xserver/hw/kdrive/ati ati.h, 1.11, 1.12 ati_dma.h, 1.1,
1.2 ati_draw.c, 1.20, 1.21 ati_draw.h, 1.7, 1.8 ati_reg.h, 1.7,
1.8 r128_composite.c, 1.4, 1.5 radeon_composite.c, 1.10, 1.11
Eric Anholt
xserver-commit at pdx.freedesktop.org
Sat Sep 11 02:28:22 PDT 2004
Committed by: anholt
Update of /cvs/xserver/xserver/hw/kdrive/ati
In directory gabe:/tmp/cvs-serv16272/hw/kdrive/ati
Modified Files:
ati.h ati_dma.h ati_draw.c ati_draw.h ati_reg.h
r128_composite.c radeon_composite.c
Log Message:
- Add disabled WIP trapezoid code for R128 and R100. The R128 rendering
is not doing an add of 1 per triangle like I hoped, and instead seems
to be saturating all the pixels or something. The R100 acceleration
renders pretty well, with some gaps. Note that both are slower than
software due to lack of DMA to submit vertices.
- Mostly fix R128 and Radeon transform support, including supporting
bilinear filtering on R128. Subpixel offsets are still probably an
issue (reported by rendercheck), but I want to make 100% sure about my
understanding of the protocol before changing everybody, including fb.
- Add support for dst formats without alpha to R128 Composite.
- Remove the R128 Blend code, which has long outlived its usefulness.
(I kept it around for one reason: It could be useful for the
w/h > 1024 case with no mask and a non-src op. That seems pretty
infrequent and not worth the trouble).
Index: ati.h
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/ati.h,v
retrieving revision 1.11
retrieving revision 1.12
diff -u -d -r1.11 -r1.12
--- ati.h 22 Jul 2004 18:17:59 -0000 1.11
+++ ati.h 11 Sep 2004 09:28:19 -0000 1.12
@@ -319,6 +319,20 @@
#define getATIScreenInfo(kd) ((ATIScreenInfo *) ((kd)->screen->driver))
#define ATIScreenInfo(kd) ATIScreenInfo *atis = getATIScreenInfo(kd)
+typedef union { float f; CARD32 i; } fi_type;
+
+/* Surely there's a better way to go about this */
+static inline CARD32
+ATIFloatAsInt(float val)
+{
+ fi_type fi;
+
+ fi.f = val;
+ return fi.i;
+}
+
+#define GET_FLOAT_BITS(x) ATIFloatAsInt(x)
+
/* ati.c */
Bool
ATIMapReg(KdCardInfo *card, ATICardInfo *atic);
Index: ati_dma.h
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/ati_dma.h,v
retrieving revision 1.1
retrieving revision 1.2
diff -u -d -r1.1 -r1.2
--- ati_dma.h 17 May 2004 20:18:02 -0000 1.1
+++ ati_dma.h 11 Sep 2004 09:28:19 -0000 1.2
@@ -76,6 +76,8 @@
__head[__count++] = (x); \
} while (0)
+#define OUT_RING_F(x) OUT_RING(GET_FLOAT_BITS(x))
+
#define OUT_REG(reg, val) \
do { \
OUT_RING(DMA_PACKET0(reg, 1)); \
Index: ati_draw.c
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/ati_draw.c,v
retrieving revision 1.20
retrieving revision 1.21
diff -u -d -r1.20 -r1.21
--- ati_draw.c 19 Jul 2004 11:19:12 -0000 1.20
+++ ati_draw.c 11 Sep 2004 09:28:19 -0000 1.21
@@ -77,6 +77,10 @@
Bool is_24bpp;
CARD32 settings, color, src_pitch_offset, dst_pitch_offset;
+int sample_count;
+float sample_offsets_x[255];
+float sample_offsets_y[255];
+
#define DRAW_USING_PACKET3 0
void
@@ -719,6 +723,9 @@
atis->kaa.offscreenPitch = 32;
}
+ kaaInitTrapOffsets(8, sample_offsets_x, sample_offsets_y, 0.0, 0.0);
+ sample_count = (1 << 8) - 1;
+
if (!kaaDrawInit(pScreen, &atis->kaa))
return FALSE;
@@ -751,15 +758,6 @@
atis->kaa.PrepareComposite = NULL;
atis->kaa.Composite = NULL;
atis->kaa.DoneComposite = NULL;
- /* XXX: The R128 Blend code has several issues, according to
- * rendercheck. Source coordinates are sometimes wrong it appears, and
- * in some cases the blending results are wrong.
