Commit 7f379c05ea9e6b60c081af2c18a84ec290af2df4

zecke
Mon Nov 09 07:28:13 GMT 2009

Tree SHA1: d4d79ff
Parent SHA1: a69e152 (Create a specialized SourceOver routine in plain C)
raw diff | raw patch

Improve BYTE_MUL when we don't know the alpha value

In case we don't know the alpha value for byte mul provide
specialized routine to check for 0 and 255 values. This is
gives almost a frame speed up.

Diff rendering mode:
inline
side by side


122	122	// -- I could not verify this on a Cortex-A8 with the above UNROLL_LOOP_WITH_PLD
123	123	# define PRELOAD_COND2(x,y) if (((uint)&x[i])%L2CacheLineLength == 0) preload(&x[i] + L2CacheLineLengthInInts); \
124	124	if (((uint)&y[i])%L2CacheLineLength == 16) preload(&y[i] + L2CacheLineLengthInInts);
	125
	126	Q_STATIC_INLINE_FUNCTION uint BYTE_MUL_ARM(uint x, uint a) {
	127	switch (a) {
	128	case 255:
	129	return x;
	130	case 0:
	131	return 0;
	132	default: {
	133	uint t = (x & 0xff00ff) * a;
	134	t = (t + ((t >> 8) & 0xff00ff) + 0x800080) >> 8;
	135	t &= 0xff00ff;
	136
	137	x = ((x >> 8) & 0xff00ff) * a;
	138	x = (x + ((x >> 8) & 0xff00ff) + 0x800080);
	139	x &= 0xff00ff00;
	140	x \|= t;
	141	return x;
	142	}
	143	}
	144	}
	145
125	146	void comp_func_SourceOver_arm(uint dest, const uint src, int length, uint const_alpha)
126	147	{
127	148	PRELOAD_INIT2(dest, src)
…	…
155	155	dest[i] = src[i];
156	156	} else {
157	157	uint s = src[i];
158		dest[i] = s + BYTE_MUL(dest[i], qAlpha(~s));
	158	dest[i] = s + BYTE_MUL_ARM(dest[i], qAlpha(~s));
159	159	}
160	160
161	161	++i;
…	…
164	164	int i = 0;
165	165	while (i < length) {
166	166	uint s = BYTE_MUL(src[i], const_alpha);
167		dest[i] = s + BYTE_MUL(dest[i], qAlpha(~s));
	167	dest[i] = s + BYTE_MUL_ARM(dest[i], qAlpha(~s));
168	168	PRELOAD_COND2(dest, src)
169	169	++i;
170	170	}

Commit 7f379c05ea9e6b60c081af2c18a84ec290af2df4

Comments

Add a new comment:

Add your comment