mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2026-10-28 18:07:12 +08:00
2404f9730e
Reduce loop unrolling across LoongArch LSX (128-bit) and LASX (256-bit) compare, row, rotate, and scale kernels to match SSE2/SSSE3 and AVX2 (1-2 vectors per iteration), rewrite intrinsics/macro implementations into GCC/Clang inline assembly, update all _Any wrapper masks and IS_ALIGNED dispatch checks, and port all remaining LSX Convert (+19) and Rotate (+2) kernels to LASX for full parity: ISA | Compare | Convert | Rotate | Scale -----------------|---------|---------|---------|-------- MIPS/Loong (Any) | 1 | 68 | 4 | 20 LSX | 1 | 68 | 4 | 20 LASX | 1 | 68 | 4 | 0 Removes 1,935 macro/helper lines from loongson_intrinsics.h (-99.3%) and reduces total LoongArch compiled .text size by 9,799 bytes (-25.1% overall, -37.4% row_lsx.cc, -52.3% scale_lsx.cc, -31.2% rotate_lsx.cc, -4.9% row_lasx.cc while adding +25 LASX functions): File | Before | After | Change -------------------|----------|----------|-------- scale_lsx.cc.o | 6,055 B | 2,888 B | -52.3% row_lsx.cc.o | 17,252 B | 10,792 B | -37.4% rotate_lsx.cc.o | 1,488 B | 1,024 B | -31.2% row_lasx.cc.o | 13,740 B | 13,072 B | -4.9% rotate_lasx.cc.o | 544 B | 1,504 B | +176.5% TOTAL .text | 39,079 B | 29,280 B | -25.1% R=wtc@google.com, yuanhecai@loongson.cn Test: qemu-loongarch64 -cpu la464 libyuv_unittest (LASX and LIBYUV_DISABLE_LASX=1) Change-Id: I11c1143ad2aad0399f10b2c3fc9bc4c02c740c59 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8564659 Reviewed-by: richard winterton <rrwinterton@gmail.com>
915 lines
44 KiB
C++
915 lines
44 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 <assert.h>
|
|
|
|
#include "libyuv/scale_row.h"
|
|
|
|
#if !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)
|
|
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
static const uint64_t kScale38Shuf[2] = {0x13100e0b08060300ULL,
|
|
0x000000001e1b1816ULL};
|
|
static const uint64_t kScale38BoxShuf[2] = {0x0a08160604120200ULL,
|
|
0x000000001e0e0c1aULL};
|
|
static const uint64_t kScale34Shuf[2] = {0x0908070504030100ULL,
|
|
0x000000000f0d0c0bULL};
|
|
static const uint64_t kScale34BoxLo[2] = {0x0504030202010100ULL,
|
|
0x0000000007060605ULL};
|
|
static const uint64_t kScale34BoxHi[2] = {0x0d0c0b0a0a090908ULL,
|
|
0x000000000f0e0e0dULL};
|
|
static const uint64_t kScale34BoxWt[2] = {0x0103030102020103ULL,
|
|
0x0000000003010202ULL};
|
|
static const uint64_t kScale34Pack12[2] = {0x0908050403020100ULL,
|
|
0x000000000d0c0b0aULL};
|
|
static const uint64_t kRamp0123[2] = {0x0000000100000000ULL,
|
|
0x0000000300000002ULL};
|
|
|
|
void ScaleARGBRowDown2_LSX(const uint8_t* src_argb,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_argb,
|
|
int dst_width) {
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vpickod.w $vr0, $vr1, $vr0 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_argb), // %[src]
|
|
[dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1");
|
|
}
|
|
|
|
void ScaleARGBRowDown2Linear_LSX(const uint8_t* src_argb,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_argb,
|
|
int dst_width) {
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vpickev.w $vr2, $vr1, $vr0 \n"
|
|