- */
- if (0 && !atic->is_radeon) {
- atis->kaa.PrepareBlend = R128PrepareBlend;
- atis->kaa.Blend = R128Blend;
- atis->kaa.DoneBlend = R128DoneBlend;
- }
/* We can't dispatch 3d commands in PIO mode. */
if (!atis->using_pio) {
@@ -768,11 +766,17 @@
atis->kaa.PrepareComposite = R128PrepareComposite;
atis->kaa.Composite = R128Composite;
atis->kaa.DoneComposite = R128DoneComposite;
+ /*atis->kaa.PrepareTrapezoids = R128PrepareTrapezoids;
+ atis->kaa.Trapezoids = R128Trapezoids;
+ atis->kaa.DoneTrapezoids = R128DoneTrapezoids;*/
} else if (atic->is_r100) {
atis->kaa.CheckComposite = R100CheckComposite;
atis->kaa.PrepareComposite = R100PrepareComposite;
atis->kaa.Composite = RadeonComposite;
atis->kaa.DoneComposite = RadeonDoneComposite;
+ /*atis->kaa.PrepareTrapezoids = RadeonPrepareTrapezoids;
+ atis->kaa.Trapezoids = RadeonTrapezoids;
+ atis->kaa.DoneTrapezoids = RadeonDoneTrapezoids;*/
} else if (0 && atic->is_r200) { /* XXX */
atis->kaa.CheckComposite = R200CheckComposite;
atis->kaa.PrepareComposite = R200PrepareComposite;
Index: ati_draw.h
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/ati_draw.h,v
retrieving revision 1.7
retrieving revision 1.8
diff -u -d -r1.7 -r1.8
--- ati_draw.h 19 Jul 2004 07:53:54 -0000 1.7
+++ ati_draw.h 11 Sep 2004 09:28:19 -0000 1.8
@@ -24,69 +24,44 @@
#ifndef _ATI_DRAW_H_
#define _ATI_DRAW_H_
-Bool
-ATIGetOffsetPitch(ATIScreenInfo *atis, int bpp, CARD32 *pitch_offset,
+Bool ATIGetOffsetPitch(ATIScreenInfo *atis, int bpp, CARD32 *pitch_offset,
int offset, int pitch);
+Bool ATIGetPixmapOffsetPitch(PixmapPtr pPix, CARD32 *pitch_offset);
-Bool
-ATIGetPixmapOffsetPitch(PixmapPtr pPix, CARD32 *pitch_offset);
-
-Bool
-R128PrepareBlend(int op, PicturePtr pSrcPicture, PicturePtr pDstPicture,
- PixmapPtr pSrc, PixmapPtr pDst);
-
-void
-R128Blend(int srcX, int srcY, int dstX, int dstY, int width, int height);
-
-void
-R128DoneBlend(void);
-
-Bool
-R128CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
+Bool R128CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
PicturePtr pDstPicture);
-
-Bool
-R128PrepareComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
- PicturePtr pDstPicture, PixmapPtr pSrc, PixmapPtr pMask, PixmapPtr pDst);
-
-void
-R128Composite(int srcX, int srcY, int maskX, int maskY, int dstX, int dstY,
+Bool R128PrepareComposite(int op, PicturePtr pSrcPicture,
+ PicturePtr pMaskPicture, PicturePtr pDstPicture, PixmapPtr pSrc,
+ PixmapPtr pMask, PixmapPtr pDst);
+void R128Composite(int srcX, int srcY, int maskX, int maskY, int dstX, int dstY,
int w, int h);
+void R128DoneComposite(void);
-void
-R128DoneComposite(void);
+Bool R128PrepareTrapezoids(PicturePtr pDstPicture, PixmapPtr pDst);
+void R128Trapezoids(KaaTrapezoid *traps, int ntraps);
+void R128DoneTrapezoids(void);
-Bool
-R100CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
+Bool R100CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
PicturePtr pDstPicture);
-
-Bool
-R100PrepareComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
- PicturePtr pDstPicture, PixmapPtr pSrc, PixmapPtr pMask, PixmapPtr pDst);
-
-Bool
-R200CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
+Bool R100PrepareComposite(int op, PicturePtr pSrcPicture,
+ PicturePtr pMaskPicture, PicturePtr pDstPicture, PixmapPtr pSrc,
+ PixmapPtr pMask, PixmapPtr pDst);
+Bool R200CheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
PicturePtr pDstPicture);
-
-Bool
-R200PrepareComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
- PicturePtr pDstPicture, PixmapPtr pSrc, PixmapPtr pMask, PixmapPtr pDst);
-
-void
