mirror of
https://github.com/webmproject/libwebp.git
synced 2025-08-30 07:42:27 +02:00
fix rescaling bug (uninitialized read, see bug #254).
the x_add/x_sub increments were wrong for u/v in the upscaling case. They shouldn't be left to the caller's discretion, but set up by WebPRescalerInit to their exact necessary values. -> Cleaned-up WebPRescalerInit() param list. -> added safety asserts -> removed the mips32/mips_r2 variant of "ImportRow" which were buggy prior Change-Id: I347c75804d835811e7025de92a0758d7929dfc09
This commit is contained in:
Pascal Massimino
@@ -9,6 +9,8 @@
|
||||
//
|
||||
// Rescaling functions
|
||||
|
||||
#include <assert.h>
|
||||
|
||||
#include "./dsp.h"
|
||||
#include "../utils/rescaler.h"
|
||||
|
||||
@@ -24,14 +26,15 @@ static void RescalerImportRowC(WebPRescaler* const wrk,
|
||||
const int x_out_max = wrk->dst_width * wrk->num_channels;
|
||||
int x_in = channel;
|
||||
int x_out;
|
||||
int accum = 0;
|
||||
if (!wrk->x_expand) {
|
||||
int sum = 0;
|
||||
uint32_t sum = 0;
|
||||
int accum = 0;
|
||||
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;
|
||||
@@ -43,18 +46,27 @@ static void RescalerImportRowC(WebPRescaler* const wrk,
|
||||
sum = (int)MULT_FIX(frac, wrk->fx_scale);
|
||||
}
|
||||
}
|
||||
assert(accum == 0);
|
||||
} else { // simple bilinear interpolation
|
||||
int left = src[channel], right = src[channel];
|
||||
for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
|
||||
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;
|
||||
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;
|
||||
}
|
||||
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
|
||||
accum -= wrk->x_sub;
|
||||
}
|
||||
assert(wrk->x_sub == 0 /* <- special case for src_width=1 */ || accum == 0);
|
||||
}
|
||||
// Accumulate the contribution of the new row.
|
||||
for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
|
||||
|
||||
@@ -17,103 +17,6 @@
|
||||
|
||||
#include "../utils/rescaler.h"
|
||||
|
||||
static void ImportRow(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
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 - wrk->x_sub;
|
||||
const int x_sub = wrk->x_sub;
|
||||
int* frow = wrk->frow + channel;
|
||||
int* irow = wrk->irow + 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;
|
||||
|
||||
if (!wrk->x_expand) {
|
||||
__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"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[temp3], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[temp3] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"addu %[temp3], %[sum], %[base] \n\t"
|
||||
"mul %[temp3], %[temp3], %[x_sub] \n\t"
|
||||
"lw %[base], 0(%[irow]) \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"
|
||||
"add %[base], %[base], %[temp3] \n\t"
|
||||
"sw %[base], 0(%[irow]) \n\t"
|
||||
"addu %[irow], %[irow], %[x_stride1] \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), [irow] "+r" (irow), [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"
|
||||
);
|
||||
} else {
|
||||
__asm__ volatile (
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"move %[temp2], %[temp1] \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"bgez %[accum], 2f \n\t"
|
||||
"move %[temp2], %[temp1] \n\t"
|
||||
"addu %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[x_add] \n\t"
|
||||
"2: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[base], %[temp1], %[x_add] \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"lw %[frac], 0(%[irow]) \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"addu %[temp3], %[base], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[x_stride1] \n\t"
|
||||
"addu %[frac], %[temp3] \n\t"
|
||||
"sw %[frac], 0(%[irow]) \n\t"
|
||||
"addu %[irow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
|
||||
: [src1] "+r" (src1), [accum] "=&r" (accum), [temp1] "=&r" (temp1),
|
||||
[temp2] "=&r" (temp2), [temp3] "=&r" (temp3), [base] "=&r" (base),
|
||||