"vpickod.w $vr3, $vr1, $vr0 \n"
|
|
"vavgr.bu $vr0, $vr3, $vr2 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_argb), // %[src]
|
|
[dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3");
|
|
}
|
|
|
|
void ScaleARGBRowDown2Box_LSX(const uint8_t* src_argb,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_argb,
|
|
int dst_width) {
|
|
const uint8_t* t = src_argb + src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[s], 0 \n"
|
|
"vld $vr1, %[s], 16 \n"
|
|
"vld $vr2, %[t], 0 \n"
|
|
"vld $vr3, %[t], 16 \n"
|
|
"vpickev.w $vr4, $vr1, $vr0 \n"
|
|
"vpickod.w $vr5, $vr1, $vr0 \n"
|
|
"vpickev.w $vr6, $vr3, $vr2 \n"
|
|
"vpickod.w $vr7, $vr3, $vr2 \n"
|
|
"vaddwev.h.bu $vr0, $vr4, $vr5 \n"
|
|
"vaddwod.h.bu $vr1, $vr4, $vr5 \n"
|
|
"vaddwev.h.bu $vr2, $vr6, $vr7 \n"
|
|
"vaddwod.h.bu $vr3, $vr6, $vr7 \n"
|
|
"vadd.h $vr0, $vr0, $vr2 \n"
|
|
"vadd.h $vr1, $vr1, $vr3 \n"
|
|
"vsrari.h $vr0, $vr0, 2 \n"
|
|
"vsrari.h $vr1, $vr1, 2 \n"
|
|
"vpackev.b $vr0, $vr1, $vr0 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[s], %[s], 32 \n"
|
|
"addi.d %[t], %[t], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [s] "+r"(src_argb), // %[s]
|
|
[t] "+r"(t), // %[t]
|
|
[dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7");
|
|
}
|
|
|
|
void ScaleARGBRowDownEven_LSX(const uint8_t* src_argb,
|
|
ptrdiff_t src_stride,
|
|
int32_t src_stepx,
|
|
uint8_t* dst_argb,
|
|
int dst_width) {
|
|
const ptrdiff_t step = (ptrdiff_t)src_stepx * 4;
|
|
uint32_t r0, r1, r2, r3;
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"ld.w %[r0], %[src], 0 \n"
|
|
"add.d %[src], %[src], %[step] \n"
|
|
"ld.w %[r1], %[src], 0 \n"
|
|
"add.d %[src], %[src], %[step] \n"
|
|
"ld.w %[r2], %[src], 0 \n"
|
|
"add.d %[src], %[src], %[step] \n"
|
|
"ld.w %[r3], %[src], 0 \n"
|
|
"add.d %[src], %[src], %[step] \n"
|
|
"st.w %[r0], %[dst], 0 \n"
|
|
"st.w %[r1], %[dst], 4 \n"
|
|
"st.w %[r2], %[dst], 8 \n"
|
|
"st.w %[r3], %[dst], 12 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_argb), // %[src]
|
|
[dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width), // %[w]
|
|
[r0] "=&r"(r0), // %[r0]
|
|
[r1] "=&r"(r1), // %[r1]
|
|
[r2] "=&r"(r2), // %[r2]
|
|
[r3] "=&r"(r3) // %[r3]
|
|
: [step] "r"(step) // %[step]
|
|
: "memory");
|
|
}
|
|
|
|
void ScaleARGBRowDownEvenBox_LSX(const uint8_t* src_argb,
|
|
ptrdiff_t src_stride,
|
|
int src_stepx,
|
|
uint8_t* dst_argb,
|
|
int dst_width) {
|
|
const ptrdiff_t step = (ptrdiff_t)src_stepx * 4;
|
|
const uint8_t* next_argb = src_argb + src_stride;
|
|
uint64_t r0, r1;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"ld.d %[r0], %[s], 0 \n"
|
|
"add.d %[s], %[s], %[step] \n"
|
|
"ld.d %[r1], %[s], 0 \n"
|
|
"add.d %[s], %[s], %[step] \n"
|
|
"vinsgr2vr.d $vr0, %[r0], 0 \n"
|
|
"vinsgr2vr.d $vr0, %[r1], 1 \n"
|
|
"ld.d %[r0], %[s], 0 \n"
|
|
"add.d %[s], %[s], %[step] \n"
|
|
"ld.d %[r1], %[s], 0 \n"
|
|
"add.d %[s], %[s], %[step] \n"
|
|
"vinsgr2vr.d $vr1, %[r0], 0 \n"
|
|
"vinsgr2vr.d $vr1, %[r1], 1 \n"
|
|
|
|
"ld.d %[r0], %[t], 0 \n"
|
|
"add.d %[t], %[t], %[step] \n"
|
|
"ld.d %[r1], %[t], 0 \n"
|
|
"add.d %[t], %[t], %[step] \n"
|
|
"vinsgr2vr.d $vr2, %[r0], 0 \n"
|
|
"vinsgr2vr.d $vr2, %[r1], 1 \n"
