Hi,
The attached patch adds readpixel acceleration to the r100 based on what's done for the r200.
This speeds-up readpix by a factor of 2 on x86, and much more on non-x86 systems which don't benefit from Ian's patch (like 8 times on itanium 2).
Stephane
diff -Naur radeon.old/Makefile radeon/Makefile
--- radeon.old/Makefile 2004-11-08 21:38:48.000000000 +0100
+++ radeon/Makefile 2004-11-03 17:07:59.000000000 +0100
@@ -22,6 +22,7 @@
radeon_context.c \
radeon_ioctl.c \
radeon_lock.c \
+ radeon_pixel.c \
radeon_screen.c \
radeon_state.c \
radeon_state_init.c \
diff -Naur radeon.old/radeon_context.c radeon/radeon_context.c
--- radeon.old/radeon_context.c 2004-11-08 21:38:48.000000000 +0100
+++ radeon/radeon_context.c 2004-11-08 21:00:47.000000000 +0100
@@ -343,7 +343,14 @@
ctx->Const.MaxLineWidth = 10.0;
ctx->Const.MaxLineWidthAA = 10.0;
ctx->Const.LineWidthGranularity = 0.0625;
-
+ if (screen->cpp == 4)
+ {
+ ctx->Const.ColorReadFormat = GL_BGRA;
+ ctx->Const.ColorReadType = GL_UNSIGNED_INT_8_8_8_8_REV;
+ } else {
+ ctx->Const.ColorReadFormat = GL_BGR;
+ ctx->Const.ColorReadType = GL_UNSIGNED_SHORT_5_6_5_REV;
+ }
/* Set maxlocksize (and hence vb size) small enough to avoid
* fallbacks in radeon_tcl.c. ie. guarentee that all vertices can
* fit in a single dma buffer for indexed rendering of quad strips,
diff -Naur radeon.old/radeon_context.h radeon/radeon_context.h
--- radeon.old/radeon_context.h 2004-11-08 21:38:48.000000000 +0100
+++ radeon/radeon_context.h 2004-11-08 21:25:50.000000000 +0100
@@ -847,6 +847,7 @@
#define DEBUG_DRI 0x200
#define DEBUG_DMA 0x400
#define DEBUG_SANITY 0x800
+#define DEBUG_PIXEL 0x2000
#endif
#endif /* __RADEON_CONTEXT_H__ */
diff -Naur radeon.old/radeon_ioctl.c radeon/radeon_ioctl.c
--- radeon.old/radeon_ioctl.c 2004-11-08 21:38:48.000000000 +0100
+++ radeon/radeon_ioctl.c 2004-11-08 21:01:33.000000000 +0100
@@ -59,7 +59,7 @@
static void radeonWaitForIdle( radeonContextPtr rmesa );
-static int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
+int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
const char * caller );
static void radeonSaveHwState( radeonContextPtr rmesa )
@@ -518,7 +518,7 @@
}
-static int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
+int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
const char * caller )
{
int ret, i;
@@ -752,6 +752,8 @@
rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */
rmesa->dma.current.start =
rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;
+
+ assert( rmesa->dma.current.ptr <= rmesa->dma.current.end );
}
void radeonAllocDmaRegionVerts( radeonContextPtr rmesa,
@@ -1233,6 +1235,30 @@
radeonWaitForIdle( rmesa );
}
+GLboolean radeonIsGartMemory( radeonContextPtr rmesa, const GLvoid *pointer,
+ GLint size )
+{
+ ptrdiff_t offset = (char *)pointer - (char
*)rmesa->radeonScreen->gartTextures.map;
+ int valid = (size >= 0 &&
+ offset >= 0 &&
+ offset + size < rmesa->radeonScreen->gartTextures.size);
+
+ if (RADEON_DEBUG & DEBUG_IOCTL)
+ fprintf(stderr, "radeonIsGartMemory( %p ) : %d\n", pointer, valid );
+
+ return valid;
+}
+
+GLuint radeonGartOffsetFromVirtual( radeonContextPtr rmesa, const GLvoid
*pointer )
+{
+ ptrdiff_t offset = (char *)pointer - (char
*)rmesa->radeonScreen->gartTextures.map;
+
+ if (offset < 0 || offset > rmesa->radeonScreen->gartTextures.size)
+ return ~0;
+ else
+ return rmesa->radeonScreen->gart_texture_offset + offset;
+}
+
void radeonInitIoctlFuncs( GLcontext *ctx )
{
