rescaler: simplify ImportRow logic

incorporates the loop over 'channel' and removes one parameter

Change-Id: I4e3b33c111ca825fe96461583420413b17326409
This commit is contained in:
Pascal Massimino 2015-09-19 09:59:32 -07:00
parent 5ff0079ece
commit 9ba1894b9b
5 changed files with 250 additions and 239 deletions

View File

@ -338,8 +338,7 @@ struct WebPRescaler;
// 'channel' denotes the channel number to be imported. 'Expand' corresponds to // 'channel' denotes the channel number to be imported. 'Expand' corresponds to
// the wrk->x_expand case. Otherwise, 'Shrink' is to be used. // the wrk->x_expand case. Otherwise, 'Shrink' is to be used.
typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk, typedef void (*WebPRescalerImportRowFunc)(struct WebPRescaler* const wrk,
const uint8_t* const src, const uint8_t* src);
int channel);
extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand; extern WebPRescalerImportRowFunc WebPRescalerImportRowExpand;
extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink; extern WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
@ -353,15 +352,15 @@ extern WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
// Plain-C implementation, as fall-back. // Plain-C implementation, as fall-back.
extern void WebPRescalerImportRowExpandC(struct WebPRescaler* const wrk, 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, 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 WebPRescalerExportRowExpandC(struct WebPRescaler* const wrk);
extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk); extern void WebPRescalerExportRowShrinkC(struct WebPRescaler* const wrk);
// Main entry calls: // Main entry calls:
extern void WebPRescalerImportRow(struct WebPRescaler* const wrk, 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. // Export one row (starting at x_out position) from rescaler.
extern void WebPRescalerExportRow(struct WebPRescaler* const wrk); extern void WebPRescalerExportRow(struct WebPRescaler* const wrk);

View File

@ -23,21 +23,20 @@
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
// Row import // Row import
void WebPRescalerImportRowExpandC(WebPRescaler* const wrk, void WebPRescalerImportRowExpandC(WebPRescaler* wrk, const uint8_t* src) {
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels; const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
int channel;
assert(!WebPRescalerInputDone(wrk));
assert(wrk->x_expand);
for (channel = 0; channel < x_stride; ++channel) {
int x_in = channel; int x_in = channel;
int x_out; int x_out = channel;
// simple bilinear interpolation // simple bilinear interpolation
int accum = wrk->x_add; int accum = wrk->x_add;
int left = src[x_in]; int left = src[x_in];
int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left; int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
x_in += x_stride; x_in += x_stride;
x_out = channel;
assert(!WebPRescalerInputDone(wrk));
assert(wrk->x_expand);
while (1) { while (1) {
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum; wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
x_out += x_stride; x_out += x_stride;
@ -53,19 +52,21 @@ void WebPRescalerImportRowExpandC(WebPRescaler* const wrk,
} }
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, 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_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
int x_in = channel; int channel;
int x_out;
uint32_t sum = 0;
int accum = 0;
assert(!WebPRescalerInputDone(wrk)); assert(!WebPRescalerInputDone(wrk));
assert(!wrk->x_expand); assert(!wrk->x_expand);
for (x_out = channel; x_out < x_out_max; x_out += x_stride) { 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; uint32_t base = 0;
accum += wrk->x_add; accum += wrk->x_add;
while (accum > 0) { while (accum > 0) {
@ -81,9 +82,11 @@ void WebPRescalerImportRowShrinkC(WebPRescaler* const wrk,
// fresh fractional start for next pixel // fresh fractional start for next pixel
sum = (int)MULT_FIX(frac, wrk->fx_scale); sum = (int)MULT_FIX(frac, wrk->fx_scale);
} }
x_out += x_stride;
} }
assert(accum == 0); assert(accum == 0);
} }
}
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
// Row export // Row export
@ -145,13 +148,12 @@ void WebPRescalerExportRowShrinkC(WebPRescaler* const wrk) {
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
// Main entry calls // Main entry calls
void WebPRescalerImportRow(WebPRescaler* const wrk, void WebPRescalerImportRow(WebPRescaler* const wrk, const uint8_t* const src) {
const uint8_t* const src, int channel) {
assert(!WebPRescalerInputDone(wrk)); assert(!WebPRescalerInputDone(wrk));
if (!wrk->x_expand) { if (!wrk->x_expand) {
WebPRescalerImportRowShrink(wrk, src, channel); WebPRescalerImportRowShrink(wrk, src);
} else { } else {
WebPRescalerImportRowExpand(wrk, src, channel); WebPRescalerImportRowExpand(wrk, src);
} }
} }