|
|
"ld.d %[r0], %[t], 0 \n"
|
|
"add.d %[t], %[t], %[step] \n"
|
|
"ld.d %[r1], %[t], 0 \n"
|
|
"add.d %[t], %[t], %[step] \n"
|
|
"vinsgr2vr.d $vr3, %[r0], 0 \n"
|
|
"vinsgr2vr.d $vr3, %[r1], 1 \n"
|
|
|
|
"vpickev.w $vr4, $vr1, $vr0 \n"
|
|
"vpickod.w $vr5, $vr1, $vr0 \n"
|
|
"vpickev.w $vr6, $vr3, $vr2 \n"
|
|
"vpickod.w $vr7, $vr3, $vr2 \n"
|
|
"vaddwev.h.bu $vr0, $vr4, $vr5 \n"
|
|
"vaddwod.h.bu $vr1, $vr4, $vr5 \n"
|
|
"vaddwev.h.bu $vr2, $vr6, $vr7 \n"
|
|
"vaddwod.h.bu $vr3, $vr6, $vr7 \n"
|
|
"vadd.h $vr0, $vr0, $vr2 \n"
|
|
"vadd.h $vr1, $vr1, $vr3 \n"
|
|
"vsrari.h $vr0, $vr0, 2 \n"
|
|
"vsrari.h $vr1, $vr1, 2 \n"
|
|
"vpackev.b $vr0, $vr1, $vr0 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [s] "+r"(src_argb), // %[s]
|
|
[t] "+r"(next_argb), // %[t]
|
|
[dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width), // %[w]
|
|
[r0] "=&r"(r0), // %[r0]
|
|
[r1] "=&r"(r1) // %[r1]
|
|
: [step] "r"(step) // %[step]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7");
|
|
}
|
|
|
|
void ScaleRowDown2_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 4 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vpickod.b $vr0, $vr1, $vr0 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1");
|
|
}
|
|
|
|
void ScaleRowDown2Linear_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 4 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vpickev.b $vr2, $vr1, $vr0 \n"
|
|
"vpickod.b $vr3, $vr1, $vr0 \n"
|
|
"vavgr.bu $vr0, $vr3, $vr2 \n"
|
|
"vst $vr0, %[dst], 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3");
|
|
}
|
|
|
|
void ScaleRowDown2Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
const uint8_t* src_nex = src_ptr + src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 4 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vld $vr2, %[nex], 0 \n"
|
|
"vld $vr3, %[nex], 16 \n"
|
|
"vaddwev.h.bu $vr4, $vr0, $vr2 \n"
|
|
"vaddwod.h.bu $vr5, $vr0, $vr2 \n"
|
|
"vaddwev.h.bu $vr6, $vr1, $vr3 \n"
|
|
"vaddwod.h.bu $vr7, $vr1, $vr3 \n"
|
|
"vadd.h $vr4, $vr4, $vr5 \n"
|
|
"vadd.h $vr6, $vr6, $vr7 \n"
|
|
"vsrarni.b.h $vr6, $vr4, 2 \n"
|
|
"vst $vr6, %[dst], 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[nex], %[nex], 32 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[nex] "+r"(src_nex), // %[nex]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7");
|
|
}
|
|
|
|
void ScaleRowDown4_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 3 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vpickev.b $vr0, $vr1, $vr0 \n"
|
|
"vpickod.b $vr0, $vr0, $vr0 \n"
|
|
"vstelm.d $vr0, %[dst], 0, 0 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 8 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1");
|
|
}
|
|
|
|
void ScaleRowDown4Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
const uint8_t* ptr1 = src_ptr + src_stride;
|
|
const uint8_t* ptr2 = ptr1 + src_stride;
|
|
const uint8_t* ptr3 = ptr2 + src_stride;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 3 \n"
|
|
"beqz %[w], 2f \n"
|
|
"1: \n"
|
|
"vld $vr0, %[p0], 0 \n"
|
|
"vld $vr1, %[p0], 16 \n"
|
|
"vld $vr2, %[p1], 0 \n"
|
|
"vld $vr3, %[p1], 16 \n"
|
|
"vld $vr4, %[p2], 0 \n"
|
|
"vld $vr5, %[p2], 16 \n"
|
|
