InplaceTransformFinal get_invhtransformfinal_sse4_2(int wavelet_index, int active_bits, int sample_size) {
if (sample_size == 4) {
if (active_bits == 10) {
switch(wavelet_index) {
case VC2DECODER_WFT_LEGALL_5_3:
return LeGall_5_3_invtransform_H_final_1_10_sse4_2<int32_t>;
case VC2DECODER_WFT_HAAR_NO_SHIFT:
return Haar_invtransform_H_final_1_10_sse4_2_int32_t<0>;
case VC2DECODER_WFT_HAAR_SINGLE_SHIFT:
return Haar_invtransform_H_final_1_10_sse4_2_int32_t<1>;
default:
break;
}
}
} else if (sample_size == 2) {
if (active_bits == 10) {
switch(wavelet_index) {
case VC2DECODER_WFT_LEGALL_5_3:
return LeGall_5_3_invtransform_H_final_1_10_sse4_2<int16_t>;
case VC2DECODER_WFT_HAAR_NO_SHIFT:
return Haar_invtransform_H_final_1_10_sse4_2_int16_t<0>;
case VC2DECODER_WFT_HAAR_SINGLE_SHIFT:
return Haar_invtransform_H_final_1_10_sse4_2_int16_t<1>;
default:
break;
}
}
}
return get_invhtransformfinal_c(wavelet_index, active_bits, sample_size);
}
int perform_invhtransformfinaltest(invhtransformfinaltest_data &data,
void *idata_pre,
const int width,
const int height,
const int stride,
bool HAS_SSE4_2, bool HAS_AVX, bool HAS_AVX2) {
int r = 0;
(void)HAS_AVX;(void)HAS_AVX2;
void *idata = ALIGNED_ALLOC(32, height*stride*data.sample_size);
memcpy(idata, idata_pre, height*stride*data.sample_size);
struct offsets_t offsets[] = { { 0, 0, 0, 0 },
{ 32, 0, 0, 0 },
{ 0, 32, 0, 0 },
{ 32, 32, 0, 0 },
{ 0, 0, 32, 0 },
{ 0, 0, 0, 32 },
{ 0, 0, 32, 32 },
{ 32, 0, 32, 0 },
{ 32, 0, 0, 32 },
{ 32, 0, 32, 32 },
{ 0, 32, 32, 0 },
{ 0, 32, 0, 32 },
{ 0, 0, 32, 32 },
{ 32, 32, 32, 0 },
{ 32, 32, 0, 32 },
{ 32, 32, 32, 32 },
};
for (int i = 0; r==0 && i < (int)(sizeof(offsets)/sizeof(struct offsets_t)); i++) {
char *cdata = (char *)malloc(height*stride*sizeof(uint16_t));
memset(cdata, 0, height*stride*sizeof(uint16_t));
printf("%-20s: H 0/* (%2d,%2d,%2d,%2d) (active %d-bit) ", VC2DecoderWaveletFilterTypeString[data.wavelet], offsets[i].left, offsets[i].right, offsets[i].top, offsets[i].bottom, data.active_bits);
if (data.sample_size == 2)
printf("16-bit ");
else
printf("32-bit ");
/* Use C version to generate comparison value */
printf("C [ ");
InplaceTransformFinal ctrans = NULL;
try {
ctrans = get_invhtransformfinal_c(data.wavelet, data.active_bits, data.sample_size);
} catch(...) {
printf(" NONE ]");
r = 1;
free(cdata);
break;
}
ctrans(idata, stride, cdata + (offsets[i].top*stride + offsets[i].left)*2, stride, width, height, offsets[i].left, offsets[i].top, width - offsets[i].left - offsets[i].right, height - offsets[i].top - offsets[i].bottom);
if (memcmp(idata, idata_pre, height*stride*data.sample_size) != 0) {
printf(" BAD ]\n");
printf(" c function overwrites input data!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
for (int y = 0; y < offsets[i].top; y++) {
for (int x = 0; x < stride; x++) {
if (((uint16_t *)cdata)[y*stride + x] != 0) {
printf(" BAD ]\n");
printf(" c function writes outside of specified memory area!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
}
}
for (int y = offsets[i].top; y < height - offsets[i].bottom; y++) {
for (int x = 0; x < offsets[i].left; x++) {
if (((uint16_t *)cdata)[y*stride + x] != 0) {
printf(" BAD ]\n");
printf(" c function writes outside of specified memory area!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
}
for (int x = offsets[i].left; x < width - offsets[i].right; x++) {
if (((uint16_t *)cdata)[y*stride + x] >= (1 << data.active_bits)) {
printf(" BAD ]\n");
printf(" c function does not clip values to correct number of bits!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
}
for (int x = width - offsets[i].right; x < stride; x++) {
if (((uint16_t *)cdata)[y*stride + x] != 0) {
printf(" BAD ]\n");
printf(" c function writes outside of specified memory area!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
}
}
for (int y = height - offsets[i].bottom; y < height; y++) {
for (int x = 0; x < stride; x++) {
if (((uint16_t *)cdata)[y*stride + x] != 0) {
printf(" BAD ]\n");
printf(" c function writes outside of specified memory area!\n");
r = 1;
free(cdata);
printf("\n");
break;
}
}
}
printf(" GOOD ] ");
/* Test SSE4_2 version */
if (HAS_SSE4_2 && data.SSE4_2) {
printf("SSE4.2 [");
InplaceTransformFinal trans = get_invhtransformfinal_sse4_2(data.wavelet, data.active_bits, data.sample_size);
if (trans == ctrans) {
printf("NONE]");
} else {
char *tdata = (char *)malloc(height*stride*sizeof(uint16_t));
memset(tdata, 0, height*stride*sizeof(uint16_t));
trans(idata, stride, tdata + (offsets[i].top*stride + offsets[i].left)*2, stride, width, height, offsets[i].left, offsets[i].top, width - offsets[i].left - offsets[i].right, height - offsets[i].top - offsets[i].bottom);
if (memcmp(cdata, tdata, height*stride*sizeof(uint16_t))) {
printf("FAIL]\n");
for (int i = 0; i < (int)(height*stride*sizeof(uint16_t)); i++) {
if (cdata[i] != tdata[i]) {
printf("\nFirst difference at byte %d, 0x%02x =/= 0x%02x\n\n", i, ((uint8_t *)cdata)[i], ((uint8_t *)tdata)[i]);
break;
}
}
r = 1;
} else {
printf(" OK ] ");
}
free(tdata);
}
}
free(cdata);
printf("\n");
}
ALIGNED_FREE(idata);
return r;
}