View File

@ -18,23 +18,25 @@
#include <assert.h> #include <assert.h>
#include "../utils/rescaler.h" #include "../utils/rescaler.h"
static void ImportRowShrink(WebPRescaler* const wrk, static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels; const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
const int fx_scale = wrk->fx_scale; const int fx_scale = wrk->fx_scale;
const int x_add = wrk->x_add; const int x_add = wrk->x_add;
const int x_sub = wrk->x_sub; const int x_sub = wrk->x_sub;
int* frow = wrk->frow + channel; const int x_stride1 = x_stride << 2;
int channel;
assert(!wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
const uint8_t* src1 = src + channel; const uint8_t* src1 = src + channel;
int* frow = wrk->frow + channel;
int temp1, temp2, temp3; int temp1, temp2, temp3;
int base, frac, sum; int base, frac, sum;
int accum, accum1; int accum, accum1;
const int x_stride1 = x_stride << 2; int loop_c = x_out_max;
int loop_c = x_out_max - channel;
assert(!wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
__asm__ volatile ( __asm__ volatile (
"li %[temp1], 0x8000 \n\t" "li %[temp1], 0x8000 \n\t"
"li %[temp2], 0x10000 \n\t" "li %[temp2], 0x10000 \n\t"
@ -73,24 +75,27 @@ static void ImportRowShrink(WebPRescaler* const wrk,
: "memory", "hi", "lo" : "memory", "hi", "lo"
); );
} }
}
static void ImportRowExpand(WebPRescaler* const wrk, static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels; const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
const int x_add = wrk->x_add; const int x_add = wrk->x_add;
const int x_sub = wrk->x_sub; const int x_sub = wrk->x_sub;
const int src_width = wrk->src_width; const int src_width = wrk->src_width;
int* frow = wrk->frow + channel; const int x_stride1 = x_stride << 2;
int channel;
assert(wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
const uint8_t* src1 = src + channel; const uint8_t* src1 = src + channel;
int* frow = wrk->frow + channel;
int temp1, temp2, temp3, temp4; int temp1, temp2, temp3, temp4;
int frac; int frac;
int accum; int accum;
const int x_stride1 = x_stride << 2; int x_out = 0;
int x_out = channel;
assert(wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
__asm__ volatile ( __asm__ volatile (
"addiu %[temp3], %[src_width], -1 \n\t" "addiu %[temp3], %[src_width], -1 \n\t"
"lbu %[temp2], 0(%[src1]) \n\t" "lbu %[temp2], 0(%[src1]) \n\t"
@ -130,6 +135,7 @@ static void ImportRowExpand(WebPRescaler* const wrk,
: "memory", "hi", "lo" : "memory", "hi", "lo"
); );
} }
}
static void ExportRowShrink(WebPRescaler* const wrk) { static void ExportRowShrink(WebPRescaler* const wrk) {
assert(!WebPRescalerOutputDone(wrk)); assert(!WebPRescalerOutputDone(wrk));