"vld $vr6, %[p3], 0 \n"
|
|
"vld $vr7, %[p3], 16 \n"
|
|
"vhaddw.hu.bu $vr0, $vr0, $vr0 \n"
|
|
"vhaddw.hu.bu $vr1, $vr1, $vr1 \n"
|
|
"vhaddw.hu.bu $vr2, $vr2, $vr2 \n"
|
|
"vhaddw.hu.bu $vr3, $vr3, $vr3 \n"
|
|
"vhaddw.hu.bu $vr4, $vr4, $vr4 \n"
|
|
"vhaddw.hu.bu $vr5, $vr5, $vr5 \n"
|
|
"vhaddw.hu.bu $vr6, $vr6, $vr6 \n"
|
|
"vhaddw.hu.bu $vr7, $vr7, $vr7 \n"
|
|
"vadd.h $vr0, $vr0, $vr2 \n"
|
|
"vadd.h $vr1, $vr1, $vr3 \n"
|
|
"vadd.h $vr4, $vr4, $vr6 \n"
|
|
"vadd.h $vr5, $vr5, $vr7 \n"
|
|
"vadd.h $vr0, $vr0, $vr4 \n"
|
|
"vadd.h $vr1, $vr1, $vr5 \n"
|
|
"vhaddw.wu.hu $vr0, $vr0, $vr0 \n"
|
|
"vhaddw.wu.hu $vr1, $vr1, $vr1 \n"
|
|
"vsrarni.h.w $vr1, $vr0, 4 \n"
|
|
"vsrlni.b.h $vr1, $vr1, 0 \n"
|
|
"vstelm.d $vr1, %[dst], 0, 0 \n"
|
|
"addi.d %[p0], %[p0], 32 \n"
|
|
"addi.d %[p1], %[p1], 32 \n"
|
|
"addi.d %[p2], %[p2], 32 \n"
|
|
"addi.d %[p3], %[p3], 32 \n"
|
|
"addi.d %[dst], %[dst], 8 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [p0] "+r"(src_ptr), // %[p0]
|
|
[p1] "+r"(ptr1), // %[p1]
|
|
[p2] "+r"(ptr2), // %[p2]
|
|
[p3] "+r"(ptr3), // %[p3]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[w] "+r"(dst_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7");
|
|
}
|
|
|
|
void ScaleRowDown38_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
int count = dst_width / 12;
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vld $vr2, %[shuf], 0 \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vshuf.b $vr0, $vr1, $vr0, $vr2 \n"
|
|
"vstelm.d $vr0, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr0, %[dst], 8, 2 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [shuf] "r"(kScale38Shuf) // %[shuf]
|
|
: "memory", "$vr0", "$vr1", "$vr2");
|
|
}
|
|
|
|
void ScaleRowDown38_2_Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_ptr,
|
|
int dst_width) {
|
|
const uint8_t* src_nex = src_ptr + src_stride;
|
|
int count = dst_width / 12;
|
|
int c2aaa = 0x2aaa;
|
|
int c4000 = 0x4000;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vld $vr8, %[shuf], 0 \n"
|
|
"vreplgr2vr.h $vr9, %[c2aaa] \n"
|
|
"vreplgr2vr.w $vr10, %[c4000] \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vld $vr1, %[src], 16 \n"
|
|
"vld $vr2, %[nex], 0 \n"
|
|
"vld $vr3, %[nex], 16 \n"
|
|
"vaddwev.h.bu $vr4, $vr0, $vr2 \n"
|
|
"vaddwod.h.bu $vr5, $vr0, $vr2 \n"
|
|
"vaddwev.h.bu $vr6, $vr1, $vr3 \n"
|
|
"vaddwod.h.bu $vr7, $vr1, $vr3 \n"
|
|
"vpickev.h $vr0, $vr6, $vr4 \n"
|
|
"vpickev.h $vr1, $vr7, $vr5 \n"
|
|
"vpackod.h $vr2, $vr5, $vr4 \n"
|
|
"vpackod.h $vr3, $vr7, $vr6 \n"
|
|
"vpickev.w $vr4, $vr3, $vr2 \n"
|
|
"vadd.h $vr0, $vr0, $vr1 \n"
|
|
"vadd.h $vr0, $vr0, $vr4 \n"
|
|
"vmuh.h $vr0, $vr0, $vr9 \n"
|
|
"vpickod.w $vr2, $vr3, $vr2 \n"
|
|
"vhaddw.wu.hu $vr2, $vr2, $vr2 \n"
|
|
"vmul.w $vr2, $vr2, $vr10 \n"
|
|
"vshuf.b $vr0, $vr2, $vr0, $vr8 \n"
|
|
"vstelm.d $vr0, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr0, %[dst], 8, 2 \n"
|
|
"addi.d %[src], %[src], 32 \n"
|
|
"addi.d %[nex], %[nex], 32 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[nex] "+r"(src_nex), // %[nex]
|
|
[dst] "+r"(dst_ptr), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [shuf] "r"(kScale38BoxShuf), // %[shuf]
|
|