[frac] "=&r" (frac), [frow] "+r" (frow), [irow] "+r" (irow)
|
||||
: [x_stride] "r" (x_stride), [x_add] "r" (x_add), [x_sub] "r" (x_sub),
|
||||
[x_stride1] "r" (x_stride1), [loop_c] "r" (loop_c)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ExportRow(WebPRescaler* const wrk, int x_out) {
|
||||
if (wrk->y_accum <= 0) {
|
||||
uint8_t* const dst = wrk->dst;
|
||||
@@ -183,7 +86,6 @@ static void ExportRow(WebPRescaler* const wrk, int x_out) {
|
||||
extern void WebPRescalerDspInitMIPS32(void);
|
||||
|
||||
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPS32(void) {
|
||||
WebPRescalerImportRow = ImportRow;
|
||||
WebPRescalerExportRow = ExportRow;
|
||||
}
|
||||
|
||||
|
||||
@@ -17,99 +17,6 @@
|
||||
|
||||
#include "../utils/rescaler.h"
|
||||
|
||||
static void ImportRow(WebPRescaler* const wrk,
|
||||
const uint8_t* const src, int channel) {
|
||||
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 - wrk->x_sub;
|
||||
const int x_sub = wrk->x_sub;
|
||||
int* frow = wrk->frow + channel;
|
||||
int* irow = wrk->irow + 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;
|
||||
|
||||
if (!wrk->x_expand) {
|
||||
__asm__ volatile (
|
||||
"li %[sum], 0 \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"addu %[accum], %[accum], %[x_add] \n\t"
|
||||
"blez %[accum], 3f \n\t"
|
||||
"2: \n\t"
|
||||
"lbu %[temp3], 0(%[src1]) \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"addu %[sum], %[sum], %[temp3] \n\t"
|
||||
"bgtz %[accum], 2b \n\t"
|
||||
"3: \n\t"
|
||||
"lbu %[base], 0(%[src1]) \n\t"
|
||||
"addu %[src1], %[src1], %[x_stride] \n\t"
|
||||
"negu %[accum1], %[accum] \n\t"
|
||||
"mul %[frac], %[base], %[accum1] \n\t"
|
||||
"addu %[temp3], %[sum], %[base] \n\t"
|
||||
"mul %[temp3], %[temp3], %[x_sub] \n\t"
|
||||
"lw %[base], 0(%[irow]) \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"
|
||||
"add %[base], %[base], %[temp3] \n\t"
|
||||
"sw %[base], 0(%[irow]) \n\t"
|
||||
"addu %[irow], %[irow], %[x_stride1] \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), [irow]"+&r"(irow), [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"
|
||||
);
|
||||
} else {
|
||||
__asm__ volatile (
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"move %[temp2], %[temp1] \n\t"
|
||||
"li %[accum], 0 \n\t"
|
||||
"1: \n\t"
|
||||
"bgez %[accum], 2f \n\t"
|
||||
"move %[temp2], %[temp1] \n\t"
|
||||
"addu %[src1], %[x_stride] \n\t"
|
||||
"lbu %[temp1], 0(%[src1]) \n\t"
|
||||
"addu %[accum], %[x_add] \n\t"
|
||||
"2: \n\t"
|
||||
"subu %[temp3], %[temp2], %[temp1] \n\t"
|
||||
"mul %[temp3], %[temp3], %[accum] \n\t"
|
||||
"mul %[base], %[temp1], %[x_add] \n\t"
|
||||
"subu %[accum], %[accum], %[x_sub] \n\t"
|
||||
"lw %[frac], 0(%[irow]) \n\t"
|
||||
"subu %[loop_c], %[loop_c], %[x_stride] \n\t"
|
||||
"addu %[temp3], %[base], %[temp3] \n\t"
|
||||
"sw %[temp3], 0(%[frow]) \n\t"
|
||||
"addu %[frow], %[x_stride1] \n\t"
|
||||
"addu %[frac], %[temp3] \n\t"
|
||||
"sw %[frac], 0(%[irow]) \n\t"
|
||||
"addu %[irow], %[x_stride1] \n\t"
|
||||
"bgtz %[loop_c], 1b \n\t"
|
||||
|
||||
: [src1]"+&r"(src1), [accum]"=&r"(accum), [temp1]"=&r"(temp1),
|
||||
[temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [base]"=&r"(base),
|
||||
[frac]"=&r"(frac), [frow]"+&r"(frow), [irow]"+&r"(irow),
|
||||
[loop_c]"+&r"(loop_c)
|
||||
: [x_stride]"r"(x_stride), [x_add]"r"(x_add), [x_sub]"r"(x_sub),
|
||||
[x_stride1]"r"(x_stride1)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
static void ExportRow(WebPRescaler* const wrk, int x_out) {
|
||||
if (wrk->y_accum <= 0) {
|
||||
// if wrk->fxy_scale can fit into 32 bits use optimized code,
|
||||
@@ -201,7 +108,6 @@ static void ExportRow(WebPRescaler* const wrk, int x_out) {
|
||||
extern void WebPRescalerDspInitMIPSdspR2(void);
|
||||
|
||||
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPSdspR2(void) {
|
||||
WebPRescalerImportRow = ImportRow;
|
||||
WebPRescalerExportRow = ExportRow;
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user