diff -Naur radeon.old/radeon_ioctl.h radeon/radeon_ioctl.h
--- radeon.old/radeon_ioctl.h 2004-11-08 21:38:48.000000000 +0100
+++ radeon/radeon_ioctl.h 2004-11-03 17:04:48.000000000 +0100
@@ -77,6 +77,8 @@
extern void radeonEmitWait( radeonContextPtr rmesa, GLuint flags );
+extern int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
+ const char * caller );
extern void radeonFlushCmdBuf( radeonContextPtr rmesa, const char * );
extern void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa );
@@ -99,6 +101,12 @@
extern void radeonPageFlip( const __DRIdrawablePrivate *drawable );
extern void radeonFlush( GLcontext *ctx );
extern void radeonFinish( GLcontext *ctx );
+extern GLboolean radeonIsGartMemory( radeonContextPtr rmesa,
+ const GLvoid *pointer,
+ GLint size );
+extern GLuint radeonGartOffsetFromVirtual( radeonContextPtr rmesa,
+ const GLvoid *pointer );
+
extern void radeonWaitForIdleLocked( radeonContextPtr rmesa );
extern void radeonWaitForVBlank( radeonContextPtr rmesa );
extern void radeonInitIoctlFuncs( GLcontext *ctx );
diff -Naur radeon.old/radeon_pixel.c radeon/radeon_pixel.c
--- radeon.old/radeon_pixel.c 1970-01-01 01:00:00.000000000 +0100
+++ radeon/radeon_pixel.c 2004-11-08 21:21:42.000000000 +0100
@@ -0,0 +1,313 @@
+#include "glheader.h"
+#include "enums.h"
+#include "mtypes.h"
+#include "macros.h"
+#include "swrast/swrast.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_pixel.h"
+#include "radeon_swtcl.h"
+
+
+static GLboolean
+check_color( const GLcontext *ctx, GLenum type, GLenum format,
+ const struct gl_pixelstore_attrib *packing,
+ const void *pixels, GLint sz )
+{
+ radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+ GLuint cpp = rmesa->radeonScreen->cpp;
+
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s\n", __FUNCTION__);
+
+ if ( ctx->_ImageTransferState ||
+ packing->SwapBytes ||
+ packing->LsbFirst) {
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s: failed 1\n", __FUNCTION__);
+ return GL_FALSE;
+ }
+
+ if ( type == GL_UNSIGNED_SHORT_5_6_5_REV &&
+ cpp == 2 &&
+ format == GL_BGR ) {
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
+ return GL_TRUE;
+ }
+
+ if ( type == GL_UNSIGNED_INT_8_8_8_8_REV &&
+ cpp == 4 &&
+ format == GL_BGRA ) {
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
+ return GL_TRUE;
+ }
+
+ fprintf(stderr, "type %d %d cpp %d %d format %d
%d\n",type,GL_UNSIGNED_INT_8_8_8_8_REV,cpp,4,format,GL_BGRA);
+
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s: failed\n", __FUNCTION__);
+
+ return GL_FALSE;
+}
+
+
+static GLboolean
+clip_pixelrect( const GLcontext *ctx,
+ const GLframebuffer *buffer,
+ GLint *x, GLint *y,
+ GLsizei *width, GLsizei *height,
+ GLint *size )
+{
+ radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+ /* left clipping */
+ if (*x < buffer->_Xmin) {
+ *width -= (buffer->_Xmin - *x);
+ *x = buffer->_Xmin;
+ }
+
+ /* right clipping */
+ if (*x + *width > buffer->_Xmax)
+ *width -= (*x + *width - buffer->_Xmax - 1);
+
+ if (*width <= 0)
+ return GL_FALSE;
+
+ /* bottom clipping */
+ if (*y < buffer->_Ymin) {
+ *height -= (buffer->_Ymin - *y);
+ *y = buffer->_Ymin;
+ }
+
+ /* top clipping */
+ if (*y + *height > buffer->_Ymax)
+ *height -= (*y + *height - buffer->_Ymax - 1);
+
+ if (*height <= 0)
+ return GL_FALSE;
+
+ *size = ((*y + *height - 1) * rmesa->radeonScreen->frontPitch +
+ (*x + *width - 1) * rmesa->radeonScreen->cpp);
+
+ return GL_TRUE;
+}
+
+static void do_read_pix( GLcontext *ctx,
+ GLint x, GLint y, GLsizei width, GLsizei height,
