aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/char/drm
diff options
context:
space:
mode:
authorDave Airlie <airlied@starflyer.(none)>2005-08-16 06:43:16 -0400
committerDave Airlie <airlied@linux.ie>2005-08-16 06:43:16 -0400
commit414ed537995617f4cbcab65e193f26a2b2dcfa5e (patch)
tree7bd1b77bfd80819f7253fe4cbdb0c25560d8924c /drivers/char/drm
parent282a16749ba63256bcdce2766817f46aaac4dc20 (diff)
drm: add initial r300 3D support.
This adds initial r300 3D support to the radeon DRM. From: Nicolai Haehnle, Vladimir Dergachev, and others. Signed-off-by: David Airlie <airlied@linux.ie>
Diffstat (limited to 'drivers/char/drm')
-rw-r--r--drivers/char/drm/Makefile2
-rw-r--r--drivers/char/drm/drm_pciids.h13
-rw-r--r--drivers/char/drm/r300_cmdbuf.c801
-rw-r--r--drivers/char/drm/r300_reg.h1412
-rw-r--r--drivers/char/drm/radeon_cp.c13
-rw-r--r--drivers/char/drm/radeon_drm.h46
-rw-r--r--drivers/char/drm/radeon_drv.h29
-rw-r--r--drivers/char/drm/radeon_state.c11
8 files changed, 2325 insertions, 2 deletions
diff --git a/drivers/char/drm/Makefile b/drivers/char/drm/Makefile
index 1945138cb8fb..e41060c76226 100644
--- a/drivers/char/drm/Makefile
+++ b/drivers/char/drm/Makefile
@@ -14,7 +14,7 @@ mga-objs := mga_drv.o mga_dma.o mga_state.o mga_warp.o mga_irq.o
14i810-objs := i810_drv.o i810_dma.o 14i810-objs := i810_drv.o i810_dma.o
15i830-objs := i830_drv.o i830_dma.o i830_irq.o 15i830-objs := i830_drv.o i830_dma.o i830_irq.o
16i915-objs := i915_drv.o i915_dma.o i915_irq.o i915_mem.o 16i915-objs := i915_drv.o i915_dma.o i915_irq.o i915_mem.o
17radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o 17radeon-objs := radeon_drv.o radeon_cp.o radeon_state.o radeon_mem.o radeon_irq.o r300_cmdbuf.o
18ffb-objs := ffb_drv.o ffb_context.o 18ffb-objs := ffb_drv.o ffb_context.o
19sis-objs := sis_drv.o sis_ds.o sis_mm.o 19sis-objs := sis_drv.o sis_ds.o sis_mm.o
20savage-objs := savage_drv.o savage_bci.o savage_state.o 20savage-objs := savage_drv.o savage_bci.o savage_state.o
diff --git a/drivers/char/drm/drm_pciids.h b/drivers/char/drm/drm_pciids.h
index 8e264f9c1a1e..1874c1fd1717 100644
--- a/drivers/char/drm/drm_pciids.h
+++ b/drivers/char/drm/drm_pciids.h
@@ -25,6 +25,8 @@
25 {0x1002, 0x4965, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \ 25 {0x1002, 0x4965, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
26 {0x1002, 0x4966, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \ 26 {0x1002, 0x4966, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
27 {0x1002, 0x4967, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \ 27 {0x1002, 0x4967, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250}, \
28 {0x1002, 0x4A49, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R420}, \
29 {0x1002, 0x4A4B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R420}, \
28 {0x1002, 0x4C57, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \ 30 {0x1002, 0x4C57, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \
29 {0x1002, 0x4C58, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \ 31 {0x1002, 0x4C58, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV200|CHIP_IS_MOBILITY}, \
30 {0x1002, 0x4C59, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV100|CHIP_IS_MOBILITY}, \ 32 {0x1002, 0x4C59, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV100|CHIP_IS_MOBILITY}, \
@@ -33,7 +35,17 @@
33 {0x1002, 0x4C65, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \ 35 {0x1002, 0x4C65, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
34 {0x1002, 0x4C66, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \ 36 {0x1002, 0x4C66, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
35 {0x1002, 0x4C67, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \ 37 {0x1002, 0x4C67, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R250|CHIP_IS_MOBILITY}, \
38 {0x1002, 0x4E44, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
39 {0x1002, 0x4E45, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
40 {0x1002, 0x4E46, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
41 {0x1002, 0x4E47, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R300}, \
42 {0x1002, 0x4E48, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
43 {0x1002, 0x4E49, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
44 {0x1002, 0x4E4A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
45 {0x1002, 0x4E4B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R350}, \
36 {0x1002, 0x4E50, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \ 46 {0x1002, 0x4E50, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
47 {0x1002, 0x4E51, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
48 {0x1002, 0x4E54, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350|CHIP_IS_MOBILITY}, \
37 {0x1002, 0x5144, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \ 49 {0x1002, 0x5144, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
38 {0x1002, 0x5145, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \ 50 {0x1002, 0x5145, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
39 {0x1002, 0x5146, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \ 51 {0x1002, 0x5146, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R100|CHIP_SINGLE_CRTC}, \
@@ -56,6 +68,7 @@
56 {0x1002, 0x516A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \ 68 {0x1002, 0x516A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
57 {0x1002, 0x516B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \ 69 {0x1002, 0x516B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
58 {0x1002, 0x516C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \ 70 {0x1002, 0x516C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_R200}, \
71 {0x1002, 0x5460, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RV350}, \
59 {0x1002, 0x5834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \ 72 {0x1002, 0x5834, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \
60 {0x1002, 0x5835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP|CHIP_IS_MOBILITY}, \ 73 {0x1002, 0x5835, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP|CHIP_IS_MOBILITY}, \
61 {0x1002, 0x5836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \ 74 {0x1002, 0x5836, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS300|CHIP_IS_IGP}, \
diff --git a/drivers/char/drm/r300_cmdbuf.c b/drivers/char/drm/r300_cmdbuf.c
new file mode 100644
index 000000000000..623f1f460cb5
--- /dev/null
+++ b/drivers/char/drm/r300_cmdbuf.c
@@ -0,0 +1,801 @@
1/* r300_cmdbuf.c -- Command buffer emission for R300 -*- linux-c -*-
2 *
3 * Copyright (C) The Weather Channel, Inc. 2002.