View File

@ -18,23 +18,25 @@
#include <assert.h> #include <assert.h>
#include "../utils/rescaler.h" #include "../utils/rescaler.h"
static void ImportRowShrink(WebPRescaler* const wrk, static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels; const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
const int fx_scale = wrk->fx_scale; const int fx_scale = wrk->fx_scale;
const int x_add = wrk->x_add; const int x_add = wrk->x_add;
const int x_sub = wrk->x_sub; const int x_sub = wrk->x_sub;
const int x_stride1 = x_stride << 2;
int channel;
assert(!wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
int* frow = wrk->frow + channel; int* frow = wrk->frow + channel;
const uint8_t* src1 = src + channel; const uint8_t* src1 = src + channel;
int temp3; int temp3;
int base, frac, sum; int base, frac, sum;
int accum, accum1; int accum, accum1;
const int x_stride1 = x_stride << 2; int loop_c = x_out_max;
int loop_c = x_out_max - channel;
assert(!wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
__asm__ volatile ( __asm__ volatile (
"li %[sum], 0 \n\t" "li %[sum], 0 \n\t"
"li %[accum], 0 \n\t" "li %[accum], 0 \n\t"
@ -68,24 +70,27 @@ static void ImportRowShrink(WebPRescaler* const wrk,
: "memory", "hi", "lo" : "memory", "hi", "lo"
); );
} }
}
static void ImportRowExpand(WebPRescaler* const wrk, static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels; const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels; const int x_out_max = wrk->dst_width * wrk->num_channels;
const int x_add = wrk->x_add; const int x_add = wrk->x_add;
const int x_sub = wrk->x_sub; const int x_sub = wrk->x_sub;
const int src_width = wrk->src_width; const int src_width = wrk->src_width;
const int x_stride1 = x_stride << 2;
int channel;
assert(wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
int* frow = wrk->frow + channel; int* frow = wrk->frow + channel;
const uint8_t* src1 = src + channel; const uint8_t* src1 = src + channel;
int temp1, temp2, temp3, temp4; int temp1, temp2, temp3, temp4;
int frac; int frac;
int accum; int accum;
const int x_stride1 = x_stride << 2;
int x_out = channel; int x_out = channel;
assert(wrk->x_expand);
assert(!WebPRescalerInputDone(wrk));
__asm__ volatile ( __asm__ volatile (
"addiu %[temp3], %[src_width], -1 \n\t" "addiu %[temp3], %[src_width], -1 \n\t"
"lbu %[temp2], 0(%[src1]) \n\t" "lbu %[temp2], 0(%[src1]) \n\t"
@ -125,6 +130,7 @@ static void ImportRowExpand(WebPRescaler* const wrk,
: "memory", "hi", "lo" : "memory", "hi", "lo"
); );
} }
}
static void ExportRowShrink(WebPRescaler* const wrk) { static void ExportRowShrink(WebPRescaler* const wrk) {
assert(!WebPRescalerOutputDone(wrk)); assert(!WebPRescalerOutputDone(wrk));

View File

@ -102,16 +102,14 @@ int WebPRescalerImport(WebPRescaler* const wrk, int num_lines,
const uint8_t* src, int src_stride) { const uint8_t* src, int src_stride) {
int total_imported = 0; int total_imported = 0;
while (total_imported < num_lines && !WebPRescalerHasPendingOutput(wrk)) { while (total_imported < num_lines && !WebPRescalerHasPendingOutput(wrk)) {
int x, channel;
if (wrk->y_expand) { if (wrk->y_expand) {
rescaler_t* const tmp = wrk->irow; rescaler_t* const tmp = wrk->irow;
wrk->irow = wrk->frow; wrk->irow = wrk->frow;
wrk->frow = tmp; wrk->frow = tmp;
} }
for (channel = 0; channel < wrk->num_channels; ++channel) { WebPRescalerImportRow(wrk, src);
WebPRescalerImportRow(wrk, src, channel);
}
if (!wrk->y_expand) { // Accumulate the contribution of the new row. if (!wrk->y_expand) { // Accumulate the contribution of the new row.
int x;
for (x = 0; x < wrk->num_channels * wrk->dst_width; ++x) { for (x = 0; x < wrk->num_channels * wrk->dst_width; ++x) {
wrk->irow[x] += wrk->frow[x]; wrk->irow[x] += wrk->frow[x];
} }