[c2aaa] "r"(c2aaa), // %[c2aaa]
|
|
[c4000] "r"(c4000) // %[c4000]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7", "$vr8", "$vr9", "$vr10");
|
|
}
|
|
|
|
void ScaleRowDown38_3_Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_ptr,
|
|
int dst_width) {
|
|
const uint8_t* ptr1 = src_ptr + src_stride;
|
|
const uint8_t* ptr2 = ptr1 + src_stride;
|
|
int count = dst_width / 12;
|
|
int c1c71 = 0x1c71;
|
|
int c2aaa = 0x2aaa;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vldi $vr11, 0 \n"
|
|
"vld $vr8, %[shuf], 0 \n"
|
|
"vreplgr2vr.h $vr9, %[c1c71] \n"
|
|
"vreplgr2vr.w $vr10, %[c2aaa] \n"
|
|
"1: \n"
|
|
"vld $vr0, %[p0], 0 \n"
|
|
"vld $vr1, %[p0], 16 \n"
|
|
"vld $vr2, %[p1], 0 \n"
|
|
"vld $vr3, %[p1], 16 \n"
|
|
"vaddwev.h.bu $vr4, $vr0, $vr2 \n"
|
|
"vaddwod.h.bu $vr5, $vr0, $vr2 \n"
|
|
"vaddwev.h.bu $vr6, $vr1, $vr3 \n"
|
|
"vaddwod.h.bu $vr7, $vr1, $vr3 \n"
|
|
"vld $vr0, %[p2], 0 \n"
|
|
"vld $vr1, %[p2], 16 \n"
|
|
"vpackev.b $vr2, $vr11, $vr0 \n"
|
|
"vpackod.b $vr3, $vr11, $vr0 \n"
|
|
"vadd.h $vr4, $vr4, $vr2 \n"
|
|
"vadd.h $vr5, $vr5, $vr3 \n"
|
|
"vpackev.b $vr2, $vr11, $vr1 \n"
|
|
"vpackod.b $vr3, $vr11, $vr1 \n"
|
|
"vadd.h $vr6, $vr6, $vr2 \n"
|
|
"vadd.h $vr7, $vr7, $vr3 \n"
|
|
"vpickev.h $vr0, $vr6, $vr4 \n"
|
|
"vpickev.h $vr1, $vr7, $vr5 \n"
|
|
"vpackod.h $vr2, $vr5, $vr4 \n"
|
|
"vpackod.h $vr3, $vr7, $vr6 \n"
|
|
"vpickev.w $vr4, $vr3, $vr2 \n"
|
|
"vadd.h $vr0, $vr0, $vr1 \n"
|
|
"vadd.h $vr0, $vr0, $vr4 \n"
|
|
"vmuh.h $vr0, $vr0, $vr9 \n"
|
|
"vpickod.w $vr2, $vr3, $vr2 \n"
|
|
"vhaddw.wu.hu $vr2, $vr2, $vr2 \n"
|
|
"vmul.w $vr2, $vr2, $vr10 \n"
|
|
"vshuf.b $vr0, $vr2, $vr0, $vr8 \n"
|
|
"vstelm.d $vr0, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr0, %[dst], 8, 2 \n"
|
|
"addi.d %[p0], %[p0], 32 \n"
|
|
"addi.d %[p1], %[p1], 32 \n"
|
|
"addi.d %[p2], %[p2], 32 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [p0] "+r"(src_ptr), // %[p0]
|
|
[p1] "+r"(ptr1), // %[p1]
|
|
[p2] "+r"(ptr2), // %[p2]
|
|
[dst] "+r"(dst_ptr), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [shuf] "r"(kScale38BoxShuf), // %[shuf]
|
|
[c1c71] "r"(c1c71), // %[c1c71]
|
|
[c2aaa] "r"(c2aaa) // %[c2aaa]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7", "$vr8", "$vr9", "$vr10", "$vr11");
|
|
}
|
|
|
|
void ScaleAddRow_LSX(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) {
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 4 \n"
|
|
"beqz %[w], 2f \n"
|
|
"vldi $vr0, 0 \n"
|
|
"1: \n"
|
|
"vld $vr1, %[src], 0 \n"
|
|
"vld $vr2, %[dst], 0 \n"
|
|
"vld $vr3, %[dst], 16 \n"
|
|
"vilvl.b $vr4, $vr0, $vr1 \n"
|
|
"vilvh.b $vr5, $vr0, $vr1 \n"
|
|
"vadd.h $vr2, $vr2, $vr4 \n"
|
|
"vadd.h $vr3, $vr3, $vr5 \n"
|
|
"vst $vr2, %[dst], 0 \n"
|
|
"vst $vr3, %[dst], 16 \n"
|
|
"addi.d %[src], %[src], 16 \n"
|
|
"addi.d %[dst], %[dst], 32 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst_ptr), // %[dst]
|
|
[w] "+r"(src_width) // %[w]
|
|
:
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5");
|
|
}
|
|
|
|
void ScaleFilterCols_LSX(uint8_t* dst_ptr,
|
|
const uint8_t* src_ptr,
|
|
int dst_width,
|
|
int x,
|
|
int dx) {
|
|
ptrdiff_t i0, i1, i2, i3;
|
|
uint32_t p0, p1;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"vld $vr0, %[ramp], 0 \n"