-RadeonComposite(int srcX, int srcY, int maskX, int maskY, int dstX,
+Bool R200PrepareComposite(int op, PicturePtr pSrcPicture,
+ PicturePtr pMaskPicture, PicturePtr pDstPicture, PixmapPtr pSrc,
+ PixmapPtr pMask, PixmapPtr pDst);
+void RadeonComposite(int srcX, int srcY, int maskX, int maskY, int dstX,
int dstY, int w, int h);
+void RadeonDoneComposite(void);
-void
-RadeonDoneComposite(void);
-
-void
-RadeonSwitchTo2D(ATIScreenInfo *atis);
-
-void
-RadeonSwitchTo3D(ATIScreenInfo *atis);
+Bool RadeonPrepareTrapezoids(PicturePtr pDstPicture, PixmapPtr pDst);
+void RadeonTrapezoids(KaaTrapezoid *traps, int ntraps);
+void RadeonDoneTrapezoids(void);
-void
-ATIWaitIdle(ATIScreenInfo *atis);
+void RadeonSwitchTo2D(ATIScreenInfo *atis);
+void RadeonSwitchTo3D(ATIScreenInfo *atis);
+void ATIWaitIdle(ATIScreenInfo *atis);
#if 0
#define ATI_FALLBACK(x) \
Index: ati_reg.h
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/ati_reg.h,v
retrieving revision 1.7
retrieving revision 1.8
diff -u -d -r1.7 -r1.8
--- ati_reg.h 22 Jul 2004 18:17:59 -0000 1.7
+++ ati_reg.h 11 Sep 2004 09:28:19 -0000 1.8
@@ -327,7 +327,27 @@
#define ATI_REG_DP_BRUSH_FRGD_CLR 0x147c
#define ATI_REG_DST_WIDTH_HEIGHT 0x1598
#define ATI_REG_CLR_CMP_CNTL 0x15c0
-#define ATI_REG_AUX_SC_CNTL 0x1660
+
+#define R128_REG_AUX_SC_CNTL 0x1660
+# define R128_AUX1_SC_ENB (1 << 0)
+# define R128_AUX1_SC_MODE_SUB (1 << 1)
+# define R128_AUX2_SC_ENB (1 << 0)
+# define R128_AUX2_SC_MODE_SUB (1 << 1)
+# define R128_AUX3_SC_ENB (1 << 0)
+# define R128_AUX3_SC_MODE_SUB (1 << 1)
+
+#define R128_REG_AUX1_SC_LEFT 0x1664
+#define R128_REG_AUX1_SC_RIGHT 0x1668
+#define R128_REG_AUX1_SC_TOP 0x166c
+#define R128_REG_AUX1_SC_BOTTOM 0x1670
+#define R128_REG_AUX2_SC_LEFT 0x1674
+#define R128_REG_AUX2_SC_RIGHT 0x1678
+#define R128_REG_AUX2_SC_TOP 0x167c
+#define R128_REG_AUX2_SC_BOTTOM 0x1680
+#define R128_REG_AUX3_SC_LEFT 0x1684
+#define R128_REG_AUX3_SC_RIGHT 0x1688
+#define R128_REG_AUX3_SC_TOP 0x168c
+#define R128_REG_AUX3_SC_BOTTOM 0x1690
#define ATI_REG_DP_CNTL 0x16c0
# define ATI_DST_X_LEFT_TO_RIGHT (1 << 0)
Index: r128_composite.c
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/r128_composite.c,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -d -r1.4 -r1.5
--- r128_composite.c 19 Jul 2004 11:16:13 -0000 1.4
+++ r128_composite.c 11 Sep 2004 09:28:19 -0000 1.5
@@ -27,14 +27,13 @@
#include "ati_draw.h"
extern ATIScreenInfo *accel_atis;
+extern int sample_count;
+extern float sample_offsets_x[255];
+extern float sample_offsets_y[255];
extern CARD8 ATIBltRop[16];
-static int src_pitch;
-static int src_offset;
-static int src_bpp;
static int widths[2] = {1,1};
static int heights[2] = {1,1};
-static Bool is_repeat;
static Bool is_transform[2];
static PictTransform *transform[2];
@@ -78,12 +77,14 @@
{
switch (format) {
case PICT_a1r5g5b5:
+ case PICT_x1r5g5b5:
*type = R128_DATATYPE_ARGB1555;
return TRUE;
case PICT_r5g6b5:
*type = R128_DATATYPE_RGB565;
return TRUE;
case PICT_a8r8g8b8:
+ case PICT_x8r8g8b8:
*type = R128_DATATYPE_ARGB8888;
return TRUE;
default:
@@ -92,120 +93,6 @@
}
-Bool
-R128PrepareBlend(int op, PicturePtr pSrcPicture, PicturePtr pDstPicture,
- PixmapPtr pSrc, PixmapPtr pDst)
-{
- KdScreenPriv(pDst->drawable.pScreen);
- ATIScreenInfo(pScreenPriv);
- CARD32 dstDatatype, srcDatatype;
- RING_LOCALS;
- CARD32 xinc, yinc, dst_pitch_offset;
-
- accel_atis = atis;
-
- src_offset = (CARD8 *)pSrc->devPrivate.ptr -
- pScreenPriv->screen->memory_base;
- src_pitch = pSrc->devKind;
- src_bpp = pSrc->drawable.bitsPerPixel;
- is_repeat = pSrcPicture->repeat;
-