+ GLint pitch,
+ const void *pixels,
+ int blit_format)
+{
+ radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+ __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+ int nbox = dPriv->numClipRects;
+ int src_offset = rmesa->state.color.drawOffset
+ + rmesa->radeonScreen->fbLocation;
+ int dst_offset;
+ drm_clip_rect_t *box = dPriv->pClipRects;
+ int i;
+ int src_pitch = rmesa->state.color.drawPitch * rmesa->radeonScreen->cpp;
+ int dst_pitch = pitch;
+ GLint mx1,mx2,my1,my2,mw,mh;
+
+ if (!nbox)
+ return;
+
+ y = dPriv->h - y - height;
+ x += dPriv->x;
+ y += dPriv->y;
+
+
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "readpixel blit src_pitch %d dst_pitch %d\n",
+ src_pitch, dst_pitch);
+
+ mx1 = box[0].x1;
+ my1 = box[0].y1;
+ mx2 = box[0].x2;
+ my2 = box[0].y2;
+ /* merge the rects */
+ for (i = 1 ; i < nbox ; i++)
+ {
+ if (box[i].x1 < mx1) mx1 = box[i].x1;
+ if (box[i].y1 < my1) my1 = box[i].y1;
+ if (box[i].x2 > mx2) mx2 = box[i].x2;
+ if (box[i].y2 > my2) my2 = box[i].y2;
+ }
+ mw=mx2-mx1;
+ mh=my2-my1;
+
+ /* intersect with the area we are supposed to read */
+ if (mx1 < x) mw -= x - mx1, mx1 = x;
+ if (my1 < y) mh -= y - my1, my1 = y;
+ if (mx1 + mw > x + width) mw = x + width - mx1;
+ if (my1 + mh > y + height) mh = y + height - my1;
+ if (mw <= 0) return;
+ if (mh <= 0) return;
+ mx2=mx1+mw;
+ my2=my1+mh;
+
+ dst_offset = radeonGartOffsetFromVirtual( rmesa, pixels );
+ radeonEmitWait( rmesa, RADEON_WAIT_3D );
+
+ radeonEmitBlit( rmesa,
+ blit_format,
+ src_pitch, src_offset,
+ dst_pitch, dst_offset,
+ mx1, my1,
+ mx1 - x, my1 - y,
+ mx2 - mx1, my2 - my1 );
+
+ radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
+}
+
+static GLboolean
+radeonTryReadPixels( GLcontext *ctx,
+ GLint x, GLint y, GLsizei width, GLsizei height,
+ GLenum format, GLenum type,
+ const struct gl_pixelstore_attrib *pack,
+ GLvoid *pixels )
+{
+ radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+ GLuint cpp = rmesa->radeonScreen->cpp;
+ GLint pitch = (pack->RowLength ? pack->RowLength : width) * cpp;
+ GLint size;
+ GLint blit_format;
+
+ pitch= (pitch+63)&~63;
+
+ size=pitch * height;
+
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s\n", __FUNCTION__);
+
+ /* this is the threshold above which a blit is faster than video memory
acccess by the CPU
+ * tuning for this is approximate */
+ if (size<64)
+ return GL_FALSE;
+
+ if (!check_color(ctx, type, format, pack, pixels, size))
+ return GL_FALSE;
+
+ switch ( rmesa->radeonScreen->cpp ) {
+ case 2:
+ blit_format = RADEON_GMC_DST_16BPP;
+ break;
+ case 4:
+ blit_format = RADEON_GMC_DST_32BPP;
+ break;
+ default:
+ return GL_FALSE;
+ }
+
+
+ /* Although the blits go on the command buffer, need to do this and
+ * fire with lock held to guarentee cliprects and drawOffset are
+ * correct.
+ *
+ * This is an unusual situation however, as the code which flushes
+ * a full command buffer expects to be called unlocked. As a
+ * workaround, immediately flush the buffer on aquiring the lock.
+ */
+ LOCK_HARDWARE( rmesa );
+
+ if (rmesa->store.cmd_used)
+ radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
+
+ if (!clip_pixelrect(ctx, ctx->ReadBuffer, &x, &y, &width, &height,
+ &size)) {
+ UNLOCK_HARDWARE( rmesa );
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s totally clipped -- nothing to do\n",
+ __FUNCTION__);
+ return GL_TRUE;
+ }
+
+ {
+ /* Pixels is in regular memory -- get dma buffers and perform
+ * upload through them.
