mirror of
https://github.com/webmproject/libwebp.git
synced 2024-12-27 06:08:21 +01:00
rescaler: simplify ImportRow logic
incorporates the loop over 'channel' and removes one parameter Change-Id: I4e3b33c111ca825fe96461583420413b17326409
This commit is contained in:
Pascal Massimino
parent
5ff0079ece
commit
9ba1894b9b
@ -338,8 +338,7 @@ struct WebPRescaler;
|
||||
// 'channel' denotes the channel number to be imported. 'Expand' corresponds to
|
||||
// the wrk->x_expand case. Otherwise, 'Shrink' is to be used.
|
||||
typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk,
|
||||
const uint8_t* const src,
|
||||
int channel);
|
||||
const uint8_t* src);
|
||||
|
||||
extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand;
|
||||
extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
|
||||
@ -353,15 +352,15 @@ extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
|
||||
|
||||
// Plain-C implementation, as fall-back.
|
||||
extern void WebPRescalerImportRowExpandC(struct WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel);
|
||||
const uint8_t* src);
|
||||
extern void WebPRescalerImportRowShrinkC(struct WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel);
|
||||
const uint8_t* src);
|
||||
extern void WebPRescalerExportRowExpandC(struct WebPRescaler* const wrk);
|
||||
extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk);
|
||||
|
||||
// Main entry calls:
|
||||
extern void WebPRescalerImportRow(struct WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel);
|
||||
const uint8_t* src);
|
||||
// Export one row (starting at x_out position) from rescaler.
|
||||
extern void WebPRescalerExportRow(struct WebPRescaler* const wrk);
|
||||
|
||||
|
||||
@ -23,66 +23,69 @@
|
||||
//------------------------------------------------------------------------------
|
||||
// Row import
|
||||
|
||||
void WebPRescalerImportRowExpandC(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
void WebPRescalerImportRowExpandC(WebPRescaler* wrk, const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
int x_in = channel;
|
||||
int x_out;
|
||||
// simple bilinear interpolation
|
||||
int accum = wrk->x_add;
|
||||
int left = src[x_in];
|
||||
int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
|
||||
x_in += x_stride;
|
||||
x_out = channel;
|
||||
|
||||
int channel;
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
assert(wrk->x_expand);
|
||||
while (1) {
|
||||
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
|
||||
x_out += x_stride;
|
||||
if (x_out >= x_out_max) break;
|
||||
accum -= wrk->x_sub;
|
||||
if (accum < 0) {
|
||||
left = right;
|
||||
x_in += x_stride;
|
||||
assert(x_in < wrk->src_width * x_stride);
|
||||
right = src[x_in];
|
||||
accum += wrk->x_add;
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
int x_in = channel;
|
||||
int x_out = channel;
|
||||
// simple bilinear interpolation
|
||||
int accum = wrk->x_add;
|
||||
int left = src[x_in];
|
||||
int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
|
||||
x_in += x_stride;
|
||||
while (1) {
|
||||
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
|
||||
x_out += x_stride;
|
||||
if (x_out >= x_out_max) break;
|
||||
accum -= wrk->x_sub;
|
||||
if (accum < 0) {
|
||||
left = right;
|
||||
x_in += x_stride;
|
||||
assert(x_in < wrk->src_width * x_stride);
|
||||
right = src[x_in];
|
||||
accum += wrk->x_add;
|
||||
}
|
||||
}
|
||||
assert(wrk->x_sub == 0 /* <- special case for src_width=1 */ || accum == 0);
|
||||
}
|
||||
assert(wrk->x_sub == 0 /* <- special case for src_width=1 */ || accum == 0);
|
||||
}
|
||||
|
||||
void WebPRescalerImportRowShrinkC(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
int x_in = channel;
|
||||
int x_out;
|
||||
uint32_t sum = 0;
|
||||
int accum = 0;
|
||||
|
||||
int channel;
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
assert(!wrk->x_expand);
|
||||
for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
|
||||
uint32_t base = 0;
|
||||
accum += wrk->x_add;
|
||||
while (accum > 0) {
|
||||
accum -= wrk->x_sub;
|
||||
assert(x_in < wrk->src_width * x_stride);
|
||||
base = src[x_in];
|
||||
sum += base;
|
||||
x_in += x_stride;
|
||||
}
|
||||
{ // Emit next horizontal pixel.