- if (op >= sizeof(R128BlendOp)/sizeof(R128BlendOp[0]))
- ATI_FALLBACK(("Unsupported op 0x%x\n", op));
- if (pSrcPicture->repeat && (pSrc->drawable.width != 1 ||
- pSrc->drawable.height != 1))
- ATI_FALLBACK(("repeat unsupported\n"));
- if (pSrcPicture->transform != NULL)
- ATI_FALLBACK(("transform unsupported\n"));
- if (!R128GetDatatypePict(pDstPicture->format, &dstDatatype))
- ATI_FALLBACK(("Unsupported dest format 0x%x\n",
- pDstPicture->format));
- if (!R128GetDatatypePict(pSrcPicture->format, &srcDatatype))
- ATI_FALLBACK(("Unsupported src format 0x%x\n",
- pSrcPicture->format));
- if (src_pitch % src_bpp != 0)
- ATI_FALLBACK(("Bad src pitch 0x%x\n", src_pitch));
- if (!ATIGetPixmapOffsetPitch(pDst, &dst_pitch_offset))
- return FALSE;
-
- if (is_repeat) {
- xinc = 0;
- yinc = 0;
- } else {
- xinc = 65536;
- yinc = 65536;
- }
-
- BEGIN_DMA(18);
- OUT_REG(ATI_REG_DST_PITCH_OFFSET, dst_pitch_offset);
- OUT_REG(ATI_REG_DP_GUI_MASTER_CNTL,
- ATI_GMC_DST_PITCH_OFFSET_CNTL |
- ATI_GMC_BRUSH_SOLID_COLOR |
- (dstDatatype << 8) |
- ATI_GMC_SRC_DATATYPE_COLOR |
- (ATIBltRop[GXcopy] << 16) |
- ATI_DP_SRC_SOURCE_MEMORY |
- R128_GMC_3D_FCN_EN |
- ATI_GMC_CLR_CMP_CNTL_DIS |
- R128_GMC_AUX_CLIP_DIS);
- OUT_REG(ATI_REG_DP_CNTL,
- ATI_DST_X_LEFT_TO_RIGHT | ATI_DST_Y_TOP_TO_BOTTOM );
- OUT_REG(R128_REG_SCALE_3D_CNTL,
- R128_SCALE_3D_SCALE |
- R128_SCALE_PIX_REPLICATE |
- R128BlendOp[op].blendctl |
- R128_TEX_MAP_ALPHA_IN_TEXTURE);
- OUT_REG(R128_REG_TEX_CNTL_C, R128_ALPHA_ENABLE | R128_TEX_CACHE_FLUSH);
- OUT_REG(R128_REG_SCALE_3D_DATATYPE, srcDatatype);
-
- /* R128_REG_SCALE_PITCH,
- * R128_REG_SCALE_X_INC,
- * R128_REG_SCALE_Y_INC,
- * R128_REG_SCALE_HACC
- * R128_REG_SCALE_VACC */
- OUT_RING(DMA_PACKET0(R128_REG_SCALE_PITCH, 5));
- OUT_RING(src_pitch / src_bpp);
- OUT_RING(xinc);
- OUT_RING(yinc);
- OUT_RING(0x0);
- OUT_RING(0x0);
- END_DMA();
-
- return TRUE;
-}
-
-void
-R128Blend(int srcX, int srcY, int dstX, int dstY, int width, int height)
-{
- ATIScreenInfo *atis = accel_atis;
- RING_LOCALS;
-
- if (is_repeat) {
- srcX = 0;
- srcY = 0;
- }
-
- BEGIN_DMA(6);
- /* R128_REG_SCALE_SRC_HEIGHT_WIDTH,
- * R128_REG_SCALE_OFFSET_0
- */
- OUT_RING(DMA_PACKET0(R128_REG_SCALE_SRC_HEIGHT_WIDTH, 2));
- OUT_RING((height << 16) | width);
- OUT_RING(src_offset + srcY * src_pitch + srcX * (src_bpp >> 3));
- /* R128_REG_SCALE_DST_X_Y
- * R128_REG_SCALE_DST_HEIGHT_WIDTH
- */
- OUT_RING(DMA_PACKET0(R128_REG_SCALE_DST_X_Y, 2));
- OUT_RING((dstX << 16) | dstY);
- OUT_RING((height << 16) | width);
- END_DMA();
-}
-
-void
-R128DoneBlend(void)
-{
-}
-
static Bool
R128CheckCompositeTexture(PicturePtr pPict)
{
@@ -302,6 +189,9 @@
*tex_cntl_c |= R128_MIP_MAP_DISABLE;
+ if (pPict->filter == PictFilterBilinear)
+ *tex_cntl_c |= R128_MIN_BLEND_LINEAR | R128_MAG_BLEND_LINEAR;
+
if (unit == 0)
shift = 0;
else {
@@ -344,6 +234,7 @@
ATIScreenInfo(pScreenPriv);
CARD32 txsize = 0, prim_tex_cntl_c, sec_tex_cntl_c = 0, dstDatatype;
CARD32 dst_pitch_offset, color_factor, in_color_factor, alpha_comb;
+ CARD32 blend_cntl;
int i;
RING_LOCALS;
@@ -365,6 +256,19 @@
if (!ATIGetPixmapOffsetPitch(pDst, &dst_pitch_offset))
return FALSE;
+ blend_cntl = R128BlendOp[op].blendctl;
+ if (PICT_FORMAT_A(pDstPicture->format) == 0 &&
+ R128BlendOp[op].dst_alpha) {
+ if ((blend_cntl & R128_SBLEND_MASK) ==
+ R128_SBLEND_DST_ALPHA)
+ blend_cntl = (blend_cntl & ~R128_SBLEND_MASK) |
+ R128_SBLEND_ONE;
+ else if ((blend_cntl & R128_SBLEND_MASK) ==
+ R128_SBLEND_INV_DST_ALPHA)
+ blend_cntl = (blend_cntl & ~R128_SBLEND_MASK) |
+ R128_SBLEND_ZERO;
+ }
+
BEGIN_DMA(12);
OUT_REG(R128_REG_SCALE_3D_CNTL,
