fix softblur

This commit is contained in:
veejay
2023-10-30 16:58:54 +01:00
parent e561655917
commit a2a2f8ca60

View File

@@ -38,19 +38,46 @@ vj_effect *softblur_init(int w,int h)
ve->defaults[0] = 0; ve->defaults[0] = 0;
ve->limits[0][0] = 0; ve->limits[0][0] = 0;
ve->limits[1][0] = 1; /* 3*/ ve->limits[1][0] = 2; /* 3*/
ve->description = "Soft Blur (1x3) and (3x3)"; ve->description = "Soft Blur";
ve->sub_format = -1; ve->sub_format = -1;
ve->extra_frame = 0; ve->extra_frame = 0;
ve->has_user = 0; ve->has_user = 0;
ve->param_description = vje_build_param_list(ve->num_params, "Kernel size"); ve->param_description = vje_build_param_list(ve->num_params, "Kernel Size");
ve->hints = vje_init_value_hint_list( ve->num_params );
vje_build_value_hint_list( ve->hints, ve->limits[1][0], 0, "1x3", "3x3","5x5");
return ve; return ve;
} }
static void softblur5_apply(VJFrame *frame)
{
int r, c, i, j;
uint8_t *restrict Y = frame->data[0];
const int len = frame->len;
const int width = frame->width;
for (r = 2 * width; r < len - 2 * width; r += width) {
#pragma omp simd
for (c = 2; c < width - 2; c++) {
int sum = 0;
for (i = -2; i <= 2; i++) {
for (j = -2; j <= 2; j++) {
sum += Y[r + i * width + c + j];
}
}
Y[r + c] = sum / 25;
}
}
}
static void softblur3_apply(VJFrame *frame ) static void softblur3_apply(VJFrame *frame )
{ {
int r,c; int r,c;
uint8_t *Y = frame->data[0]; uint8_t *restrict Y = frame->data[0];
const int len = frame->len; const int len = frame->len;
const int width = frame->width; const int width = frame->width;
@@ -61,6 +88,7 @@ static void softblur3_apply(VJFrame *frame )
) / 3; ) / 3;
for(r=width; r < (len-width); r+=width) { for(r=width; r < (len-width); r+=width) {
#pragma omp simd
for(c=1; c < (width-1); c++) { for(c=1; c < (width-1); c++) {
Y[r+c] = ( Y[r - width + c - 1] + Y[r+c] = ( Y[r - width + c - 1] +
Y[r - width + c ] + Y[r - width + c ] +
@@ -80,188 +108,29 @@ static void softblur3_apply(VJFrame *frame )
Y[c] + Y[c] +
Y[c + 1] Y[c + 1]
) / 3; ) / 3;
} }
#ifdef HAVE_ASM_MMX
/* mmx_blur() taken from libvisual plugins
*
* Libvisual-plugins - Standard plugins for libvisual
*
* Copyright (C) 2002, 2003, 2004, 2005 Dennis Smit <ds@nerds-incorporated.org>
*
* Authors: Dennis Smit <ds@nerds-incorporated.org>
*/
static void mmx_blur(VJFrame *frame)
{
__asm __volatile
("\n\t pxor %%mm6, %%mm6"
::);
const int len = frame->len;
const int width = frame->width;
int scrsh = (len) >> 1;
int i;
uint8_t *buf = frame->data[0];
/* Prepare substraction register */
for (i = 0; i < scrsh; i += 4) {
__asm __volatile
("\n\t movd %[buf], %%mm0"
"\n\t movd %[add1], %%mm1"
"\n\t punpcklbw %%mm6, %%mm0"
"\n\t movd %[add2], %%mm2"
"\n\t punpcklbw %%mm6, %%mm1"
"\n\t movd %[add3], %%mm3"
"\n\t punpcklbw %%mm6, %%mm2"
"\n\t paddw %%mm1, %%mm0"
"\n\t punpcklbw %%mm6, %%mm3"
"\n\t paddw %%mm2, %%mm0"
"\n\t paddw %%mm3, %%mm0"
"\n\t psrlw $2, %%mm0"
"\n\t packuswb %%mm6, %%mm0"
"\n\t movd %%mm0, %[buf]"
:: [buf] "m" (*(buf + i))
, [add1] "m" (*(buf + i + width))
, [add2] "m" (*(buf + i + width + 1))
, [add3] "m" (*(buf + i + width - 1))
);
// : "mm0", "mm1", "mm2", "mm3", "mm6");
}
for (i = len; i > scrsh; i -= 4) {
__asm __volatile
("\n\t movd %[buf], %%mm0"
"\n\t movd %[add1], %%mm1"
"\n\t punpcklbw %%mm6, %%mm0"
"\n\t movd %[add2], %%mm2"
"\n\t punpcklbw %%mm6, %%mm1"
"\n\t movd %[add3], %%mm3"
"\n\t punpcklbw %%mm6, %%mm2"
"\n\t paddw %%mm1, %%mm0"
"\n\t punpcklbw %%mm6, %%mm3"
"\n\t paddw %%mm2, %%mm0"
"\n\t paddw %%mm3, %%mm0"
"\n\t psrlw $2, %%mm0"
"\n\t packuswb %%mm6, %%mm0"
"\n\t movd %%mm0, %[buf]"
:: [buf] "m" (*(buf + i))
, [add1] "m" (*(buf + i + width))
, [add2] "m" (*(buf + i + 1))
, [add3] "m" (*(buf + i + width - 1))
);// : "mm0", "mm1", "mm2", "mm3", "mm6");
}
do_emms;
}
#endif
#if !defined(HAVE_ASM_MMX) && !defined(HAVE_ARM)
static void softblur1_apply( VJFrame *frame) static void softblur1_apply( VJFrame *frame)
{ {