4 * Copyright (C) 2004 Nicolai Haehnle.
5 * All Rights Reserved.
6 *
7 * The Weather Channel (TM) funded Tungsten Graphics to develop the
8 * initial release of the Radeon 8500 driver under the XFree86 license.
9 * This notice must be preserved.
10 *
11 * Permission is hereby granted, free of charge, to any person obtaining a
12 * copy of this software and associated documentation files (the "Software"),
13 * to deal in the Software without restriction, including without limitation
14 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
15 * and/or sell copies of the Software, and to permit persons to whom the
16 * Software is furnished to do so, subject to the following conditions:
17 *
18 * The above copyright notice and this permission notice (including the next
19 * paragraph) shall be included in all copies or substantial portions of the
20 * Software.
21 *
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
25 * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
26 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
27 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
28 * DEALINGS IN THE SOFTWARE.
29 *
30 * Authors:
31 * Nicolai Haehnle <prefect_@gmx.net>
32 */
33
34#include "drmP.h"
35#include "drm.h"
36#include "radeon_drm.h"
37#include "radeon_drv.h"
38#include "r300_reg.h"
39
40
41#define R300_SIMULTANEOUS_CLIPRECTS 4
42
43/* Values for R300_RE_CLIPRECT_CNTL depending on the number of cliprects
44 */
45static const int r300_cliprect_cntl[4] = {
46 0xAAAA,
47 0xEEEE,
48 0xFEFE,
49 0xFFFE
50};
51
52
53/**
54 * Emit up to R300_SIMULTANEOUS_CLIPRECTS cliprects from the given command
55 * buffer, starting with index n.
56 */
57static int r300_emit_cliprects(drm_radeon_private_t* dev_priv,
58 drm_radeon_cmd_buffer_t* cmdbuf,
59 int n)
60{
61 drm_clip_rect_t box;
62 int nr;
63 int i;
64 RING_LOCALS;
65
66 nr = cmdbuf->nbox - n;
67 if (nr > R300_SIMULTANEOUS_CLIPRECTS)
68 nr = R300_SIMULTANEOUS_CLIPRECTS;
69
70 DRM_DEBUG("%i cliprects\n", nr);
71
72 if (nr) {
73 BEGIN_RING(6 + nr*2);
74 OUT_RING( CP_PACKET0( R300_RE_CLIPRECT_TL_0, nr*2 - 1 ) );
75
76 for(i = 0; i < nr; ++i) {
77 if (DRM_COPY_FROM_USER_UNCHECKED(&box, &cmdbuf->boxes[n+i], sizeof(box))) {
78 DRM_ERROR("copy cliprect faulted\n");
79 return DRM_ERR(EFAULT);
80 }
81
82 box.x1 = (box.x1 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
83 box.y1 = (box.y1 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
84 box.x2 = (box.x2 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
85 box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) & R300_CLIPRECT_MASK;
86
87 OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
88 (box.y1 << R300_CLIPRECT_Y_SHIFT));
89 OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
90 (box.y2 << R300_CLIPRECT_Y_SHIFT));
91 }
92
93 OUT_RING_REG( R300_RE_CLIPRECT_CNTL, r300_cliprect_cntl[nr-1] );
94
95 /* TODO/SECURITY: Force scissors to a safe value, otherwise the
96 * client might be able to trample over memory.
97 * The impact should be very limited, but I'd rather be safe than
98 * sorry.
99 */
100 OUT_RING( CP_PACKET0( R300_RE_SCISSORS_TL, 1 ) );
101 OUT_RING( 0 );
102 OUT_RING( R300_SCISSORS_X_MASK | R300_SCISSORS_Y_MASK );
103 ADVANCE_RING();
104 } else {
105 /* Why we allow zero cliprect rendering:
106 * There are some commands in a command buffer that must be submitted
107 * even when there are no cliprects, e.g. DMA buffer discard
108 * or state setting (though state setting could be avoided by
109 * simulating a loss of context).
110 *
111 * Now since the cmdbuf interface is so chaotic right now (and is
112 * bound to remain that way for a bit until things settle down),
113 * it is basically impossible to filter out the commands that are
114 * necessary and those that aren't.
115 *
116 * So I choose the safe way and don't do any filtering at all;
117 * instead, I simply set up the engine so that all rendering
118 * can't produce any fragments.
