comparison libpostproc/postprocess_template.c @ 2642:240e17c3cb2d libavcodec

GCC4 fix by (Keenan Pepper (keenanpepper gmail com)

2641:c337f851d0f9

#ifdef HAVE_MMX
/**
 * accurate deblock filter
 */
static always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
	int64_t dc_mask, eq_mask;
	int64_t sums[10*8*2];
	src+= step*3; // src points to begin of the 8x8 Block
//START_TIMER
asm volatile(
		"movq %0, %%mm7					\n\t" 

		: "=m" (eq_mask), "=m" (dc_mask)
		: "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
		: "%"REG_a
		);

	if(dc_mask & eq_mask){
		long offset= -8*step;
		int64_t *temp_sums= sums;

		asm volatile(
		"movq %2, %%mm0					\n\t"  // QP,..., QP

		"add $16, %1					\n\t"
		"add %2, %0					\n\t"
		" js 1b						\n\t"

		: "+r"(offset), "+r"(temp_sums)
		: "r" ((long)step), "r"(src - offset), "m"(dc_mask & eq_mask)
		);
	}else
		src+= step; // src points to begin of the 8x8 Block

	if(eq_mask != -1LL){

2642:240e17c3cb2d

#ifdef HAVE_MMX
/**
 * accurate deblock filter
 */
static always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
	int64_t dc_mask, eq_mask, both_masks;
	int64_t sums[10*8*2];
	src+= step*3; // src points to begin of the 8x8 Block
//START_TIMER
asm volatile(
		"movq %0, %%mm7					\n\t" 

		: "=m" (eq_mask), "=m" (dc_mask)
		: "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
		: "%"REG_a
		);

	both_masks = dc_mask & eq_mask;

	if(both_masks){
		long offset= -8*step;
		int64_t *temp_sums= sums;

		asm volatile(
		"movq %2, %%mm0					\n\t"  // QP,..., QP

		"add $16, %1					\n\t"
		"add %2, %0					\n\t"
		" js 1b						\n\t"

		: "+r"(offset), "+r"(temp_sums)
		: "r" ((long)step), "r"(src - offset), "m"(both_masks)
		);
	}else
		src+= step; // src points to begin of the 8x8 Block

	if(eq_mask != -1LL){