|
||||
const rescaler_t frac = base * (-accum);
|
||||
wrk->frow[x_out] = sum * wrk->x_sub - frac;
|
||||
// fresh fractional start for next pixel
|
||||
sum = (int)MULT_FIX(frac, wrk->fx_scale);
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
int x_in = channel;
|
||||
int x_out = channel;
|
||||
uint32_t sum = 0;
|
||||
int accum = 0;
|
||||
while (x_out < x_out_max) {
|
||||
uint32_t base = 0;
|
||||
accum += wrk->x_add;
|
||||
while (accum > 0) {
|
||||
accum -= wrk->x_sub;
|
||||
assert(x_in < wrk->src_width * x_stride);
|
||||
base = src[x_in];
|
||||
sum += base;
|
||||
x_in += x_stride;
|
||||
}
|
||||
{ // Emit next horizontal pixel.
|
||||
const rescaler_t frac = base * (-accum);
|
||||
wrk->frow[x_out] = sum * wrk->x_sub - frac;
|
||||
// fresh fractional start for next pixel
|
||||
sum = (int)MULT_FIX(frac, wrk->fx_scale);
|
||||
}
|
||||
x_out += x_stride;
|
||||
}
|
||||
assert(accum == 0);
|
||||
}
|
||||
assert(accum == 0);
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
@ -145,13 +148,12 @@ void WebPRescalerExportRowShrinkC(WebPRescaler* const wrk) {
|
||||
//------------------------------------------------------------------------------
|
||||
// Main entry calls
|
||||
|
||||
void WebPRescalerImportRow(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
void WebPRescalerImportRow(WebPRescaler* const wrk, const uint8_t* const src) {
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
if (!wrk->x_expand) {
|
||||
WebPRescalerImportRowShrink(wrk, src, channel);
|
||||
WebPRescalerImportRowShrink(wrk, src);
|
||||
} else {
|
||||
WebPRescalerImportRowExpand(wrk, src, channel);
|
||||
WebPRescalerImportRowExpand(wrk, src);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -18,117 +18,123 @@
|
||||
#include <assert.h>
|
||||
#include "../utils/rescaler.h"
|
||||
|
||||
static void ImportRowShrink(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
const int fx_scale = wrk->fx_scale;
|
||||
const int x_add = wrk->x_add;
|
||||
const int x_sub = wrk->x_sub;
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp1, temp2, temp3;
|
||||
int base, frac, sum;
|
||||
int accum, accum1;
|
||||
const int x_stride1 = x_stride << 2;
|
||||
int loop_c = x_out_max - channel;
|
||||
|
||||
int channel;
|
||||
assert(!wrk->x_expand);
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
__asm__ volatile (
|
||||
"li %[temp1], 0x8000 \n\t"
|
||||
"li %[temp2], 0x10000 \n\t"
|
||||
"li %[sum], 0 \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"li %[base], 0 \n\t"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[base] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"mul %[temp3], %[sum], %[x_sub] \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"sll %[accum1], %[frac], 2 \n\t"
|
||||
"mult %[temp1], %[temp2] \n\t"
|
||||
"madd %[accum1], %[fx_scale] \n\t"
|
||||
"mfhi %[sum] \n\t"
|
||||
"subu %[temp3], %[temp3], %[frac] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
: [accum] "=&r" (accum), [src1] "+r" (src1), [temp3] "=&r" (temp3),
|
||||
[sum] "=&r" (sum), [base] "=&r" (base), [frac] "=&r" (frac),
|
||||
[frow] "+r" (frow), [accum1] "=&r" (accum1),
|
||||
[temp2] "=&r" (temp2), [temp1] "=&r" (temp1)
|
||||
: [x_stride] "r" (x_stride), [fx_scale] "r" (fx_scale),
|
||||
[x_sub] "r" (x_sub), [x_add] "r" (x_add),
|
||||
[loop_c] "r" (loop_c), [x_stride1] "r" (x_stride1)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
const uint8_t* src1 = src + channel;
|
||||
int* frow = wrk->frow + channel;
|
||||
int temp1, temp2, temp3;
|
||||
int base, frac, sum;
|
||||
int accum, accum1;
|
||||
int loop_c = x_out_max;
|
||||
|
||||
__asm__ volatile (
|
||||
"li %[temp1], 0x8000 \n\t"
|
||||