119 */
120 BEGIN_RING(2);
121 OUT_RING_REG( R300_RE_CLIPRECT_CNTL, 0 );
122 ADVANCE_RING();
123 }
124
125 return 0;
126}
127
128u8 r300_reg_flags[0x10000>>2];
129
130
131void r300_init_reg_flags(void)
132{
133 int i;
134 memset(r300_reg_flags, 0, 0x10000>>2);
135 #define ADD_RANGE_MARK(reg, count,mark) \
136 for(i=((reg)>>2);i<((reg)>>2)+(count);i++)\
137 r300_reg_flags[i]|=(mark);
138
139 #define MARK_SAFE 1
140 #define MARK_CHECK_OFFSET 2
141
142 #define ADD_RANGE(reg, count) ADD_RANGE_MARK(reg, count, MARK_SAFE)
143
144 /* these match cmducs() command in r300_driver/r300/r300_cmdbuf.c */
145 ADD_RANGE(R300_SE_VPORT_XSCALE, 6);
146 ADD_RANGE(0x2080, 1);
147 ADD_RANGE(R300_SE_VTE_CNTL, 2);
148 ADD_RANGE(0x2134, 2);
149 ADD_RANGE(0x2140, 1);
150 ADD_RANGE(R300_VAP_INPUT_CNTL_0, 2);
151 ADD_RANGE(0x21DC, 1);
152 ADD_RANGE(0x221C, 1);
153 ADD_RANGE(0x2220, 4);
154 ADD_RANGE(0x2288, 1);
155 ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
156 ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
157 ADD_RANGE(R300_GB_ENABLE, 1);
158 ADD_RANGE(R300_GB_MSPOS0, 5);
159 ADD_RANGE(R300_TX_ENABLE, 1);
160 ADD_RANGE(0x4200, 4);
161 ADD_RANGE(0x4214, 1);
162 ADD_RANGE(R300_RE_POINTSIZE, 1);
163 ADD_RANGE(0x4230, 3);
164 ADD_RANGE(R300_RE_LINE_CNT, 1);
165 ADD_RANGE(0x4238, 1);
166 ADD_RANGE(0x4260, 3);
167 ADD_RANGE(0x4274, 4);
168 ADD_RANGE(0x4288, 5);
169 ADD_RANGE(0x42A0, 1);
170 ADD_RANGE(R300_RE_ZBIAS_T_FACTOR, 4);
171 ADD_RANGE(0x42B4, 1);
172 ADD_RANGE(R300_RE_CULL_CNTL, 1);
173 ADD_RANGE(0x42C0, 2);
174 ADD_RANGE(R300_RS_CNTL_0, 2);
175 ADD_RANGE(R300_RS_INTERP_0, 8);
176 ADD_RANGE(R300_RS_ROUTE_0, 8);
177 ADD_RANGE(0x43A4, 2);
178 ADD_RANGE(0x43E8, 1);
179 ADD_RANGE(R300_PFS_CNTL_0, 3);
180 ADD_RANGE(R300_PFS_NODE_0, 4);
181 ADD_RANGE(R300_PFS_TEXI_0, 64);
182 ADD_RANGE(0x46A4, 5);
183 ADD_RANGE(R300_PFS_INSTR0_0, 64);
184 ADD_RANGE(R300_PFS_INSTR1_0, 64);
185 ADD_RANGE(R300_PFS_INSTR2_0, 64);
186 ADD_RANGE(R300_PFS_INSTR3_0, 64);
187 ADD_RANGE(0x4BC0, 1);
188 ADD_RANGE(0x4BC8, 3);
189 ADD_RANGE(R300_PP_ALPHA_TEST, 2);
190 ADD_RANGE(0x4BD8, 1);
191 ADD_RANGE(R300_PFS_PARAM_0_X, 64);
192 ADD_RANGE(0x4E00, 1);
193 ADD_RANGE(R300_RB3D_CBLEND, 2);
194 ADD_RANGE(R300_RB3D_COLORMASK, 1);
195 ADD_RANGE(0x4E10, 3);
196 ADD_RANGE_MARK(R300_RB3D_COLOROFFSET0, 1, MARK_CHECK_OFFSET); /* check offset */
197 ADD_RANGE(R300_RB3D_COLORPITCH0, 1);
198 ADD_RANGE(0x4E50, 9);
199 ADD_RANGE(0x4E88, 1);
200 ADD_RANGE(0x4EA0, 2);
201 ADD_RANGE(R300_RB3D_ZSTENCIL_CNTL_0, 3);
202 ADD_RANGE(0x4F10, 4);
203 ADD_RANGE_MARK(R300_RB3D_DEPTHOFFSET, 1, MARK_CHECK_OFFSET); /* check offset */
204 ADD_RANGE(R300_RB3D_DEPTHPITCH, 1);
205 ADD_RANGE(0x4F28, 1);
206 ADD_RANGE(0x4F30, 2);
207 ADD_RANGE(0x4F44, 1);
208 ADD_RANGE(0x4F54, 1);
209
210 ADD_RANGE(R300_TX_FILTER_0, 16);
211 ADD_RANGE(R300_TX_UNK1_0, 16);
212 ADD_RANGE(R300_TX_SIZE_0, 16);
213 ADD_RANGE(R300_TX_FORMAT_0, 16);
214 /* Texture offset is dangerous and needs more checking */
215 ADD_RANGE_MARK(R300_TX_OFFSET_0, 16, MARK_CHECK_OFFSET);
216 ADD_RANGE(R300_TX_UNK4_0, 16);
217 ADD_RANGE(R300_TX_BORDER_COLOR_0, 16);
218
219 /* Sporadic registers used as primitives are emitted */
220 ADD_RANGE(0x4f18, 1);
221 ADD_RANGE(R300_RB3D_DSTCACHE_CTLSTAT, 1);
222 ADD_RANGE(R300_VAP_INPUT_ROUTE_0_0, 8);
223 ADD_RANGE(R300_VAP_INPUT_ROUTE_1_0, 8);
224
225}
226
227static __inline__ int r300_check_range(unsigned reg, int count)
228{
229 int i;
230 if(reg & ~0xffff)return -1;
231 for(i=(reg>>2);i<(reg>>2)+count;i++)
232 if(r300_reg_flags[i]!=MARK_SAFE)return 1;
233 return 0;
234}
235
236 /* we expect offsets passed to the framebuffer to be either within video memory or
237 within AGP space */
238static __inline__ int r300_check_offset(drm_radeon_private_t* dev_priv, u32 offset)
239{
240 /* we realy want to check against end of video aperture
241 but this value is not being kept.
242 This code is correct for now (does the same thing as the
243 code that sets MC_FB_LOCATION) in radeon_cp.c */
244 if((offset>=dev_priv->fb_location) &&
245 (offset<dev_priv->gart_vm_start))return 0;
246 if((offset>=dev_priv->gart_vm_start) &&
247 (offset<dev_priv->gart_vm_start+dev_priv->gart_size))return 0;
248 return 1;
249}
250
251static __inline__ int r300_emit_carefully_checked_packet0(drm_radeon_private_t* dev_priv,
252 drm_radeon_cmd_buffer_t* cmdbuf,
253 drm_r300_cmd_header_t header)
254{
255 int reg;
256 int sz;
257 int i;
258 int values[64];
259 RING_LOCALS;
260
261 sz = header.packet0.count;
262 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
263
264 if((sz>64)||(sz<0)){
265 DRM_ERROR("Cannot emit more than 64 values at a time (reg=%04x sz=%d)\n", reg, sz);
266 return DRM_ERR(EINVAL);
267 }
268 for(i=0;i<sz;i++){
269 values[i]=((int __user*)cmdbuf->buf)[i];
270 switch(r300_reg_flags[(reg>>2)+i]){
271 case MARK_SAFE:
272 break;
273 case MARK_CHECK_OFFSET:
274 if(r300_check_offset(dev_priv, (u32)values[i])){
275 DRM_ERROR("Offset failed range check (reg=%04x sz=%d)\n", reg, sz);
276 return DRM_ERR(EINVAL);
277 }
278 break;
279 default:
280 DRM_ERROR("Register %04x failed check as flag=%02x\n", reg+i*4, r300_reg_flags[(reg>>2)+i]);
281 return DRM_ERR(EINVAL);
282 }
283 }
284
285 BEGIN_RING(1+sz);
286 OUT_RING( CP_PACKET0( reg, sz-1 ) );
287 OUT_RING_TABLE( values, sz );
288 ADVANCE_RING();
289
290 cmdbuf->buf += sz*4;
291 cmdbuf->bufsz -= sz*4;
292
293 return 0;
294}
295
296/**
297 * Emits a packet0 setting arbitrary registers.