|
|
"vreplgr2vr.w $vr1, %[dx] \n"
|
|
"vreplgr2vr.w $vr2, %[x] \n"
|
|
"vmul.w $vr0, $vr1, $vr0 \n"
|
|
"vslli.w $vr1, $vr1, 2 \n"
|
|
"vadd.w $vr2, $vr2, $vr0 \n"
|
|
"vldi $vr3, 0x7f \n"
|
|
"vsrli.w $vr3, $vr3, 24 \n"
|
|
"vaddi.wu $vr4, $vr3, 1 \n"
|
|
"1: \n"
|
|
"vsrai.w $vr5, $vr2, 16 \n"
|
|
"vsrai.w $vr6, $vr2, 9 \n"
|
|
"vand.v $vr6, $vr6, $vr3 \n"
|
|
"vsub.w $vr7, $vr4, $vr6 \n"
|
|
"vslli.w $vr6, $vr6, 8 \n"
|
|
"vor.v $vr6, $vr6, $vr7 \n"
|
|
"vadd.w $vr2, $vr2, $vr1 \n"
|
|
"vpickve2gr.w %[i0], $vr5, 0 \n"
|
|
"vpickve2gr.w %[i1], $vr5, 1 \n"
|
|
"vpickve2gr.w %[i2], $vr5, 2 \n"
|
|
"vpickve2gr.w %[i3], $vr5, 3 \n"
|
|
"ldx.h %[p0], %[src], %[i0] \n"
|
|
"ldx.h %[p1], %[src], %[i1] \n"
|
|
"vinsgr2vr.h $vr5, %[p0], 0 \n"
|
|
"vinsgr2vr.h $vr5, %[p1], 2 \n"
|
|
"ldx.h %[p0], %[src], %[i2] \n"
|
|
"ldx.h %[p1], %[src], %[i3] \n"
|
|
"vinsgr2vr.h $vr5, %[p0], 4 \n"
|
|
"vinsgr2vr.h $vr5, %[p1], 6 \n"
|
|
"vmulwev.h.bu $vr7, $vr5, $vr6 \n"
|
|
"vmaddwod.h.bu $vr7, $vr5, $vr6 \n"
|
|
"vsrari.h $vr5, $vr7, 7 \n"
|
|
"vpickev.b $vr5, $vr5, $vr5 \n"
|
|
"vpickev.b $vr5, $vr5, $vr5 \n"
|
|
"vstelm.w $vr5, %[dst], 0, 0 \n"
|
|
"addi.d %[dst], %[dst], 4 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [dst] "+r"(dst_ptr), // %[dst]
|
|
[w] "+r"(dst_width), // %[w]
|
|
[i0] "=&r"(i0), // %[i0]
|
|
[i1] "=&r"(i1), // %[i1]
|
|
[i2] "=&r"(i2), // %[i2]
|
|
[i3] "=&r"(i3), // %[i3]
|
|
[p0] "=&r"(p0), // %[p0]
|
|
[p1] "=&r"(p1) // %[p1]
|
|
: [src] "r"(src_ptr), // %[src]
|
|
[x] "r"(x), // %[x]
|
|
[dx] "r"(dx), // %[dx]
|
|
[ramp] "r"(kRamp0123) // %[ramp]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7");
|
|
}
|
|
|
|
void ScaleARGBCols_LSX(uint8_t* dst_argb,
|
|
const uint8_t* src_argb,
|
|
int dst_width,
|
|
int x,
|
|
int dx) {
|
|
ptrdiff_t i0, i1, i2, i3;
|
|
uint32_t p0, p1;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"vld $vr0, %[ramp], 0 \n"
|
|
"vreplgr2vr.w $vr1, %[dx] \n"
|
|
"vreplgr2vr.w $vr2, %[x] \n"
|
|
"vmul.w $vr0, $vr1, $vr0 \n"
|
|
"vslli.w $vr1, $vr1, 2 \n"
|
|
"vadd.w $vr2, $vr2, $vr0 \n"
|
|
"1: \n"
|
|
"vsrai.w $vr3, $vr2, 16 \n"
|
|
"vslli.w $vr3, $vr3, 2 \n"
|
|
"vadd.w $vr2, $vr2, $vr1 \n"
|
|
"vpickve2gr.w %[i0], $vr3, 0 \n"
|
|
"vpickve2gr.w %[i1], $vr3, 1 \n"
|
|
"vpickve2gr.w %[i2], $vr3, 2 \n"
|
|
"vpickve2gr.w %[i3], $vr3, 3 \n"
|
|
"ldx.w %[p0], %[src], %[i0] \n"
|
|
"ldx.w %[p1], %[src], %[i1] \n"
|
|
"st.w %[p0], %[dst], 0 \n"
|
|
"st.w %[p1], %[dst], 4 \n"
|
|
"ldx.w %[p0], %[src], %[i2] \n"
|
|
"ldx.w %[p1], %[src], %[i3] \n"
|
|
"st.w %[p0], %[dst], 8 \n"
|
|
"st.w %[p1], %[dst], 12 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width), // %[w]
|
|
[i0] "=&r"(i0), // %[i0]
|
|
[i1] "=&r"(i1), // %[i1]
|
|
[i2] "=&r"(i2), // %[i2]
|
|
[i3] "=&r"(i3), // %[i3]
|
|
[p0] "=&r"(p0), // %[p0]
|
|
[p1] "=&r"(p1) // %[p1]
|
|
: [src] "r"(src_argb), // %[src]
|
|
[x] "r"(x), // %[x]
|
|
[dx] "r"(dx), // %[dx]
|
|
[ramp] "r"(kRamp0123) // %[ramp]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3");
|
|
}
|
|
|
|
void ScaleARGBFilterCols_LSX(uint8_t* dst_argb,
|
|
const uint8_t* src_argb,
|
|
int dst_width,
|
|
int x,
|
|
int dx) {