"li %[temp2], 0x10000 \n\t"
|
||||
"li %[sum], 0 \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"li %[base], 0 \n\t"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[base] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"mul %[temp3], %[sum], %[x_sub] \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"sll %[accum1], %[frac], 2 \n\t"
|
||||
"mult %[temp1], %[temp2] \n\t"
|
||||
"madd %[accum1], %[fx_scale] \n\t"
|
||||
"mfhi %[sum] \n\t"
|
||||
"subu %[temp3], %[temp3], %[frac] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
: [accum] "=&r" (accum), [src1] "+r" (src1), [temp3] "=&r" (temp3),
|
||||
[sum] "=&r" (sum), [base] "=&r" (base), [frac] "=&r" (frac),
|
||||
[frow] "+r" (frow), [accum1] "=&r" (accum1),
|
||||
[temp2] "=&r" (temp2), [temp1] "=&r" (temp1)
|
||||
: [x_stride] "r" (x_stride), [fx_scale] "r" (fx_scale),
|
||||
[x_sub] "r" (x_sub), [x_add] "r" (x_add),
|
||||
[loop_c] "r" (loop_c), [x_stride1] "r" (x_stride1)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ImportRowExpand(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
const int x_add = wrk->x_add;
|
||||
const int x_sub = wrk->x_sub;
|
||||
const int src_width = wrk->src_width;
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp1, temp2, temp3, temp4;
|
||||
int frac;
|
||||
int accum;
|
||||
const int x_stride1 = x_stride << 2;
|
||||
int x_out = channel;
|
||||
|
||||
int channel;
|
||||
assert(wrk->x_expand);
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
__asm__ volatile (
|
||||
"addiu %[temp3], %[src_width], -1 \n\t"
|
||||
"lbu %[temp2], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"bgtz %[temp3], 0f \n\t"
|
||||
"addiu %[temp1], %[temp2], 0 \n\t"
|
||||
"b 3f \n\t"
|
||||
"0: \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"3: \n\t"
|
||||
"addiu %[accum], %[x_add], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[temp4], %[temp1], %[x_add] \n\t"
|
||||
"addu %[temp3], %[temp4], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"addu %[x_out], %[x_out], %[x_stride] \n\t"
|
||||
"subu %[temp3], %[x_out], %[x_out_max] \n\t"
|
||||
"bgez %[temp3], 2f \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"bgez %[accum], 4f \n\t"
|
||||
"addiu %[temp2], %[temp1], 0 \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"4: \n\t"
|
||||
"b 1b \n\t"
|
||||
"2: \n\t"
|
||||
: [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
|
||||
[temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [temp4] "=&r" (temp4),
|
||||
[x_out] "+r" (x_out), [frac] "=&r" (frac), [frow] "+r" (frow)
|
||||
: [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
|
||||
[x_stride1] "r" (x_stride1), [src_width] "r" (src_width),
|
||||
[x_out_max] "r" (x_out_max)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
const uint8_t* src1 = src + channel;
|
||||
int* frow = wrk->frow + channel;
|
||||
int temp1, temp2, temp3, temp4;
|
||||
int frac;
|
||||
int accum;
|
||||
int x_out = 0;
|
||||
|
||||
__asm__ volatile (
|
||||
"addiu %[temp3], %[src_width], -1 \n\t"
|
||||
"lbu %[temp2], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"bgtz %[temp3], 0f \n\t"
|
||||
"addiu %[temp1], %[temp2], 0 \n\t"
|
||||
"b 3f \n\t"
|
||||
"0: \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"3: \n\t"
|
||||
"addiu %[accum], %[x_add], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[temp4], %[temp1], %[x_add] \n\t"
|
||||
"addu %[temp3], %[temp4], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"addu %[x_out], %[x_out], %[x_stride] \n\t"
|
||||
"subu %[temp3], %[x_out], %[x_out_max] \n\t"
|
||||
"bgez %[temp3], 2f \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"bgez %[accum], 4f \n\t"
|
||||