298 * Called by r300_do_cp_cmdbuf.
299 *
300 * Note that checks are performed on contents and addresses of the registers
301 */
302static __inline__ int r300_emit_packet0(drm_radeon_private_t* dev_priv,
303 drm_radeon_cmd_buffer_t* cmdbuf,
304 drm_r300_cmd_header_t header)
305{
306 int reg;
307 int sz;
308 RING_LOCALS;
309
310 sz = header.packet0.count;
311 reg = (header.packet0.reghi << 8) | header.packet0.reglo;
312
313 if (!sz)
314 return 0;
315
316 if (sz*4 > cmdbuf->bufsz)
317 return DRM_ERR(EINVAL);
318
319 if (reg+sz*4 >= 0x10000){
320 DRM_ERROR("No such registers in hardware reg=%04x sz=%d\n", reg, sz);
321 return DRM_ERR(EINVAL);
322 }
323
324 if(r300_check_range(reg, sz)){
325 /* go and check everything */
326 return r300_emit_carefully_checked_packet0(dev_priv, cmdbuf, header);
327 }
328 /* the rest of the data is safe to emit, whatever the values the user passed */
329
330 BEGIN_RING(1+sz);
331 OUT_RING( CP_PACKET0( reg, sz-1 ) );
332 OUT_RING_TABLE( (int __user*)cmdbuf->buf, sz );
333 ADVANCE_RING();
334
335 cmdbuf->buf += sz*4;
336 cmdbuf->bufsz -= sz*4;
337
338 return 0;
339}
340
341
342/**
343 * Uploads user-supplied vertex program instructions or parameters onto
344 * the graphics card.
345 * Called by r300_do_cp_cmdbuf.
346 */
347static __inline__ int r300_emit_vpu(drm_radeon_private_t* dev_priv,
348 drm_radeon_cmd_buffer_t* cmdbuf,
349 drm_r300_cmd_header_t header)
350{
351 int sz;
352 int addr;
353 RING_LOCALS;
354
355 sz = header.vpu.count;
356 addr = (header.vpu.adrhi << 8) | header.vpu.adrlo;
357
358 if (!sz)
359 return 0;
360 if (sz*16 > cmdbuf->bufsz)
361 return DRM_ERR(EINVAL);
362
363 BEGIN_RING(5+sz*4);
364 /* Wait for VAP to come to senses.. */
365 /* there is no need to emit it multiple times, (only once before VAP is programmed,
366 but this optimization is for later */
367 OUT_RING_REG( R300_VAP_PVS_WAITIDLE, 0 );
368 OUT_RING_REG( R300_VAP_PVS_UPLOAD_ADDRESS, addr );
369 OUT_RING( CP_PACKET0_TABLE( R300_VAP_PVS_UPLOAD_DATA, sz*4 - 1 ) );
370 OUT_RING_TABLE( (int __user*)cmdbuf->buf, sz*4 );
371
372 ADVANCE_RING();
373
374 cmdbuf->buf += sz*16;
375 cmdbuf->bufsz -= sz*16;
376
377 return 0;
378}
379
380
381/**
382 * Emit a clear packet from userspace.
383 * Called by r300_emit_packet3.
384 */
385static __inline__ int r300_emit_clear(drm_radeon_private_t* dev_priv,
386 drm_radeon_cmd_buffer_t* cmdbuf)
387{
388 RING_LOCALS;
389
390 if (8*4 > cmdbuf->bufsz)
391 return DRM_ERR(EINVAL);
392
393 BEGIN_RING(10);
394 OUT_RING( CP_PACKET3( R200_3D_DRAW_IMMD_2, 8 ) );
395 OUT_RING( R300_PRIM_TYPE_POINT|R300_PRIM_WALK_RING|
396 (1<<R300_PRIM_NUM_VERTICES_SHIFT) );
397 OUT_RING_TABLE( (int __user*)cmdbuf->buf, 8 );
398 ADVANCE_RING();
399
400 cmdbuf->buf += 8*4;
401 cmdbuf->bufsz -= 8*4;
402
403 return 0;
404}
405
406static __inline__ int r300_emit_3d_load_vbpntr(drm_radeon_private_t* dev_priv,
407 drm_radeon_cmd_buffer_t* cmdbuf,
408 u32 header)
409{
410 int count, i,k;
411 #define MAX_ARRAY_PACKET 64
412 u32 payload[MAX_ARRAY_PACKET];
413 u32 narrays;
414 RING_LOCALS;
415
416 count=(header>>16) & 0x3fff;
417
418 if((count+1)>MAX_ARRAY_PACKET){
419 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n", count);
420 return DRM_ERR(EINVAL);
421 }
422 memset(payload, 0, MAX_ARRAY_PACKET*4);
423 memcpy(payload, cmdbuf->buf+4, (count+1)*4);
424
425 /* carefully check packet contents */
426
427 narrays=payload[0];
428 k=0;
429 i=1;
430 while((k<narrays) && (i<(count+1))){
431 i++; /* skip attribute field */
432 if(r300_check_offset(dev_priv, payload[i])){
433 DRM_ERROR("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", k, i);
434 return DRM_ERR(EINVAL);
435 }
436 k++;
437 i++;
438 if(k==narrays)break;
439 /* have one more to process, they come in pairs */
440 if(r300_check_offset(dev_priv, payload[i])){
441 DRM_ERROR("Offset failed range check (k=%d i=%d) while processing 3D_LOAD_VBPNTR packet.\n", k, i);
442 return DRM_ERR(EINVAL);
443 }
444 k++;
445 i++;
446 }
447 /* do the counts match what we expect ? */
448 if((k!=narrays) || (i!=(count+1))){
449 DRM_ERROR("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n", k, i, narrays, count+1);
450 return DRM_ERR(EINVAL);
451 }
452
453 /* all clear, output packet */
454
455 BEGIN_RING(count+2);
456 OUT_RING(header);
457 OUT_RING_TABLE(payload, count+1);
458 ADVANCE_RING();
459
460 cmdbuf->buf += (count+2)*4;
461 cmdbuf->bufsz -= (count+2)*4;
462
463 return 0;
464}
465
466static __inline__ int r300_emit_raw_packet3(drm_radeon_private_t* dev_priv,
467 drm_radeon_cmd_buffer_t* cmdbuf)
468{
469 u32 header;
470 int count;
471 RING_LOCALS;
472
473 if (4 > cmdbuf->bufsz)
474 return DRM_ERR(EINVAL);
475
476 /* Fixme !! This simply emits a packet without much checking.