R128_SCALE_3D_TEXMAP_SHADE |
@@ -388,7 +292,7 @@
R128_MISC_SCALE_3D_TEXMAP_SHADE |
R128_MISC_SCALE_PIX_REPLICATE |
R128_ALPHA_COMB_ADD_CLAMP |
- R128BlendOp[op].blendctl);
+ blend_cntl);
OUT_REG(R128_REG_TEX_CNTL_C,
R128_TEXMAP_ENABLE |
((pMask != NULL) ? R128_SEC_TEXMAP_ENABLE : 0) |
@@ -477,30 +381,18 @@
return TRUE;
}
-
-union intfloat {
- float f;
- CARD32 i;
-};
-
-struct blend_vertex {
- union intfloat x, y, z, w;
- union intfloat s0, t0;
- union intfloat s1, t1;
-};
-
#define VTX_RING_COUNT 8
-#define VTX_OUT(vtx) \
-do { \
- OUT_RING(vtx.x.i); \
- OUT_RING(vtx.y.i); \
- OUT_RING(vtx.z.i); \
- OUT_RING(vtx.w.i); \
- OUT_RING(vtx.s0.i); \
- OUT_RING(vtx.t0.i); \
- OUT_RING(vtx.s1.i); \
- OUT_RING(vtx.t1.i); \
+#define VTX_OUT(_dstX, _dstY, _srcX, _srcY, _maskX, _maskY) \
+do { \
+ OUT_RING_F((_dstX)); \
+ OUT_RING_F(((float)(_dstY)) + .125); \
+ OUT_RING_F(0.0); \
+ OUT_RING_F(1.0); \
+ OUT_RING_F((((float)(_srcX)) + 0.5) / (widths[0])); \
+ OUT_RING_F((((float)(_srcY)) + 0.5) / (heights[0])); \
+ OUT_RING_F((((float)(_maskX)) + 0.5) / (widths[1])); \
+ OUT_RING_F((((float)(_maskY)) + 0.5) / (heights[1])); \
} while (0)
void
@@ -508,95 +400,44 @@
int w, int h)
{
ATIScreenInfo *atis = accel_atis;
- struct blend_vertex vtx[4];
- int i;
int srcXend, srcYend, maskXend, maskYend;
+ PictVector v;
RING_LOCALS;
/*ErrorF("R128Composite (%d,%d) (%d,%d) (%d,%d) (%d,%d)\n",
srcX, srcY, maskX, maskY,dstX, dstY, w, h);*/
+ srcXend = srcX + w;
+ srcYend = srcY + h;
+ maskXend = maskX + w;
+ maskYend = maskY + h;
if (is_transform[0]) {
- PictVector v;
-
v.vector[0] = IntToxFixed(srcX);
v.vector[1] = IntToxFixed(srcY);
- v.vector[3] = xFixed1;
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[0], &v);
srcX = xFixedToInt(v.vector[0]);
srcY = xFixedToInt(v.vector[1]);
- v.vector[0] = IntToxFixed(srcX + w);
- v.vector[1] = IntToxFixed(srcY + h);
- v.vector[3] = xFixed1;
+ v.vector[0] = IntToxFixed(srcXend);
+ v.vector[1] = IntToxFixed(srcYend);
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[0], &v);
srcXend = xFixedToInt(v.vector[0]);
srcYend = xFixedToInt(v.vector[1]);
- } else {
- srcXend = srcX + w;
- srcYend = srcY + h;
}
if (is_transform[1]) {
- PictVector v;
-
v.vector[0] = IntToxFixed(maskX);
v.vector[1] = IntToxFixed(maskY);
- v.vector[3] = xFixed1;
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[1], &v);
maskX = xFixedToInt(v.vector[0]);
maskY = xFixedToInt(v.vector[1]);
- v.vector[0] = IntToxFixed(maskX + w);
- v.vector[1] = IntToxFixed(maskY + h);
- v.vector[3] = xFixed1;
+ v.vector[0] = IntToxFixed(maskXend);
+ v.vector[1] = IntToxFixed(maskYend);
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[1], &v);
maskXend = xFixedToInt(v.vector[0]);
maskYend = xFixedToInt(v.vector[1]);
- } else {
- maskXend = maskX + w;
- maskYend = maskY + h;
- }
- vtx[0].x.f = dstX;
- vtx[0].y.f = dstY;
- vtx[0].z.f = 0.0;
- vtx[0].w.f = 1.0;
- vtx[0].s0.f = srcX;
- vtx[0].t0.f = srcY;
- vtx[0].s1.f = maskX;
- vtx[0].t1.f = maskY;
-
- vtx[1].x.f = dstX;
- vtx[1].y.f = dstY + h;
- vtx[1].z.f = 0.0;
- vtx[1].w.f = 1.0;
- vtx[1].s0.f = srcX;
- vtx[1].t0.f = srcYend;
- vtx[1].s1.f = maskX;
- vtx[1].t1.f = maskYend;
-
- vtx[2].x.f = dstX + w;
- vtx[2].y.f = dstY + h;
- vtx[2].z.f = 0.0;
- vtx[2].w.f = 1.0;
- vtx[2].s0.f = srcXend;
- vtx[2].t0.f = srcYend;
- vtx[2].s1.f = maskXend;
- vtx[2].t1.f = maskYend;
-
- vtx[3].x.f = dstX + w;
- vtx[3].y.f = dstY;
- vtx[3].z.f = 0.0;
- vtx[3].w.f = 1.0;
- vtx[3].s0.f = srcXend;
- vtx[3].t0.f = srcY;
- vtx[3].s1.f = maskXend;
- vtx[3].t1.f = maskY;
-