"addiu %[temp2], %[temp1], 0 \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"4: \n\t"
|
||||
"b 1b \n\t"
|
||||
"2: \n\t"
|
||||
: [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
|
||||
[temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [temp4] "=&r" (temp4),
|
||||
[x_out] "+r" (x_out), [frac] "=&r" (frac), [frow] "+r" (frow)
|
||||
: [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
|
||||
[x_stride1] "r" (x_stride1), [src_width] "r" (src_width),
|
||||
[x_out_max] "r" (x_out_max)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ExportRowShrink(WebPRescaler* const wrk) {
|
||||
|
||||
@ -18,112 +18,118 @@
|
||||
#include <assert.h>
|
||||
#include "../utils/rescaler.h"
|
||||
|
||||
static void ImportRowShrink(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
const int fx_scale = wrk->fx_scale;
|
||||
const int x_add = wrk->x_add;
|
||||
const int x_sub = wrk->x_sub;
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp3;
|
||||
int base, frac, sum;
|
||||
int accum, accum1;
|
||||
const int x_stride1 = x_stride << 2;
|
||||
int loop_c = x_out_max - channel;
|
||||
|
||||
int channel;
|
||||
assert(!wrk->x_expand);
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
__asm__ volatile (
|
||||
"li %[sum], 0 \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"li %[base], 0 \n\t"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[base] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"mul %[temp3], %[sum], %[x_sub] \n\t"
|
||||
"sll %[accum1], %[frac], 1 \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"mulq_rs.w %[sum], %[accum1], %[fx_scale] \n\t"
|
||||
"subu %[temp3], %[temp3], %[frac] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
: [accum]"=&r"(accum), [src1]"+&r"(src1), [temp3]"=&r"(temp3),
|
||||
[sum]"=&r"(sum), [base]"=&r"(base), [frac]"=&r"(frac),
|
||||
[frow]"+&r"(frow), [accum1]"=&r"(accum1),
|
||||
[loop_c]"+&r"(loop_c)
|
||||
: [x_stride]"r"(x_stride), [fx_scale]"r"(fx_scale), [x_sub]"r"(x_sub),
|
||||
[x_add] "r" (x_add), [x_stride1] "r" (x_stride1)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp3;
|
||||
int base, frac, sum;
|
||||
int accum, accum1;
|
||||
int loop_c = x_out_max;
|
||||
|
||||
__asm__ volatile (
|
||||
"li %[sum], 0 \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"li %[base], 0 \n\t"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[base] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"mul %[temp3], %[sum], %[x_sub] \n\t"
|
||||
"sll %[accum1], %[frac], 1 \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"mulq_rs.w %[sum], %[accum1], %[fx_scale] \n\t"
|
||||
"subu %[temp3], %[temp3], %[frac] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
: [accum]"=&r"(accum), [src1]"+&r"(src1), [temp3]"=&r"(temp3),
|
||||
[sum]"=&r"(sum), [base]"=&r"(base), [frac]"=&r"(frac),
|
||||
[frow]"+&r"(frow), [accum1]"=&r"(accum1),
|
||||
[loop_c]"+&r"(loop_c)
|
||||
: [x_stride]"r"(x_stride), [fx_scale]"r"(fx_scale), [x_sub]"r"(x_sub),
|
||||
[x_add] "r" (x_add), [x_stride1] "r" (x_stride1)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ImportRowExpand(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
|
||||
const int x_stride = wrk->num_channels;
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
const int x_add = wrk->x_add;
|
||||
const int x_sub = wrk->x_sub;
|
||||
const int src_width = wrk->src_width;
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp1, temp2, temp3, temp4;
|
||||
int frac;
|
||||
int accum;
|
||||
const int x_stride1 = x_stride << 2;
|
||||
int x_out = channel;
|
||||
|
||||
int channel;
|
||||
assert(wrk->x_expand);
|
||||
assert(!WebPRescalerInputDone(wrk));
|
||||
__asm__ volatile (
|
||||
