
void RAWToUVRow_LASX(const uint8_t* src_raw,
|
|
int src_stride_raw,
|
|
uint8_t* dst_u,
|
|
uint8_t* dst_v,
|
|
int width) {
|
|
static const uint8_t kShuf[32] = {2, 1, 0, 0, 5, 4, 3, 0, 8, 7, 6,
|
|
0, 11, 10, 9, 0, 6, 5, 4, 0, 9, 8,
|
|
7, 0, 12, 11, 10, 0, 15, 14, 13, 0};
|
|
RGB24ToUVHelper_LASX(src_raw, src_stride_raw, dst_u, dst_v, kShuf, width);
|
|
}
|
|
|
|
#ifdef HAS_SSIM4X4ROW_LASX
|
|
void Ssim4x4Row_LASX(const uint8_t* src_a,
|
|
int stride_a,
|
|
const uint8_t* src_b,
|
|
int stride_b,
|
|
uint32_t* dst,
|
|
int width) {
|
|
const uint8_t* pa;
|
|
const uint8_t* pb;
|
|
int r;
|
|
asm volatile(
|
|
"1:\n\t"
|
|
"move %[pa], %[sa]\n\t"
|
|
"move %[pb], %[sb]\n\t"
|
|
"li.d %[r], 4\n\t"
|
|
"xvxor.v $xr8, $xr8, $xr8\n\t"
|
|
"xvxor.v $xr9, $xr9, $xr9\n\t"
|
|
"xvxor.v $xr10, $xr10, $xr10\n\t"
|
|
"xvxor.v $xr11, $xr11, $xr11\n\t"
|
|
"2:\n\t"
|
|
"xvld $xr0, %[pa], 0\n\t"
|
|
"xvld $xr1, %[pb], 0\n\t"
|
|
"add.d %[pa], %[pa], %[sta]\n\t"
|
|
"add.d %[pb], %[pb], %[stb]\n\t"
|
|
"xvhaddw.hu.bu $xr2, $xr0, $xr0\n\t"
|
|
"xvhaddw.hu.bu $xr3, $xr1, $xr1\n\t"
|
|
"xvadd.h $xr8, $xr8, $xr2\n\t"
|
|
"xvadd.h $xr9, $xr9, $xr3\n\t"
|
|
"xvmulwev.h.bu $xr2, $xr0, $xr1\n\t"
|
|
"xvmulwod.h.bu $xr3, $xr0, $xr1\n\t"
|
|
"xvhaddw.wu.hu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.wu.hu $xr3, $xr3, $xr3\n\t"
|
|
"xvadd.w $xr11, $xr11, $xr2\n\t"
|
|
"xvadd.w $xr11, $xr11, $xr3\n\t"
|
|
"xvmulwev.h.bu $xr2, $xr0, $xr0\n\t"
|
|
"xvmulwod.h.bu $xr3, $xr0, $xr0\n\t"
|
|
"xvmulwev.h.bu $xr4, $xr1, $xr1\n\t"
|
|
"xvmulwod.h.bu $xr5, $xr1, $xr1\n\t"
|
|
"xvhaddw.wu.hu $xr2, $xr2, $xr2\n\t"
|
|
"xvhaddw.wu.hu $xr3, $xr3, $xr3\n\t"
|
|
"xvhaddw.wu.hu $xr4, $xr4, $xr4\n\t"
|
|
"xvhaddw.wu.hu $xr5, $xr5, $xr5\n\t"
|
|
"xvadd.w $xr2, $xr2, $xr3\n\t"
|
|
"xvadd.w $xr4, $xr4, $xr5\n\t"
|
|
"xvadd.w $xr10, $xr10, $xr2\n\t"
|
|
"xvadd.w $xr10, $xr10, $xr4\n\t"
|
|
"addi.d %[r], %[r], -1\n\t"
|
|
"bnez %[r], 2b\n\t"
|
|
"xvhaddw.wu.hu $xr8, $xr8, $xr8\n\t"
|
|
"xvhaddw.wu.hu $xr9, $xr9, $xr9\n\t"
|
|
"xvilvl.w $xr0, $xr9, $xr8\n\t"
|
|
"xvilvh.w $xr1, $xr9, $xr8\n\t"
|
|
"xvilvl.w $xr2, $xr11, $xr10\n\t"
|
|
"xvilvh.w $xr3, $xr11, $xr10\n\t"
|
|
"xvilvl.d $xr4, $xr2, $xr0\n\t"
|
|
"xvilvh.d $xr5, $xr2, $xr0\n\t"
|
|
"xvilvl.d $xr6, $xr3, $xr1\n\t"
|
|
"xvilvh.d $xr7, $xr3, $xr1\n\t"
|
|
"xvor.v $xr0, $xr5, $xr5\n\t"
|
|
"xvpermi.q $xr0, $xr4, 0x20\n\t"
|
|
"xvpermi.q $xr5, $xr4, 0x31\n\t"
|
|
"xvor.v $xr1, $xr7, $xr7\n\t"
|
|
"xvpermi.q $xr1, $xr6, 0x20\n\t"
|
|
"xvpermi.q $xr7, $xr6, 0x31\n\t"
|
|
"xvst $xr0, %[dst], 0\n\t"
|
|
"xvst $xr1, %[dst], 32\n\t"
|
|
"xvst $xr5, %[dst], 64\n\t"
|
|
"xvst $xr7, %[dst], 96\n\t"
|
|
"addi.d %[sa], %[sa], 32\n\t"
|
|
"addi.d %[sb], %[sb], 32\n\t"
|
|
"addi.d %[dst], %[dst], 128\n\t"
|
|
"addi.d %[cnt], %[cnt], -32\n\t"
|
|
"bnez %[cnt], 1b\n\t"
|
|
: [sa] "+r"(src_a), [sb] "+r"(src_b), [dst] "+r"(dst),
|
|
[cnt] "+r"(width), [pa] "=&r"(pa), [pb] "=&r"(pb), [r] "=&r"(r)
|
|
: [sta] "r"((ptrdiff_t)stride_a), [stb] "r"((ptrdiff_t)stride_b)
|
|
: "memory", "$xr0", "$xr1", "$xr2", "$xr3", "$xr4", "$xr5", "$xr6",
|
|
"$xr7", "$xr8", "$xr9", "$xr10", "$xr11");
|
|
}
|
|
#endif // HAS_SSIM4X4ROW_LASX
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|
|
|
|
#endif // !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)
|