mirror of
https://github.com/webmproject/libwebp.git
synced 2025-07-13 06:24:27 +02:00
Merge "refactor the PSNR / SSIM calculation code"
This commit is contained in:
@ -250,7 +250,7 @@ extern VP8GetResidualCostFunc VP8GetResidualCost;
|
||||
void VP8EncDspCostInit(void);
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
// SSIM utils
|
||||
// SSIM / PSNR utils
|
||||
|
||||
// struct for accumulating statistical moments
|
||||
typedef struct {
|
||||
@ -275,6 +275,10 @@ typedef void (*VP8SSIMAccumulateFunc)(const uint8_t* src1, int stride1,
|
||||
extern VP8SSIMAccumulateFunc VP8SSIMAccumulate; // unclipped / unchecked
|
||||
extern VP8SSIMAccumulateClippedFunc VP8SSIMAccumulateClipped; // with clipping
|
||||
|
||||
typedef uint32_t (*VP8AccumulateSSEFunc)(const uint8_t* src1,
|
||||
const uint8_t* src2, int len);
|
||||
extern VP8AccumulateSSEFunc VP8AccumulateSSE;
|
||||
|
||||
// must be called before using any of the above directly
|
||||
void VP8SSIMDspInit(void);
|
||||
|
||||
|
@ -691,6 +691,7 @@ static void Copy16x8(const uint8_t* src, uint8_t* dst) {
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
// SSIM / PSNR
|
||||
|
||||
static void SSIMAccumulateClipped(const uint8_t* src1, int stride1,
|
||||
const uint8_t* src2, int stride2,
|
||||
@ -737,8 +738,23 @@ static void SSIMAccumulate(const uint8_t* src1, int stride1,
|
||||
}
|
||||
}
|
||||
|
||||
static uint32_t AccumulateSSE(const uint8_t* src1,
|
||||
const uint8_t* src2, int len) {
|
||||
int i;
|
||||
uint32_t sse2 = 0;
|
||||
assert(len <= 65535); // to ensure that accumulation fits within uint32_t
|
||||
for (i = 0; i < len; ++i) {
|
||||
const int32_t diff = src1[i] - src2[i];
|
||||
sse2 += diff * diff;
|
||||
}
|
||||
return sse2;
|
||||
}
|
||||
|
||||
VP8SSIMAccumulateFunc VP8SSIMAccumulate;
|
||||
VP8SSIMAccumulateClippedFunc VP8SSIMAccumulateClipped;
|
||||
VP8AccumulateSSEFunc VP8AccumulateSSE;
|
||||
|
||||
extern void VP8SSIMDspInitSSE2(void);
|
||||
|
||||
static volatile VP8CPUInfo ssim_last_cpuinfo_used =
|
||||
(VP8CPUInfo)&ssim_last_cpuinfo_used;
|
||||
@ -749,6 +765,15 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInit(void) {
|
||||
VP8SSIMAccumulate = SSIMAccumulate;
|
||||
VP8SSIMAccumulateClipped = SSIMAccumulateClipped;
|
||||
|
||||
VP8AccumulateSSE = AccumulateSSE;
|
||||
if (VP8GetCPUInfo != NULL) {
|
||||
#if defined(WEBP_USE_SSE2)
|
||||
if (VP8GetCPUInfo(kSSE2)) {
|
||||
VP8SSIMDspInitSSE2();
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
ssim_last_cpuinfo_used = VP8GetCPUInfo;
|
||||
}
|
||||
|
||||
|
@ -1365,8 +1365,56 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitSSE2(void) {
|
||||
VP8Mean16x4 = Mean16x4;
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
// SSIM / PSNR entry point (TODO(skal): move to its own file later)
|
||||
|
||||
static uint32_t AccumulateSSE_SSE2(const uint8_t* src1,
|
||||
const uint8_t* src2, int len) {
|
||||
int i = 0;
|
||||
uint32_t sse2 = 0;
|
||||
if (len >= 16) {
|
||||
const int limit = len - 32;
|
||||
int32_t tmp[4];
|
||||
__m128i sum1;
|
||||
__m128i sum = _mm_setzero_si128();
|
||||
__m128i a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
|
||||
__m128i b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
|
||||
i += 16;
|
||||
while (i <= limit) {
|
||||
const __m128i a1 = _mm_loadu_si128((const __m128i*)&src1[i]);
|
||||
const __m128i b1 = _mm_loadu_si128((const __m128i*)&src2[i]);
|
||||
__m128i sum2;
|
||||
i += 16;
|
||||
SubtractAndAccumulate(a0, b0, &sum1);
|
||||
sum = _mm_add_epi32(sum, sum1);
|
||||
a0 = _mm_loadu_si128((const __m128i*)&src1[i]);
|
||||
b0 = _mm_loadu_si128((const __m128i*)&src2[i]);
|
||||
i += 16;
|
||||
SubtractAndAccumulate(a1, b1, &sum2);
|
||||
sum = _mm_add_epi32(sum, sum2);
|
||||
}
|
||||
SubtractAndAccumulate(a0, b0, &sum1);
|
||||
sum = _mm_add_epi32(sum, sum1);
|
||||
_mm_storeu_si128((__m128i*)tmp, sum);
|
||||
sse2 += (tmp[3] + tmp[2] + tmp[1] + tmp[0]);
|
||||
}
|
||||
|
||||
for (; i < len; ++i) {
|
||||
const int32_t diff = src1[i] - src2[i];
|
||||
sse2 += diff * diff;
|
||||
}
|
||||
return sse2;
|
||||
}
|
||||
|
||||
extern void VP8SSIMDspInitSSE2(void);
|
||||
|
||||
WEBP_TSAN_IGNORE_FUNCTION void VP8SSIMDspInitSSE2(void) {
|
||||
VP8AccumulateSSE = AccumulateSSE_SSE2;
|
||||
}
|
||||
|
||||
#else // !WEBP_USE_SSE2
|
||||
|
||||
WEBP_DSP_INIT_STUB(VP8EncDspInitSSE2)
|
||||
WEBP_DSP_INIT_STUB(VP8SSIMDspInitSSE2)
|
||||
|
||||
#endif // WEBP_USE_SSE2
|
||||
|
Reference in New Issue
Block a user