Optimization. Saves one compare per DWORD for common case where BCI
queue has ample space for bitmap data.
Changelog:
* EXA: use memcpy instead of loop for UploadToScreen operation
--
perl -e '$x=2.4;print sprintf("%.0f + %.0f = %.0f\n",$x,$x,$x+$x);'
>From 70c9a579041f1ec588a580647966dff1e17e26d6 Mon Sep 17 00:00:00 2001
From: =?utf-8?q?Alex=20Villac=C3=ADs=20Lasso?= <[email protected]>
Date: Tue, 30 Dec 2008 01:24:42 -0500
Subject: [PATCH] EXA: Optimization to use one memcpy per scanline instead of a conditional inside a loop for every dword.
---
src/savage_exa.c | 20 ++++++++++++++------
1 files changed, 14 insertions(+), 6 deletions(-)
diff --git a/src/savage_exa.c b/src/savage_exa.c
index 7c6efb3..538e000 100644
--- a/src/savage_exa.c
+++ b/src/savage_exa.c
@@ -495,13 +495,21 @@ SavageUploadToScreen(PixmapPtr pDst, int x, int y, int w, int h, char *src, int
dwords = (((w * Bpp) + 3) >> 2);
for (i = 0; i < h; i++) {
srcp = (CARD32 *)src;
- for (j = 0; j < dwords; j++) {
- if (queue < 4) {
- BCI_RESET;
- queue = 120 * 1024;
+
+ if (4 * dwords <= queue) {
+ /* WARNING: breaking BCI_PTR abstraction here */
+ memcpy(bci_ptr, srcp, 4 * dwords);
+ bci_ptr += dwords;
+ queue -= 4 * dwords;
+ } else {
+ for (j = 0; j < dwords; j++) {
+ if (queue < 4) {
+ BCI_RESET;
+ queue = 120 * 1024;
+ }
+ BCI_SEND(*srcp++);
+ queue -= 4;
}
- BCI_SEND(*srcp++);
- queue -= 4;
}
src += src_pitch;
}
--
1.6.0.6
_______________________________________________
xorg mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/xorg
