[src/trunk]: src/sys/arch/arm/omap copy data in 32bit chunks whenever alignme...

[macallan <macallan%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/867c397d5051
branches:  trunk
changeset: 784058:867c397d5051
user:      macallan <macallan%NetBSD.org@localhost>
date:      Wed Jan 16 20:34:10 2013 +0000

description:
copy data in 32bit chunks whenever alignment permits
 -> very noticeable speedup

diffstat:

 sys/arch/arm/omap/omapfb.c |  80 +++++++++++++++++++++++++++++++++++++--------
 1 files changed, 66 insertions(+), 14 deletions(-)

diffs (125 lines):

diff -r 215243ac644b -r 867c397d5051 sys/arch/arm/omap/omapfb.c
--- a/sys/arch/arm/omap/omapfb.c        Wed Jan 16 20:32:24 2013 +0000
+++ b/sys/arch/arm/omap/omapfb.c        Wed Jan 16 20:34:10 2013 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: omapfb.c,v 1.11 2013/01/16 00:09:27 macallan Exp $     */
+/*     $NetBSD: omapfb.c,v 1.12 2013/01/16 20:34:10 macallan Exp $     */
 
 /*
  * Copyright (c) 2010 Michael Lorenz
@@ -31,7 +31,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: omapfb.c,v 1.11 2013/01/16 00:09:27 macallan Exp $");
+__KERNEL_RCSID(0, "$NetBSD: omapfb.c,v 1.12 2013/01/16 20:34:10 macallan Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -642,14 +642,43 @@
 omapfb_rectfill(struct omapfb_softc *sc, int x, int y, int wi, int he,
      uint32_t colour)
 {
-       int width_in_bytes = wi * (sc->sc_depth >> 3);
+       int bpp = sc->sc_depth >> 3;    /* bytes per pixel */
+       int width_in_bytes = wi * bpp;
+       uint32_t daddr;
+
+       daddr = sc->sc_fbhwaddr + sc->sc_stride * y + x * bpp;
+       omapfb_wait_idle(sc);
 
-       omapfb_wait_idle(sc);
+       /*
+        * stupid hardware
+        * in 32bit mode the DMA controller always writes 0 into the upper byte, so we
+        * can use this mode only if we actually want that
+        */
+       if (((colour & 0xff00) == 0) &&
+          (((daddr | width_in_bytes) & 3) == 0)) {
+               /*
+                * everything is properly aligned so we can copy stuff in
+                * 32bit chunks instead of pixel by pixel
+                */
+               wi = wi >> 1;
+
+               /* just in case */
+               colour |= colour << 16;
+
+               omapdma_write_ch_reg(0, OMAPDMAC_CSDPI,
+                   CSDPI_SRC_BURST_64 | CSDPI_DST_BURST_64 |
+                   CSDPI_DST_PACKED | CSDPI_WRITE_POSTED_EXCEPT_LAST |
+                   CSDPI_DATA_TYPE_32);
+       } else {
+               omapdma_write_ch_reg(0, OMAPDMAC_CSDPI,
+                   CSDPI_SRC_BURST_64 | CSDPI_DST_BURST_64 |
+                   CSDPI_DST_PACKED | CSDPI_WRITE_POSTED_EXCEPT_LAST |
+                   CSDPI_DATA_TYPE_16);
+       }
 
        omapdma_write_ch_reg(0, OMAPDMAC_CEN, wi);
        omapdma_write_ch_reg(0, OMAPDMAC_CFN, he);
-       omapdma_write_ch_reg(0, OMAPDMAC_CDSA,
-           sc->sc_fbhwaddr + sc->sc_stride * y + x * (sc->sc_depth >> 3));
+       omapdma_write_ch_reg(0, OMAPDMAC_CDSA, daddr);
        omapdma_write_ch_reg(0, OMAPDMAC_CCR,
            CCR_CONST_FILL_ENABLE | CCR_DST_AMODE_DOUBLE_INDEX |
            CCR_SRC_AMODE_CONST_ADDR);
@@ -665,14 +694,19 @@
 omapfb_bitblt(struct omapfb_softc *sc, int xs, int ys, int xd, int yd,
     int wi, int he, int rop)
 {
-       int width_in_bytes = wi * (sc->sc_depth >> 3);
+       int bpp = sc->sc_depth >> 3;    /* bytes per pixel */
+       int width_in_bytes = wi * bpp;
+       
        int hstep, vstep;
        uint32_t saddr, daddr;
 
-       omapfb_wait_idle(sc);
-
-       saddr = sc->sc_fbhwaddr + sc->sc_stride * ys + xs * (sc->sc_depth >> 3);
-       daddr = sc->sc_fbhwaddr + sc->sc_stride * yd + xd * (sc->sc_depth >> 3);
+       /*
+        * TODO:
+        * - use 32bit transfers if we're properly aligned
+        */
+       saddr = sc->sc_fbhwaddr + sc->sc_stride * ys + xs * bpp;
+       daddr = sc->sc_fbhwaddr + sc->sc_stride * yd + xd * bpp;
+               
        if (ys < yd) {
                /* need to go vertically backwards */
                vstep = 1 - (sc->sc_stride + width_in_bytes);
@@ -686,12 +720,30 @@
                 * and destination pixels are on the same line
                 */
                hstep = 1 - (sc->sc_depth >> 2);
-               vstep = sc->sc_stride + (sc->sc_depth >> 3) * (wi - 1) + 1;
-               saddr += (sc->sc_depth >> 3) * (wi - 1);
-               daddr += (sc->sc_depth >> 3) * (wi - 1);
+               vstep = sc->sc_stride + bpp * (wi - 1) + 1;
+               saddr += bpp * (wi - 1);
+               daddr += bpp * (wi - 1);
        } else
                hstep = 1;
 
+       omapfb_wait_idle(sc);
+       if (((saddr | daddr | width_in_bytes) & 3) == 0) {
+               /*
+                * everything is properly aligned so we can copy stuff in
+                * 32bit chunks instead of pixel by pixel
+                */
+               wi = wi >> 1;
+               omapdma_write_ch_reg(0, OMAPDMAC_CSDPI,
+                   CSDPI_SRC_BURST_64 | CSDPI_DST_BURST_64 |
+                   CSDPI_DST_PACKED | CSDPI_WRITE_POSTED_EXCEPT_LAST |
+                   CSDPI_DATA_TYPE_32);
+       } else {
+               omapdma_write_ch_reg(0, OMAPDMAC_CSDPI,
+                   CSDPI_SRC_BURST_64 | CSDPI_DST_BURST_64 |
+                   CSDPI_DST_PACKED | CSDPI_WRITE_POSTED_EXCEPT_LAST |
+                   CSDPI_DATA_TYPE_16);
+       }
+
        omapdma_write_ch_reg(0, OMAPDMAC_CEN, wi);
        omapdma_write_ch_reg(0, OMAPDMAC_CFN, he);
        omapdma_write_ch_reg(0, OMAPDMAC_CSSA, saddr);