477 We need to be smarter. */
478
479 /* obtain first word - actual packet3 header */
480 header = *(u32 __user*)cmdbuf->buf;
481
482 /* Is it packet 3 ? */
483 if( (header>>30)!=0x3 ) {
484 DRM_ERROR("Not a packet3 header (0x%08x)\n", header);
485 return DRM_ERR(EINVAL);
486 }
487
488 count=(header>>16) & 0x3fff;
489
490 /* Check again now that we know how much data to expect */
491 if ((count+2)*4 > cmdbuf->bufsz){
492 DRM_ERROR("Expected packet3 of length %d but have only %d bytes left\n",
493 (count+2)*4, cmdbuf->bufsz);
494 return DRM_ERR(EINVAL);
495 }
496
497 /* Is it a packet type we know about ? */
498 switch(header & 0xff00){
499 case RADEON_3D_LOAD_VBPNTR: /* load vertex array pointers */
500 return r300_emit_3d_load_vbpntr(dev_priv, cmdbuf, header);
501
502 case RADEON_CP_3D_DRAW_IMMD_2: /* triggers drawing using in-packet vertex data */
503 case RADEON_CP_3D_DRAW_VBUF_2: /* triggers drawing of vertex buffers setup elsewhere */
504 case RADEON_CP_3D_DRAW_INDX_2: /* triggers drawing using indices to vertex buffer */
505 case RADEON_CP_INDX_BUFFER: /* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
506 case RADEON_WAIT_FOR_IDLE:
507 case RADEON_CP_NOP:
508 /* these packets are safe */
509 break;
510 default:
511 DRM_ERROR("Unknown packet3 header (0x%08x)\n", header);
512 return DRM_ERR(EINVAL);
513 }
514
515
516 BEGIN_RING(count+2);
517 OUT_RING(header);
518 OUT_RING_TABLE( (int __user*)(cmdbuf->buf+4), count+1);
519 ADVANCE_RING();
520
521 cmdbuf->buf += (count+2)*4;
522 cmdbuf->bufsz -= (count+2)*4;
523
524 return 0;
525}
526
527
528/**
529 * Emit a rendering packet3 from userspace.
530 * Called by r300_do_cp_cmdbuf.
531 */
532static __inline__ int r300_emit_packet3(drm_radeon_private_t* dev_priv,
533 drm_radeon_cmd_buffer_t* cmdbuf,
534 drm_r300_cmd_header_t header)
535{
536 int n;
537 int ret;
538 char __user* orig_buf = cmdbuf->buf;
539 int orig_bufsz = cmdbuf->bufsz;
540
541 /* This is a do-while-loop so that we run the interior at least once,
542 * even if cmdbuf->nbox is 0. Compare r300_emit_cliprects for rationale.
543 */
544 n = 0;
545 do {
546 if (cmdbuf->nbox > R300_SIMULTANEOUS_CLIPRECTS) {
547 ret = r300_emit_cliprects(dev_priv, cmdbuf, n);
548 if (ret)
549 return ret;
550
551 cmdbuf->buf = orig_buf;
552 cmdbuf->bufsz = orig_bufsz;
553 }
554
555 switch(header.packet3.packet) {
556 case R300_CMD_PACKET3_CLEAR:
557 DRM_DEBUG("R300_CMD_PACKET3_CLEAR\n");
558 ret = r300_emit_clear(dev_priv, cmdbuf);
559 if (ret) {
560 DRM_ERROR("r300_emit_clear failed\n");
561 return ret;
562 }
563 break;
564
565 case R300_CMD_PACKET3_RAW:
566 DRM_DEBUG("R300_CMD_PACKET3_RAW\n");
567 ret = r300_emit_raw_packet3(dev_priv, cmdbuf);
568 if (ret) {
569 DRM_ERROR("r300_emit_raw_packet3 failed\n");
570 return ret;
571 }
572 break;
573
574 default:
575 DRM_ERROR("bad packet3 type %i at %p\n",
576 header.packet3.packet,
577 cmdbuf->buf - sizeof(header));
578 return DRM_ERR(EINVAL);
579 }
580
581 n += R300_SIMULTANEOUS_CLIPRECTS;
582 } while(n < cmdbuf->nbox);
583
584 return 0;
585}
586
587/* Some of the R300 chips seem to be extremely touchy about the two registers
588 * that are configured in r300_pacify.
589 * Among the worst offenders seems to be the R300 ND (0x4E44): When userspace
590 * sends a command buffer that contains only state setting commands and a
591 * vertex program/parameter upload sequence, this will eventually lead to a
592 * lockup, unless the sequence is bracketed by calls to r300_pacify.
593 * So we should take great care to *always* call r300_pacify before
594 * *anything* 3D related, and again afterwards. This is what the
595 * call bracket in r300_do_cp_cmdbuf is for.
596 */
597
598/**
599 * Emit the sequence to pacify R300.