+ */
+
+ int ret;
+ int region_offset;
+ drm_radeon_mem_alloc_t alloc;
+ drm_radeon_mem_free_t memfree;
+ char* region;
+ int yl;
+
+ alloc.region = RADEON_MEM_REGION_GART;
+ alloc.alignment = 1024;
+ alloc.size = size;
+ alloc.region_offset = ®ion_offset;
+
+ ret = drmCommandWriteRead( rmesa->radeonScreen->driScreen->fd,
+ DRM_RADEON_ALLOC,
+ &alloc, sizeof(alloc));
+ if (ret) {
+ UNLOCK_HARDWARE( rmesa );
+ return GL_FALSE;
+ }
+ region =(void *)((char *)rmesa->radeonScreen->gartTextures.map +
region_offset);
+
+ do_read_pix( ctx, x, y, width, height, pitch, region, blit_format);
+ UNLOCK_HARDWARE( rmesa );
+ radeonFinish( ctx ); /* required for GL */
+
+ /* the thing is upside down on the card, swap it */
+ for (yl=0;yl<height;yl++)
+ {
+ memcpy(((char*)pixels)+yl*(pack->RowLength ? pack->RowLength :
width)*cpp,
+ (void *)((char *)rmesa->radeonScreen->gartTextures.map
+ region_offset + (height-1-yl)*pitch),
+ (pack->RowLength ? pack->RowLength : width)*cpp);
+ }
+
+ memfree.region = RADEON_MEM_REGION_GART;
+ memfree.region_offset = region_offset;
+
+ ret = drmCommandWrite( rmesa->radeonScreen->driScreen->fd,
+ DRM_RADEON_FREE,
+ &memfree, sizeof(memfree));
+ if (ret)
+ fprintf(stderr, "%s: DRM_RADEON_FREE ret %d\n", __FUNCTION__,
ret);
+ return GL_TRUE;
+ }
+
+}
+
+static void
+radeonReadPixels( GLcontext *ctx,
+ GLint x, GLint y, GLsizei width, GLsizei height,
+ GLenum format, GLenum type,
+ const struct gl_pixelstore_attrib *pack,
+ GLvoid *pixels )
+{
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "%s\n", __FUNCTION__);
+
+ if (!radeonTryReadPixels( ctx, x, y, width, height, format, type, pack,
+ pixels))
+ _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack,
+ pixels);
+}
+
+
+void radeonInitPixelFuncs( GLcontext *ctx )
+{
+ /* Pixel path fallbacks.
+ */
+ ctx->Driver.Accum = _swrast_Accum;
+ ctx->Driver.Bitmap = _swrast_Bitmap;
+ ctx->Driver.CopyPixels = _swrast_CopyPixels;
+ ctx->Driver.DrawPixels = _swrast_DrawPixels;
+ ctx->Driver.ReadPixels = _swrast_ReadPixels;
+
+ if (!getenv("RADEON_NO_BLITS") && RADEON_CONTEXT(ctx)->dri.drmMinor >= 6) {
+ ctx->Driver.ReadPixels = radeonReadPixels;
+ }
+}
+
+
+
diff -Naur radeon.old/radeon_pixel.h radeon/radeon_pixel.h
--- radeon.old/radeon_pixel.h 1970-01-01 01:00:00.000000000 +0100
+++ radeon/radeon_pixel.h 2004-11-03 02:00:54.000000000 +0100
@@ -0,0 +1,40 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r200/r200_pixel.h,v 1.1 2002/10/30
12:51:52 alanh Exp $ */
+/*
+Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+/*
+ * Authors:
+ * Keith Whitwell <[EMAIL PROTECTED]>
+ */
+
+#ifndef __RADEON_PIXEL_H__
+#define __RADEON_PIXEL_H__
+
+extern void radeonInitPixelFuncs( GLcontext *ctx );
+
+#endif
diff -Naur radeon.old/radeon_state.c radeon/radeon_state.c
--- radeon.old/radeon_state.c 2004-11-08 21:38:48.000000000 +0100
+++ radeon/radeon_state.c 2004-11-03 17:01:57.000000000 +0100
@@ -55,6 +55,7 @@
#include "radeon_tex.h"
#include "radeon_swtcl.h"
#include "radeon_vtxfmt.h"
+#include "radeon_pixel.h"
/* =============================================================
* Alpha blending
@@ -2225,11 +2226,7 @@
/* Pixel path fallbacks
*/
- ctx->Driver.Accum = _swrast_Accum;
- ctx->Driver.Bitmap = _swrast_Bitmap;
- ctx->Driver.CopyPixels = _swrast_CopyPixels;
- ctx->Driver.DrawPixels = _swrast_DrawPixels;
- ctx->Driver.ReadPixels = _swrast_ReadPixels;
+ radeonInitPixelFuncs(ctx);
/* Swrast hooks for imaging extensions:
*/