|
|
ptrdiff_t i0, i1, i2, i3;
|
|
uint64_t p0, p1;
|
|
asm volatile(
|
|
"srli.w %[w], %[w], 2 \n"
|
|
"beqz %[w], 2f \n"
|
|
"vld $vr0, %[ramp], 0 \n"
|
|
"vreplgr2vr.w $vr1, %[dx] \n"
|
|
"vreplgr2vr.w $vr2, %[x] \n"
|
|
"vmul.w $vr0, $vr1, $vr0 \n"
|
|
"vslli.w $vr1, $vr1, 2 \n"
|
|
"vadd.w $vr2, $vr2, $vr0 \n"
|
|
"vldi $vr3, 0x7f \n"
|
|
"1: \n"
|
|
"vsrai.w $vr4, $vr2, 16 \n"
|
|
"vslli.w $vr4, $vr4, 2 \n"
|
|
"vsrai.w $vr5, $vr2, 9 \n"
|
|
"vadd.w $vr2, $vr2, $vr1 \n"
|
|
"vand.v $vr5, $vr5, $vr3 \n"
|
|
"vshuf4i.b $vr5, $vr5, 0x00 \n"
|
|
"vxor.v $vr6, $vr5, $vr3 \n"
|
|
"vilvl.b $vr7, $vr5, $vr6 \n"
|
|
"vilvh.b $vr8, $vr5, $vr6 \n"
|
|
"vpickve2gr.w %[i0], $vr4, 0 \n"
|
|
"vpickve2gr.w %[i1], $vr4, 1 \n"
|
|
"vpickve2gr.w %[i2], $vr4, 2 \n"
|
|
"vpickve2gr.w %[i3], $vr4, 3 \n"
|
|
"ldx.d %[p0], %[src], %[i0] \n"
|
|
"ldx.d %[p1], %[src], %[i1] \n"
|
|
"vinsgr2vr.d $vr4, %[p0], 0 \n"
|
|
"vinsgr2vr.d $vr4, %[p1], 1 \n"
|
|
"ldx.d %[p0], %[src], %[i2] \n"
|
|
"ldx.d %[p1], %[src], %[i3] \n"
|
|
"vinsgr2vr.d $vr5, %[p0], 0 \n"
|
|
"vinsgr2vr.d $vr5, %[p1], 1 \n"
|
|
"vpickev.w $vr6, $vr5, $vr4 \n"
|
|
"vpickod.w $vr4, $vr5, $vr4 \n"
|
|
"vilvl.b $vr5, $vr4, $vr6 \n"
|
|
"vilvh.b $vr4, $vr4, $vr6 \n"
|
|
"vmulwev.h.bu $vr6, $vr5, $vr7 \n"
|
|
"vmaddwod.h.bu $vr6, $vr5, $vr7 \n"
|
|
"vmulwev.h.bu $vr7, $vr4, $vr8 \n"
|
|
"vmaddwod.h.bu $vr7, $vr4, $vr8 \n"
|
|
"vsrani.b.h $vr7, $vr6, 7 \n"
|
|
"vst $vr7, %[dst], 0 \n"
|
|
"addi.d %[dst], %[dst], 16 \n"
|
|
"addi.w %[w], %[w], -1 \n"
|
|
"bnez %[w], 1b \n"
|
|
"2: \n"
|
|
: [dst] "+r"(dst_argb), // %[dst]
|
|
[w] "+r"(dst_width), // %[w]
|
|
[i0] "=&r"(i0), // %[i0]
|
|
[i1] "=&r"(i1), // %[i1]
|
|
[i2] "=&r"(i2), // %[i2]
|
|
[i3] "=&r"(i3), // %[i3]
|
|
[p0] "=&r"(p0), // %[p0]
|
|
[p1] "=&r"(p1) // %[p1]
|
|
: [src] "r"(src_argb), // %[src]
|
|
[x] "r"(x), // %[x]
|
|
[dx] "r"(dx), // %[dx]
|
|
[ramp] "r"(kRamp0123) // %[ramp]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7", "$vr8");
|
|
}
|
|
|
|
void ScaleRowDown34_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
int count = dst_width / 12;
|
|
(void)src_stride;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vld $vr1, %[shuf], 0 \n"
|
|
"1: \n"
|
|
"vld $vr0, %[src], 0 \n"
|
|
"vshuf.b $vr0, $vr0, $vr0, $vr1 \n"
|
|
"vstelm.d $vr0, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr0, %[dst], 8, 2 \n"
|
|
"addi.d %[src], %[src], 16 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[dst] "+r"(dst), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [shuf] "r"(kScale34Shuf) // %[shuf]
|
|
: "memory", "$vr0", "$vr1");
|
|
}
|
|
|
|
void ScaleRowDown34_0_Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* d,
|
|
int dst_width) {
|
|
const uint8_t* src_nex = src_ptr + src_stride;
|
|
int count = dst_width / 12;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vld $vr0, %[lo], 0 \n"
|
|
"vld $vr1, %[hi], 0 \n"
|
|
"vld $vr2, %[wt], 0 \n"
|
|
"vld $vr3, %[pack], 0 \n"
|
|
"1: \n"
|
|
"vld $vr4, %[src], 0 \n"
|
|
"vld $vr5, %[nex], 0 \n"
|
|
"vshuf.b $vr6, $vr4, $vr4, $vr0 \n"
|
|
"vshuf.b $vr7, $vr4, $vr4, $vr1 \n"
|
|
"vshuf.b $vr8, $vr5, $vr5, $vr0 \n"
|
|
"vshuf.b $vr9, $vr5, $vr5, $vr1 \n"
|
|