600 */
601static __inline__ void r300_pacify(drm_radeon_private_t* dev_priv)
602{
603 RING_LOCALS;
604
605 BEGIN_RING(6);
606 OUT_RING( CP_PACKET0( R300_RB3D_DSTCACHE_CTLSTAT, 0 ) );
607 OUT_RING( 0xa );
608 OUT_RING( CP_PACKET0( 0x4f18, 0 ) );
609 OUT_RING( 0x3 );
610 OUT_RING( CP_PACKET3( RADEON_CP_NOP, 0 ) );
611 OUT_RING( 0x0 );
612 ADVANCE_RING();
613}
614
615
616/**
617 * Called by r300_do_cp_cmdbuf to update the internal buffer age and state.
618 * The actual age emit is done by r300_do_cp_cmdbuf, which is why you must
619 * be careful about how this function is called.
620 */
621static void r300_discard_buffer(drm_device_t * dev, drm_buf_t * buf)
622{
623 drm_radeon_private_t *dev_priv = dev->dev_private;
624 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
625
626 buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
627 buf->pending = 1;
628 buf->used = 0;
629}
630
631
632/**
633 * Parses and validates a user-supplied command buffer and emits appropriate
634 * commands on the DMA ring buffer.
635 * Called by the ioctl handler function radeon_cp_cmdbuf.
636 */
637int r300_do_cp_cmdbuf(drm_device_t* dev,
638 DRMFILE filp,
639 drm_file_t* filp_priv,
640 drm_radeon_cmd_buffer_t* cmdbuf)
641{
642 drm_radeon_private_t *dev_priv = dev->dev_private;
643 drm_device_dma_t *dma = dev->dma;
644 drm_buf_t *buf = NULL;
645 int emit_dispatch_age = 0;
646 int ret = 0;
647
648 DRM_DEBUG("\n");
649
650 /* See the comment above r300_emit_begin3d for why this call must be here,
651 * and what the cleanup gotos are for. */
652 r300_pacify(dev_priv);
653
654 if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
655 ret = r300_emit_cliprects(dev_priv, cmdbuf, 0);
656 if (ret)
657 goto cleanup;
658 }
659
660 while(cmdbuf->bufsz >= sizeof(drm_r300_cmd_header_t)) {
661 int idx;
662 drm_r300_cmd_header_t header;
663
664 header.u = *(unsigned int *)cmdbuf->buf;
665
666 cmdbuf->buf += sizeof(header);
667 cmdbuf->bufsz -= sizeof(header);
668
669 switch(header.header.cmd_type) {
670 case R300_CMD_PACKET0:
671 DRM_DEBUG("R300_CMD_PACKET0\n");
672 ret = r300_emit_packet0(dev_priv, cmdbuf, header);
673 if (ret) {
674 DRM_ERROR("r300_emit_packet0 failed\n");
675 goto cleanup;
676 }
677 break;
678
679 case R300_CMD_VPU:
680 DRM_DEBUG("R300_CMD_VPU\n");
681 ret = r300_emit_vpu(dev_priv, cmdbuf, header);
682 if (ret) {
683 DRM_ERROR("r300_emit_vpu failed\n");
684 goto cleanup;
685 }
686 break;
687
688 case R300_CMD_PACKET3:
689 DRM_DEBUG("R300_CMD_PACKET3\n");
690 ret = r300_emit_packet3(dev_priv, cmdbuf, header);
691 if (ret) {
692 DRM_ERROR("r300_emit_packet3 failed\n");
693 goto cleanup;
694 }
695 break;
696
697 case R300_CMD_END3D:
698 DRM_DEBUG("R300_CMD_END3D\n");
699 /* TODO:
700 Ideally userspace driver should not need to issue this call,
701 i.e. the drm driver should issue it automatically and prevent
702 lockups.
703
704 In practice, we do not understand why this call is needed and what
705 it does (except for some vague guesses that it has to do with cache
706 coherence) and so the user space driver does it.
707
708 Once we are sure which uses prevent lockups the code could be moved
709 into the kernel and the userspace driver will not
710 need to use this command.
711
712 Note that issuing this command does not hurt anything
713 except, possibly, performance */
714 r300_pacify(dev_priv);
715 break;
716
717 case R300_CMD_CP_DELAY:
718 /* simple enough, we can do it here */
719 DRM_DEBUG("R300_CMD_CP_DELAY\n");
720 {
721 int i;
722 RING_LOCALS;
723
724 BEGIN_RING(header.delay.count);
725 for(i=0;i<header.delay.count;i++)
726 OUT_RING(RADEON_CP_PACKET2);
727 ADVANCE_RING();
728 }
729 break;
730
731 case R300_CMD_DMA_DISCARD:
732 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
733 idx = header.dma.buf_idx;
734 if (idx < 0 || idx >= dma->buf_count) {
735 DRM_ERROR("buffer index %d (of %d max)\n",
736 idx, dma->buf_count - 1);
737 ret = DRM_ERR(EINVAL);
738 goto cleanup;
739 }
740
741 buf = dma->buflist[idx];
742 if (buf->filp != filp || buf->pending) {
743 DRM_ERROR("bad buffer %p %p %d\n",
744 buf->filp, filp, buf->pending);
745 ret = DRM_ERR(EINVAL);
746 goto cleanup;
747 }
748
749 emit_dispatch_age = 1;
750 r300_discard_buffer(dev, buf);
751 break;
752
753 case R300_CMD_WAIT:
754 /* simple enough, we can do it here */
755 DRM_DEBUG("R300_CMD_WAIT\n");
756 if(header.wait.flags==0)break; /* nothing to do */
757
758 {
759 RING_LOCALS;
760
761 BEGIN_RING(2);
762 OUT_RING( CP_PACKET0( RADEON_WAIT_UNTIL, 0 ) );
763 OUT_RING( (header.wait.flags & 0xf)<<14 );
764 ADVANCE_RING();
765 }
766 break;
767
768 default:
769 DRM_ERROR("bad cmd_type %i at %p\n",
770 header.header.cmd_type,
771 cmdbuf->buf - sizeof(header));
772 ret = DRM_ERR(EINVAL);
773 goto cleanup;
774 }
775 }
776
777 DRM_DEBUG("END\n");
778
779cleanup:
780 r300_pacify(dev_priv);
781
782 /* We emit the vertex buffer age here, outside the pacifier "brackets"
783 * for two reasons:
784 * (1) This may coalesce multiple age emissions into a single one and
785 * (2) more importantly, some chips lock up hard when scratch registers
786 * are written inside the pacifier bracket.