int r, c; int r, c;
const int len = frame->len; const int len = frame->len;
uint8_t *Y = frame->data[0]; uint8_t *restrict Y = frame->data[0];
const int width = frame->width; const int width = frame->width;
for (r = 0; r < len; r += width) { #pragma omp simd
for (c = 1; c < width-1; c++) { for(r=0; r < len; r+=width) {
Y[c + r] = (Y[r + c - 1] + for(c=1; c < (width-1); c++) {
Y[r + c] + Y[r+c] = (
Y[r + c + 1] Y[r + c + 1] +
) / 3; Y[r + c] +
Y[r + c + 1] ) /3;
} }
} }
} }
#endif
#ifdef HAVE_ARM
static void softblur1_apply(VJFrame *frame) {
const int len = frame->len;
uint8_t *Y = frame->data[0];
const int width = frame->width;
const int aligned_width = (width / 16) * 16;
for (int r = 0; r < len; r += width) {
for (int c = 1; c < aligned_width - 1; c += 16) {
uint8x16_t prev = vld1q_u8(&Y[r + c - 1]);
uint8x16_t current = vld1q_u8(&Y[r + c]);
uint8x16_t next = vld1q_u8(&Y[r + c + 1]);
uint16x8_t sum_low = vaddl_u8(vget_low_u8(prev), vget_low_u8(current));
uint16x8_t sum_high = vaddl_u8(vget_high_u8(prev), vget_high_u8(current));
sum_low = vaddw_u8(sum_low, vget_low_u8(next));
sum_high = vaddw_u8(sum_high, vget_high_u8(next));
uint8x16_t result = vcombine_u8(vshrn_n_u16(sum_low, 2), vshrn_n_u16(sum_high, 2));
vst1q_u8(&Y[r + c], result);
}
for (int c = aligned_width; c < width - 1; c++) {
Y[r + c] = (Y[r + c - 1] + Y[r + c] + Y[r + c + 1]) / 3;
}
}
}
#endif
#ifdef HAVE_ASM_SSE2
static void sse2_blur(VJFrame *frame) {
const int len = frame->len;
const int width = frame->width;
int scrsh = len >> 1;
uint8_t *buf = frame->data[0];
for (int i = 0; i < scrsh; i += 16) {
__m128i mm0 = _mm_load_si128((__m128i *)(buf + i));
__m128i mm1 = _mm_load_si128((__m128i *)(buf + i + width));
__m128i mm2 = _mm_load_si128((__m128i *)(buf + i + width + 1));
__m128i mm3 = _mm_load_si128((__m128i *)(buf + i + width - 1));
mm0 = _mm_unpacklo_epi8(mm0, _mm_setzero_si128());
mm1 = _mm_unpacklo_epi8(mm1, _mm_setzero_si128());
mm2 = _mm_unpacklo_epi8(mm2, _mm_setzero_si128());
mm3 = _mm_unpacklo_epi8(mm3, _mm_setzero_si128());
mm0 = _mm_add_epi16(mm0, mm1);
mm0 = _mm_add_epi16(mm0, mm2);
mm0 = _mm_add_epi16(mm0, mm3);
mm0 = _mm_srli_epi16(mm0, 2);
mm0 = _mm_packus_epi16(mm0, mm0);
_mm_storel_epi64((__m128i *)(buf + i), mm0);
}
for (int i = len - 16; i > scrsh; i -= 16) {
__m128i mm0 = _mm_load_si128((__m128i *)(buf + i));
__m128i mm1 = _mm_load_si128((__m128i *)(buf + i + width));
__m128i mm2 = _mm_load_si128((__m128i *)(buf + i + 1));
__m128i mm3 = _mm_load_si128((__m128i *)(buf + i + width - 1));
mm0 = _mm_unpacklo_epi8(mm0, _mm_setzero_si128());
mm1 = _mm_unpacklo_epi8(mm1, _mm_setzero_si128());
mm2 = _mm_unpacklo_epi8(mm2, _mm_setzero_si128());
mm3 = _mm_unpacklo_epi8(mm3, _mm_setzero_si128());
mm0 = _mm_add_epi16(mm0, mm1);
mm0 = _mm_add_epi16(mm0, mm2);
mm0 = _mm_add_epi16(mm0, mm3);
mm0 = _mm_srli_epi16(mm0, 2);
mm0 = _mm_packus_epi16(mm0, mm0);
_mm_storel_epi64((__m128i *)(buf + i), mm0);
}
_mm_empty();
}
#endif
void softblur_apply(void *ptr, VJFrame *frame, int *args) void softblur_apply(void *ptr, VJFrame *frame, int *args)
@@ -271,19 +140,14 @@ void softblur_apply(void *ptr, VJFrame *frame, int *args)
switch (type) { switch (type) {
case 0: case 0:
#ifdef HAVE_ASM_SSE2
sse2_blur(frame);
#else
#ifdef HAVE_ASM_MMX
mmx_blur(frame);
#else
softblur1_apply(frame); softblur1_apply(frame);
#endif
#endif
break; break;
case 1: case 1:
softblur3_apply(frame); softblur3_apply(frame);
break; break;
case 2:
softblur5_apply(frame);
break;
} }
} }
@@ -291,18 +155,13 @@ void softblur_apply_internal(VJFrame *frame, int type)
{ {
switch (type) { switch (type) {
case 0: case 0:
#ifdef HAVE_ASM_SSE2
sse2_blur(frame);
#else
#ifdef HAVE_ASM_MMX
mmx_blur(frame);
#else
softblur1_apply(frame); softblur1_apply(frame);
#endif
#endif
break; break;
case 1: case 1:
softblur3_apply(frame); softblur3_apply(frame);
break; break;
case 2:
softblur5_apply(frame);
break;
} }
} }