- for (i = 0; i < 4; i++) {
- vtx[i].x.f += 0.0;
- vtx[i].y.f += 0.125;
- vtx[i].s0.f /= widths[0];
- vtx[i].t0.f /= heights[0];
- vtx[i].s1.f /= widths[1];
- vtx[i].t1.f /= heights[1];
}
BEGIN_DMA(3 + 4 * VTX_RING_COUNT);
@@ -609,10 +450,10 @@
R128_CCE_VC_CNTL_PRIM_WALK_RING |
(4 << R128_CCE_VC_CNTL_NUM_SHIFT));
- VTX_OUT(vtx[0]);
- VTX_OUT(vtx[1]);
- VTX_OUT(vtx[2]);
- VTX_OUT(vtx[3]);
+ VTX_OUT(dstX, dstY, srcX, srcY, maskX, maskY);
+ VTX_OUT(dstX, dstY + h, srcX, srcYend, maskX, maskYend);
+ VTX_OUT(dstX + w, dstY + h, srcXend, srcYend, maskXend, maskYend);
+ VTX_OUT(dstX + w, dstY, srcXend, srcY, maskXend, maskY);
END_DMA();
}
@@ -621,3 +462,113 @@
R128DoneComposite(void)
{
}
+
+Bool
+R128PrepareTrapezoids(PicturePtr pDstPicture, PixmapPtr pDst)
+{
+ KdScreenPriv(pDst->drawable.pScreen);
+ ATIScreenInfo(pScreenPriv);
+ CARD32 dst_pitch_offset;
+ RING_LOCALS;
+
+ accel_atis = atis;
+
+ if (!ATIGetPixmapOffsetPitch(pDst, &dst_pitch_offset))
+ return FALSE;
+
+ BEGIN_DMA(18);
+ OUT_REG(R128_REG_SCALE_3D_CNTL,
+ R128_SCALE_3D_TEXMAP_SHADE |
+ R128_SCALE_PIX_REPLICATE |
+ R128_TEX_CACHE_SPLIT |
+ R128_TEX_CACHE_LINE_SIZE_4QW);
+ OUT_REG(ATI_REG_DST_PITCH_OFFSET, dst_pitch_offset);
+ OUT_REG(ATI_REG_DP_GUI_MASTER_CNTL,
+ ATI_GMC_DST_PITCH_OFFSET_CNTL |
+ ATI_GMC_BRUSH_SOLID_COLOR |
+ (R128_DATATYPE_RGB8 << 8) |
+ ATI_GMC_SRC_DATATYPE_COLOR |
+ (ATIBltRop[GXcopy] << 16) |
+ ATI_DP_SRC_SOURCE_MEMORY |
+ R128_GMC_3D_FCN_EN |
+ ATI_GMC_CLR_CMP_CNTL_DIS |
+ ATI_GMC_WR_MSK_DIS);
+ OUT_REG(R128_REG_MISC_3D_STATE_CNTL,
+ R128_MISC_SCALE_3D_TEXMAP_SHADE |
+ R128_MISC_SCALE_PIX_REPLICATE |
+ R128_ALPHA_COMB_ADD_CLAMP |
+ R128BlendOp[PictOpAdd].blendctl);
+ OUT_REG(R128_REG_TEX_CNTL_C,
+ R128_ALPHA_ENABLE);
+ OUT_REG(R128_REG_PC_GUI_CTLSTAT, R128_PC_FLUSH_GUI);
+
+ /* R128_REG_AUX_SC_CNTL,
+ * R128_REG_AUX1_SC_LEFT
+ * R128_REG_AUX1_SC_RIGHT
+ * R128_REG_AUX1_SC_TOP
+ * R128_REG_AUX1_SC_BOTTOM
+ */
+ OUT_RING(DMA_PACKET0(R128_REG_AUX_SC_CNTL, 5));
+ OUT_RING(R128_AUX1_SC_ENB);
+ OUT_RING(0);
+ OUT_RING(pDst->drawable.width);
+ OUT_RING(0);
+ OUT_RING(pDst->drawable.height);
+ END_DMA();
+
+ return TRUE;
+}
+
+#define TRAP_VERT_RING_COUNT 4
+
+#define TRAP_VERT(_x, _y) \
+do { \
+ OUT_RING_F((_x) + sample_x); \
+ OUT_RING_F((_y) + 0.125 + sample_y); \
+ OUT_RING_F(0.0); \
+ OUT_RING(0x01010101); \
+} while (0)
+
+void
+R128Trapezoids(KaaTrapezoid *traps, int ntraps)
+{
+ ATIScreenInfo *atis = accel_atis;
+ RING_LOCALS;
+
+ while (ntraps > 0) {
+ int i, sample, count, vertcount;
+
+ count = 0xffff / 4 / sample_count;
+ if (count > ntraps)
+ count = ntraps;
+ vertcount = count * sample_count * 4;
+
+ BEGIN_DMA(3 + vertcount * TRAP_VERT_RING_COUNT);
+ OUT_RING(DMA_PACKET3(ATI_CCE_PACKET3_3D_RNDR_GEN_PRIM,
+ 2 + vertcount * TRAP_VERT_RING_COUNT));
+ OUT_RING(R128_CCE_VC_FRMT_DIFFUSE_ARGB);
+ OUT_RING(R128_CCE_VC_CNTL_PRIM_TYPE_TRI_FAN |
+ R128_CCE_VC_CNTL_PRIM_WALK_RING |
+ (vertcount << R128_CCE_VC_CNTL_NUM_SHIFT));
+
+ for (i = 0; i < count; i++) {
+ for (sample = 0; sample < sample_count; sample++) {
+ float sample_x = sample_offsets_x[sample];
+ float sample_y = sample_offsets_y[sample];
+ TRAP_VERT(traps[i].tl, traps[i].ty);
+ TRAP_VERT(traps[i].bl, traps[i].by);
+ TRAP_VERT(traps[i].br, traps[i].by);
+ TRAP_VERT(traps[i].tr, traps[i].ty);
+ }
+ }
+ END_DMA();
+
+ ntraps -= count;
+ traps += count;
+ }
+}
+
+void
+R128DoneTrapezoids(void)
+{
+}
Index: radeon_composite.c
===================================================================
RCS file: /cvs/xserver/xserver/hw/kdrive/ati/radeon_composite.c,v
retrieving revision 1.10