787 */
788 if (emit_dispatch_age) {
789 RING_LOCALS;
790
791 /* Emit the vertex buffer age */
792 BEGIN_RING(2);
793 RADEON_DISPATCH_AGE(dev_priv->sarea_priv->last_dispatch);
794 ADVANCE_RING();
795 }
796
797 COMMIT_RING();
798
799 return ret;
800}
801
diff --git a/drivers/char/drm/r300_reg.h b/drivers/char/drm/r300_reg.h
new file mode 100644
index 000000000000..c3e7ca3dbe3d
--- /dev/null
+++ b/drivers/char/drm/r300_reg.h
@@ -0,0 +1,1412 @@
1/**************************************************************************
2
3Copyright (C) 2004-2005 Nicolai Haehnle et al.
4
5Permission is hereby granted, free of charge, to any person obtaining a
6copy of this software and associated documentation files (the "Software"),
7to deal in the Software without restriction, including without limitation
8on the rights to use, copy, modify, merge, publish, distribute, sub
9license, and/or sell copies of the Software, and to permit persons to whom
10the Software is furnished to do so, subject to the following conditions:
11
12The above copyright notice and this permission notice (including the next
13paragraph) shall be included in all copies or substantial portions of the
14Software.
15
16THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22USE OR OTHER DEALINGS IN THE SOFTWARE.
23
24**************************************************************************/
25
26#ifndef _R300_REG_H
27#define _R300_REG_H
28
29#define R300_MC_INIT_MISC_LAT_TIMER 0x180
30# define R300_MC_MISC__MC_CPR_INIT_LAT_SHIFT 0
31# define R300_MC_MISC__MC_VF_INIT_LAT_SHIFT 4
32# define R300_MC_MISC__MC_DISP0R_INIT_LAT_SHIFT 8
33# define R300_MC_MISC__MC_DISP1R_INIT_LAT_SHIFT 12
34# define R300_MC_MISC__MC_FIXED_INIT_LAT_SHIFT 16
35# define R300_MC_MISC__MC_E2R_INIT_LAT_SHIFT 20
36# define R300_MC_MISC__MC_SAME_PAGE_PRIO_SHIFT 24
37# define R300_MC_MISC__MC_GLOBW_INIT_LAT_SHIFT 28
38
39
40#define R300_MC_INIT_GFX_LAT_TIMER 0x154
41# define R300_MC_MISC__MC_G3D0R_INIT_LAT_SHIFT 0
42# define R300_MC_MISC__MC_G3D1R_INIT_LAT_SHIFT 4
43# define R300_MC_MISC__MC_G3D2R_INIT_LAT_SHIFT 8
44# define R300_MC_MISC__MC_G3D3R_INIT_LAT_SHIFT 12
45# define R300_MC_MISC__MC_TX0R_INIT_LAT_SHIFT 16
46# define R300_MC_MISC__MC_TX1R_INIT_LAT_SHIFT 20
47# define R300_MC_MISC__MC_GLOBR_INIT_LAT_SHIFT 24
48# define R300_MC_MISC__MC_GLOBW_FULL_LAT_SHIFT 28
49
50/*
51This file contains registers and constants for the R300. They have been
52found mostly by examining command buffers captured using glxtest, as well
53as by extrapolating some known registers and constants from the R200.
54
55I am fairly certain that they are correct unless stated otherwise in comments.
56*/
57
58#define R300_SE_VPORT_XSCALE 0x1D98
59#define R300_SE_VPORT_XOFFSET 0x1D9C
60#define R300_SE_VPORT_YSCALE 0x1DA0
61#define R300_SE_VPORT_YOFFSET 0x1DA4
62#define R300_SE_VPORT_ZSCALE 0x1DA8
63#define R300_SE_VPORT_ZOFFSET 0x1DAC
64
65
66/* This register is written directly and also starts data section in many 3d CP_PACKET3's */
67#define R300_VAP_VF_CNTL 0x2084
68
69# define R300_VAP_VF_CNTL__PRIM_TYPE__SHIFT 0
70# define R300_VAP_VF_CNTL__PRIM_NONE (0<<0)
71# define R300_VAP_VF_CNTL__PRIM_POINTS (1<<0)
72# define R300_VAP_VF_CNTL__PRIM_LINES (2<<0)
73# define R300_VAP_VF_CNTL__PRIM_LINE_STRIP (3<<0)
74# define R300_VAP_VF_CNTL__PRIM_TRIANGLES (4<<0)
75# define R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN (5<<0)
76# define R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP (6<<0)
77# define R300_VAP_VF_CNTL__PRIM_LINE_LOOP (12<<0)
78# define R300_VAP_VF_CNTL__PRIM_QUADS (13<<0)
79# define R300_VAP_VF_CNTL__PRIM_QUAD_STRIP (14<<0)
80# define R300_VAP_VF_CNTL__PRIM_POLYGON (15<<0)
81
82# define R300_VAP_VF_CNTL__PRIM_WALK__SHIFT 4
83 /* State based - direct writes to registers trigger vertex generation */
84# define R300_VAP_VF_CNTL__PRIM_WALK_STATE_BASED (0<<4)
85# define R300_VAP_VF_CNTL__PRIM_WALK_INDICES (1<<4)
86# define R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST (2<<4)
87# define R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED (3<<4)
88
89 /* I don't think I saw these three used.. */
90# define R300_VAP_VF_CNTL__COLOR_ORDER__SHIFT 6
91# define R300_VAP_VF_CNTL__TCL_OUTPUT_CTL_ENA__SHIFT 9
92# define R300_VAP_VF_CNTL__PROG_STREAM_ENA__SHIFT 10
93
94 /* index size - when not set the indices are assumed to be 16 bit */
95# define R300_VAP_VF_CNTL__INDEX_SIZE_32bit (1<<11)
96 /* number of vertices */
97# define R300_VAP_VF_CNTL__NUM_VERTICES__SHIFT 16
98
99/* BEGIN: Wild guesses */
100#define R300_VAP_OUTPUT_VTX_FMT_0 0x2090
101# define R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT (1<<0)
102# define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT (1<<1)
103# define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT (1<<2) /* GUESS */
104# define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT (1<<3) /* GUESS */
105# define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT (1<<4) /* GUESS */
106# define R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT (1<<16) /* GUESS */
107
108#define R300_VAP_OUTPUT_VTX_FMT_1 0x2094
109# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_0_COMP_CNT_SHIFT 0
110# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_1_COMP_CNT_SHIFT 3
111# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_2_COMP_CNT_SHIFT 6
112# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_3_COMP_CNT_SHIFT 9
113# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_4_COMP_CNT_SHIFT 12
114# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_5_COMP_CNT_SHIFT 15
115# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_6_COMP_CNT_SHIFT 18
116# define R300_VAP_OUTPUT_VTX_FMT_1__TEX_7_COMP_CNT_SHIFT 21
117/* END */
118
119#define R300_SE_VTE_CNTL 0x20b0
120# define R300_VPORT_X_SCALE_ENA 0x00000001
121# define R300_VPORT_X_OFFSET_ENA 0x00000002
122# define R300_VPORT_Y_SCALE_ENA 0x00000004
123# define R300_VPORT_Y_OFFSET_ENA 0x00000008
124# define R300_VPORT_Z_SCALE_ENA 0x00000010
125# define R300_VPORT_Z_OFFSET_ENA 0x00000020
126# define R300_VTX_XY_FMT 0x00000100
127# define R300_VTX_Z_FMT 0x00000200
128# define R300_VTX_W0_FMT 0x00000400
129# define R300_VTX_W0_NORMALIZE 0x00000800
130# define R300_VTX_ST_DENORMALIZED 0x00001000
131
132/* BEGIN: Vertex data assembly - lots of uncertainties */
133/* gap */
134/* Where do we get our vertex data?