"vmulwev.h.bu $vr4, $vr6, $vr2 \n"
|
|
"vmaddwod.h.bu $vr4, $vr6, $vr2 \n"
|
|
"vmulwev.h.bu $vr5, $vr7, $vr2 \n"
|
|
"vmaddwod.h.bu $vr5, $vr7, $vr2 \n"
|
|
"vmulwev.h.bu $vr6, $vr8, $vr2 \n"
|
|
"vmaddwod.h.bu $vr6, $vr8, $vr2 \n"
|
|
"vmulwev.h.bu $vr7, $vr9, $vr2 \n"
|
|
"vmaddwod.h.bu $vr7, $vr9, $vr2 \n"
|
|
"vsrari.h $vr4, $vr4, 2 \n"
|
|
"vsrari.h $vr5, $vr5, 2 \n"
|
|
"vsrari.h $vr6, $vr6, 2 \n"
|
|
"vsrari.h $vr7, $vr7, 2 \n"
|
|
"vslli.h $vr8, $vr4, 1 \n"
|
|
"vslli.h $vr9, $vr5, 1 \n"
|
|
"vadd.h $vr4, $vr4, $vr8 \n"
|
|
"vadd.h $vr5, $vr5, $vr9 \n"
|
|
"vadd.h $vr4, $vr4, $vr6 \n"
|
|
"vadd.h $vr5, $vr5, $vr7 \n"
|
|
"vsrarni.b.h $vr5, $vr4, 2 \n"
|
|
"vshuf.b $vr5, $vr5, $vr5, $vr3 \n"
|
|
"vstelm.d $vr5, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr5, %[dst], 8, 2 \n"
|
|
"addi.d %[src], %[src], 16 \n"
|
|
"addi.d %[nex], %[nex], 16 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[nex] "+r"(src_nex), // %[nex]
|
|
[dst] "+r"(d), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [lo] "r"(kScale34BoxLo), // %[lo]
|
|
[hi] "r"(kScale34BoxHi), // %[hi]
|
|
[wt] "r"(kScale34BoxWt), // %[wt]
|
|
[pack] "r"(kScale34Pack12) // %[pack]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7", "$vr8", "$vr9");
|
|
}
|
|
|
|
void ScaleRowDown34_1_Box_LSX(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* d,
|
|
int dst_width) {
|
|
const uint8_t* src_nex = src_ptr + src_stride;
|
|
int count = dst_width / 12;
|
|
asm volatile(
|
|
"beqz %[cnt], 2f \n"
|
|
"vld $vr0, %[lo], 0 \n"
|
|
"vld $vr1, %[hi], 0 \n"
|
|
"vld $vr2, %[wt], 0 \n"
|
|
"vld $vr3, %[pack], 0 \n"
|
|
"1: \n"
|
|
"vld $vr4, %[src], 0 \n"
|
|
"vld $vr5, %[nex], 0 \n"
|
|
"vshuf.b $vr6, $vr4, $vr4, $vr0 \n"
|
|
"vshuf.b $vr7, $vr4, $vr4, $vr1 \n"
|
|
"vshuf.b $vr8, $vr5, $vr5, $vr0 \n"
|
|
"vshuf.b $vr9, $vr5, $vr5, $vr1 \n"
|
|
"vmulwev.h.bu $vr4, $vr6, $vr2 \n"
|
|
"vmaddwod.h.bu $vr4, $vr6, $vr2 \n"
|
|
"vmulwev.h.bu $vr5, $vr7, $vr2 \n"
|
|
"vmaddwod.h.bu $vr5, $vr7, $vr2 \n"
|
|
"vmulwev.h.bu $vr6, $vr8, $vr2 \n"
|
|
"vmaddwod.h.bu $vr6, $vr8, $vr2 \n"
|
|
"vmulwev.h.bu $vr7, $vr9, $vr2 \n"
|
|
"vmaddwod.h.bu $vr7, $vr9, $vr2 \n"
|
|
"vsrari.h $vr4, $vr4, 2 \n"
|
|
"vsrari.h $vr5, $vr5, 2 \n"
|
|
"vsrari.h $vr6, $vr6, 2 \n"
|
|
"vsrari.h $vr7, $vr7, 2 \n"
|
|
"vadd.h $vr4, $vr4, $vr6 \n"
|
|
"vadd.h $vr5, $vr5, $vr7 \n"
|
|
"vsrarni.b.h $vr5, $vr4, 1 \n"
|
|
"vshuf.b $vr5, $vr5, $vr5, $vr3 \n"
|
|
"vstelm.d $vr5, %[dst], 0, 0 \n"
|
|
"vstelm.w $vr5, %[dst], 8, 2 \n"
|
|
"addi.d %[src], %[src], 16 \n"
|
|
"addi.d %[nex], %[nex], 16 \n"
|
|
"addi.d %[dst], %[dst], 12 \n"
|
|
"addi.w %[cnt], %[cnt], -1 \n"
|
|
"bnez %[cnt], 1b \n"
|
|
"2: \n"
|
|
: [src] "+r"(src_ptr), // %[src]
|
|
[nex] "+r"(src_nex), // %[nex]
|
|
[dst] "+r"(d), // %[dst]
|
|
[cnt] "+r"(count) // %[cnt]
|
|
: [lo] "r"(kScale34BoxLo), // %[lo]
|
|
[hi] "r"(kScale34BoxHi), // %[hi]
|
|
[wt] "r"(kScale34BoxWt), // %[wt]
|
|
[pack] "r"(kScale34Pack12) // %[pack]
|
|
: "memory", "$vr0", "$vr1", "$vr2", "$vr3", "$vr4", "$vr5", "$vr6",
|
|
"$vr7", "$vr8", "$vr9");
|
|
}
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|
|
|
|
#endif // !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)
|