retrieving revision 1.11
diff -u -d -r1.10 -r1.11
--- radeon_composite.c 19 Jul 2004 12:07:01 -0000 1.10
+++ radeon_composite.c 11 Sep 2004 09:28:19 -0000 1.11
@@ -32,6 +32,9 @@
#include "ati_draw.h"
extern ATIScreenInfo *accel_atis;
+extern int sample_count;
+extern float sample_offsets_x[255];
+extern float sample_offsets_y[255];
static Bool is_transform[2];
static PictTransform *transform[2];
@@ -587,14 +590,14 @@
#define VTX_DWORD_COUNT 6
-#define VTX_OUT(vtx) \
-do { \
- OUT_RING(vtx.x.i); \
- OUT_RING(vtx.y.i); \
- OUT_RING(vtx.s0.i); \
- OUT_RING(vtx.t0.i); \
- OUT_RING(vtx.s1.i); \
- OUT_RING(vtx.t1.i); \
+#define VTX_OUT(_dstX, _dstY, _srcX, _srcY, _maskX, _maskY) \
+do { \
+ OUT_RING_F(_dstX); \
+ OUT_RING_F(_dstY); \
+ OUT_RING_F(_srcX); \
+ OUT_RING_F(_srcY); \
+ OUT_RING_F(_maskX); \
+ OUT_RING_F(_maskY); \
} while (0)
void
@@ -603,80 +606,46 @@
{
ATIScreenInfo *atis = accel_atis;
ATICardInfo *atic = atis->atic;
- struct blend_vertex vtx[4];
int srcXend, srcYend, maskXend, maskYend;
RING_LOCALS;
+ PictVector v;
/*ErrorF("RadeonComposite (%d,%d) (%d,%d) (%d,%d) (%d,%d)\n",
srcX, srcY, maskX, maskY,dstX, dstY, w, h);*/
+ srcXend = srcX + w;
+ srcYend = srcY + h;
+ maskXend = maskX + w;
+ maskYend = maskY + h;
if (is_transform[0]) {
- PictVector v;
-
v.vector[0] = IntToxFixed(srcX);
v.vector[1] = IntToxFixed(srcY);
- v.vector[3] = xFixed1;
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[0], &v);
srcX = xFixedToInt(v.vector[0]);
srcY = xFixedToInt(v.vector[1]);
- v.vector[0] = IntToxFixed(srcX + w);
- v.vector[1] = IntToxFixed(srcY + h);
- v.vector[3] = xFixed1;
+ v.vector[0] = IntToxFixed(srcXend);
+ v.vector[1] = IntToxFixed(srcYend);
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[0], &v);
srcXend = xFixedToInt(v.vector[0]);
srcYend = xFixedToInt(v.vector[1]);
- } else {
- srcXend = srcX + w;
- srcYend = srcY + h;
}
if (is_transform[1]) {
- PictVector v;
-
v.vector[0] = IntToxFixed(maskX);
v.vector[1] = IntToxFixed(maskY);
- v.vector[3] = xFixed1;
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[1], &v);
maskX = xFixedToInt(v.vector[0]);
maskY = xFixedToInt(v.vector[1]);
- v.vector[0] = IntToxFixed(maskX + w);
- v.vector[1] = IntToxFixed(maskY + h);
- v.vector[3] = xFixed1;
+ v.vector[0] = IntToxFixed(maskXend);
+ v.vector[1] = IntToxFixed(maskYend);
+ v.vector[2] = xFixed1;
PictureTransformPoint(transform[1], &v);
maskXend = xFixedToInt(v.vector[0]);
maskYend = xFixedToInt(v.vector[1]);
- } else {
- maskXend = maskX + w;
- maskYend = maskY + h;
}
- vtx[0].x.f = dstX;
- vtx[0].y.f = dstY;
- vtx[0].s0.f = srcX;
- vtx[0].t0.f = srcY;
- vtx[0].s1.f = maskX;
- vtx[0].t1.f = maskY;
-
- vtx[1].x.f = dstX;
- vtx[1].y.f = dstY + h;
- vtx[1].s0.f = srcX;
- vtx[1].t0.f = srcYend;
- vtx[1].s1.f = maskX;
- vtx[1].t1.f = maskYend;
-
- vtx[2].x.f = dstX + w;
- vtx[2].y.f = dstY + h;
- vtx[2].s0.f = srcXend;
- vtx[2].t0.f = srcYend;
- vtx[2].s1.f = maskXend;
- vtx[2].t1.f = maskYend;
-
- vtx[3].x.f = dstX + w;
- vtx[3].y.f = dstY;
- vtx[3].s0.f = srcXend;
- vtx[3].t0.f = srcY;
- vtx[3].s1.f = maskXend;
- vtx[3].t1.f = maskY;
-
if (atic->is_r100) {
BEGIN_DMA(4 * VTX_DWORD_COUNT + 3);
OUT_RING(DMA_PACKET3(RADEON_CP_PACKET3_3D_DRAW_IMMD,
@@ -698,10 +667,10 @@
(4 << RADEON_CP_VC_CNTL_NUM_SHIFT));
}
- VTX_OUT(vtx[0]);
- VTX_OUT(vtx[1]);
- VTX_OUT(vtx[2]);
- VTX_OUT(vtx[3]);
+ VTX_OUT(dstX, dstY, srcX, srcY, maskX, maskY);
+ VTX_OUT(dstX, dstY + h, srcX, srcYend, maskX, maskYend);
+ VTX_OUT(dstX + w, dstY + h, srcXend, srcYend, maskXend, maskYend);
+ VTX_OUT(dstX + w, dstY, srcXend, srcY, maskXend, maskY);