"addiu %[temp3], %[src_width], -1 \n\t"
|
||||
"lbu %[temp2], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"bgtz %[temp3], 0f \n\t"
|
||||
"addiu %[temp1], %[temp2], 0 \n\t"
|
||||
"b 3f \n\t"
|
||||
"0: \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"3: \n\t"
|
||||
"addiu %[accum], %[x_add], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[temp4], %[temp1], %[x_add] \n\t"
|
||||
"addu %[temp3], %[temp4], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"addu %[x_out], %[x_out], %[x_stride] \n\t"
|
||||
"subu %[temp3], %[x_out], %[x_out_max] \n\t"
|
||||
"bgez %[temp3], 2f \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"bgez %[accum], 4f \n\t"
|
||||
"addiu %[temp2], %[temp1], 0 \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"4: \n\t"
|
||||
"b 1b \n\t"
|
||||
"2: \n\t"
|
||||
: [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
|
||||
[temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [temp4] "=&r" (temp4),
|
||||
[x_out] "+r" (x_out), [frac] "=&r" (frac), [frow] "+r" (frow)
|
||||
: [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
|
||||
[x_stride1] "r" (x_stride1), [src_width] "r" (src_width),
|
||||
[x_out_max] "r" (x_out_max)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
|
||||
for (channel = 0; channel < x_stride; ++channel) {
|
||||
int* frow = wrk->frow + channel;
|
||||
const uint8_t* src1 = src + channel;
|
||||
int temp1, temp2, temp3, temp4;
|
||||
int frac;
|
||||
int accum;
|
||||
int x_out = channel;
|
||||
|
||||
__asm__ volatile (
|
||||
"addiu %[temp3], %[src_width], -1 \n\t"
|
||||
"lbu %[temp2], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"bgtz %[temp3], 0f \n\t"
|
||||
"addiu %[temp1], %[temp2], 0 \n\t"
|
||||
"b 3f \n\t"
|
||||
"0: \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"3: \n\t"
|
||||
"addiu %[accum], %[x_add], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[temp4], %[temp1], %[x_add] \n\t"
|
||||
"addu %[temp3], %[temp4], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[frow], %[x_stride1] \n\t"
|
||||
"addu %[x_out], %[x_out], %[x_stride] \n\t"
|
||||
"subu %[temp3], %[x_out], %[x_out_max] \n\t"
|
||||
"bgez %[temp3], 2f \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"bgez %[accum], 4f \n\t"
|
||||
"addiu %[temp2], %[temp1], 0 \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"4: \n\t"
|
||||
"b 1b \n\t"
|
||||
"2: \n\t"
|
||||
: [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
|
||||
[temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [temp4] "=&r" (temp4),
|
||||
[x_out] "+r" (x_out), [frac] "=&r" (frac), [frow] "+r" (frow)
|
||||
: [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
|
||||
[x_stride1] "r" (x_stride1), [src_width] "r" (src_width),
|
||||
[x_out_max] "r" (x_out_max)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ExportRowShrink(WebPRescaler* const wrk) {
|
||||
|
||||
@ -102,16 +102,14 @@ int WebPRescalerImport(WebPRescaler* const wrk, int num_lines,
|
||||
const uint8_t* src, int src_stride) {
|
||||
int total_imported = 0;
|
||||
while (total_imported < num_lines && !WebPRescalerHasPendingOutput(wrk)) {
|
||||
int x, channel;
|
||||
if (wrk->y_expand) {
|
||||
rescaler_t* const tmp = wrk->irow;
|
||||
wrk->irow = wrk->frow;
|
||||
wrk->frow = tmp;
|
||||
}
|
||||
for (channel = 0; channel < wrk->num_channels; ++channel) {
|
||||
WebPRescalerImportRow(wrk, src, channel);
|
||||
}
|
||||
WebPRescalerImportRow(wrk, src);
|
||||
if (!wrk->y_expand) { // Accumulate the contribution of the new row.
|
||||
int x;
|
||||
for (x = 0; x < wrk->num_channels * wrk->dst_width; ++x) {
|
||||
wrk->irow[x] += wrk->frow[x];
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user