END_DMA();
}
@@ -710,3 +679,129 @@
RadeonDoneComposite(void)
{
}
+
+Bool
+RadeonPrepareTrapezoids(PicturePtr pDstPicture, PixmapPtr pDst)
+{
+ KdScreenPriv(pDst->drawable.pScreen);
+ ATIScreenInfo(pScreenPriv);
+ CARD32 dst_offset, dst_pitch;
+ int pixel_shift;
+ RING_LOCALS;
+
+ pixel_shift = pDst->drawable.bitsPerPixel >> 4;
+
+ accel_atis = atis;
+
+ dst_offset = ((CARD8 *)pDst->devPrivate.ptr -
+ pScreenPriv->screen->memory_base);
+ dst_pitch = pDst->devKind;
+ if ((dst_offset & 0x0f) != 0)
+ ATI_FALLBACK(("Bad destination offset 0x%x\n", dst_offset));
+ if (((dst_pitch >> pixel_shift) & 0x7) != 0)
+ ATI_FALLBACK(("Bad destination pitch 0x%x\n", dst_pitch));
+
+ BEGIN_DMA(14);
+ OUT_REG(ATI_REG_WAIT_UNTIL,
+ RADEON_WAIT_HOST_IDLECLEAN | RADEON_WAIT_2D_IDLECLEAN);
+
+ /* RADEON_REG_PP_CNTL,
+ * RADEON_REG_RB3D_CNTL,
+ * RADEON_REG_RB3D_COLOROFFSET,
+ * RADEON_REG_RE_WIDTH_HEIGHT,
+ * RADEON_REG_RB3D_COLORPITCH
+ */
+ OUT_RING(DMA_PACKET0(RADEON_REG_PP_CNTL, 5));
+ OUT_RING(RADEON_TEX_BLEND_0_ENABLE);
+ OUT_RING(RADEON_COLOR_FORMAT_RGB8 | RADEON_ALPHA_BLEND_ENABLE);
+ OUT_RING(dst_offset);
+ OUT_RING(((pDst->drawable.height - 1) << 16) |
+ (pDst->drawable.width - 1));
+ OUT_RING(dst_pitch >> pixel_shift);
+
+ /* RADEON_REG_PP_TXCBLEND_0,
+ * RADEON_REG_PP_TXABLEND_0,
+ * RADEON_REG_PP_TFACTOR_0
+ */
+ OUT_RING(DMA_PACKET0(RADEON_REG_PP_TXCBLEND_0, 3));
+ OUT_RING(RADEON_BLEND_CTL_ADD | RADEON_CLAMP_TX |
+ RADEON_COLOR_ARG_C_TFACTOR_ALPHA);
+ OUT_RING(RADEON_BLEND_CTL_ADD | RADEON_CLAMP_TX |
+ RADEON_ALPHA_ARG_C_TFACTOR_ALPHA);
+ OUT_RING(0x01000000);
+
+ OUT_REG(RADEON_REG_RB3D_BLENDCNTL, RadeonBlendOp[PictOpAdd].blend_cntl);
+ END_DMA();
+
+ return TRUE;
+}
+
+#define TRAP_VERT_RING_COUNT 2
+
+#define TRAP_VERT(_x, _y) \
+do { \
+ OUT_RING_F((_x) + sample_x); \
+ OUT_RING_F((_y) + sample_y); \
+} while (0)
+
+void
+RadeonTrapezoids(KaaTrapezoid *traps, int ntraps)
+{
+ ATIScreenInfo *atis = accel_atis;
+ ATICardInfo *atic = atis->atic;
+ RING_LOCALS;
+
+ while (ntraps > 0) {
+ int i, sample, count, vertcount;
+
+ count = 0xffff / 4 / sample_count;
+ if (count > ntraps)
+ count = ntraps;
+ vertcount = count * sample_count * 4;
+
+ if (atic->is_r100) {
+ BEGIN_DMA(3 + vertcount * TRAP_VERT_RING_COUNT);
+ OUT_RING(DMA_PACKET3(RADEON_CP_PACKET3_3D_DRAW_IMMD,
+ 2 + vertcount * TRAP_VERT_RING_COUNT));
+ OUT_RING(RADEON_CP_VC_FRMT_XY);
+ OUT_RING(RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN |
+ RADEON_CP_VC_CNTL_PRIM_WALK_RING |
+ RADEON_CP_VC_CNTL_MAOS_ENABLE |
+ RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
+ (vertcount << RADEON_CP_VC_CNTL_NUM_SHIFT));
+ } else {
+ BEGIN_DMA(2 + vertcount * TRAP_VERT_RING_COUNT);
+ OUT_RING(DMA_PACKET3(R200_CP_PACKET3_3D_DRAW_IMMD_2,
+ 1 + vertcount * TRAP_VERT_RING_COUNT));
+ OUT_RING(RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN |
+ RADEON_CP_VC_CNTL_PRIM_WALK_RING |
+ (vertcount << RADEON_CP_VC_CNTL_NUM_SHIFT));
+ }
+
+ for (i = 0; i < count; i++) {
+ for (sample = 0; sample < sample_count; sample++) {
+ float sample_x = sample_offsets_x[sample];
+ float sample_y = sample_offsets_y[sample];
+ TRAP_VERT(traps[i].tl, traps[i].ty);
+ TRAP_VERT(traps[i].bl, traps[i].by);
+ TRAP_VERT(traps[i].br, traps[i].by);
+ TRAP_VERT(traps[i].tr, traps[i].ty);
+ }
+ }
+ END_DMA();
+
+ ntraps -= count;
+ traps += count;
+ }
+}
+
+void
+RadeonDoneTrapezoids(void)
+{
+ ATIScreenInfo *atis = accel_atis;
+ RING_LOCALS;
+
+ BEGIN_DMA(2);
+ OUT_REG(RADEON_REG_RE_WIDTH_HEIGHT, 0xffffffff);
+ END_DMA();
+}
More information about the xserver-commit
mailing list