135//
136// Vertex data either comes either from immediate mode registers or from
137// vertex arrays.
138// There appears to be no mixed mode (though we can force the pitch of
139// vertex arrays to 0, effectively reusing the same element over and over
140// again).
141//
142// Immediate mode is controlled by the INPUT_CNTL registers. I am not sure
143// if these registers influence vertex array processing.
144//
145// Vertex arrays are controlled via the 3D_LOAD_VBPNTR packet3.
146//
147// In both cases, vertex attributes are then passed through INPUT_ROUTE.
148
149// Beginning with INPUT_ROUTE_0_0 is a list of WORDs that route vertex data
150// into the vertex processor's input registers.
151// The first word routes the first input, the second word the second, etc.
152// The corresponding input is routed into the register with the given index.
153// The list is ended by a word with INPUT_ROUTE_END set.
154//
155// Always set COMPONENTS_4 in immediate mode. */
156
157#define R300_VAP_INPUT_ROUTE_0_0 0x2150
158# define R300_INPUT_ROUTE_COMPONENTS_1 (0 << 0)
159# define R300_INPUT_ROUTE_COMPONENTS_2 (1 << 0)
160# define R300_INPUT_ROUTE_COMPONENTS_3 (2 << 0)
161# define R300_INPUT_ROUTE_COMPONENTS_4 (3 << 0)
162# define R300_INPUT_ROUTE_COMPONENTS_RGBA (4 << 0) /* GUESS */
163# define R300_VAP_INPUT_ROUTE_IDX_SHIFT 8
164# define R300_VAP_INPUT_ROUTE_IDX_MASK (31 << 8) /* GUESS */
165# define R300_VAP_INPUT_ROUTE_END (1 << 13)
166# define R300_INPUT_ROUTE_IMMEDIATE_MODE (0 << 14) /* GUESS */
167# define R300_INPUT_ROUTE_FLOAT (1 << 14) /* GUESS */
168# define R300_INPUT_ROUTE_UNSIGNED_BYTE (2 << 14) /* GUESS */
169# define R300_INPUT_ROUTE_FLOAT_COLOR (3 << 14) /* GUESS */
170#define R300_VAP_INPUT_ROUTE_0_1 0x2154
171#define R300_VAP_INPUT_ROUTE_0_2 0x2158
172#define R300_VAP_INPUT_ROUTE_0_3 0x215C
173#define R300_VAP_INPUT_ROUTE_0_4 0x2160
174#define R300_VAP_INPUT_ROUTE_0_5 0x2164
175#define R300_VAP_INPUT_ROUTE_0_6 0x2168
176#define R300_VAP_INPUT_ROUTE_0_7 0x216C
177
178/* gap */
179/* Notes:
180// - always set up to produce at least two attributes:
181// if vertex program uses only position, fglrx will set normal, too
182// - INPUT_CNTL_0_COLOR and INPUT_CNTL_COLOR bits are always equal */
183#define R300_VAP_INPUT_CNTL_0 0x2180
184# define R300_INPUT_CNTL_0_COLOR 0x00000001
185#define R300_VAP_INPUT_CNTL_1 0x2184
186# define R300_INPUT_CNTL_POS 0x00000001
187# define R300_INPUT_CNTL_NORMAL 0x00000002
188# define R300_INPUT_CNTL_COLOR 0x00000004
189# define R300_INPUT_CNTL_TC0 0x00000400
190# define R300_INPUT_CNTL_TC1 0x00000800
191# define R300_INPUT_CNTL_TC2 0x00001000 /* GUESS */
192# define R300_INPUT_CNTL_TC3 0x00002000 /* GUESS */
193# define R300_INPUT_CNTL_TC4 0x00004000 /* GUESS */
194# define R300_INPUT_CNTL_TC5 0x00008000 /* GUESS */
195# define R300_INPUT_CNTL_TC6 0x00010000 /* GUESS */
196# define R300_INPUT_CNTL_TC7 0x00020000 /* GUESS */
197
198/* gap */
199/* Words parallel to INPUT_ROUTE_0; All words that are active in INPUT_ROUTE_0
200// are set to a swizzling bit pattern, other words are 0.
201//
202// In immediate mode, the pattern is always set to xyzw. In vertex array
203// mode, the swizzling pattern is e.g. used to set zw components in texture
204// coordinates with only tweo components. */
205#define R300_VAP_INPUT_ROUTE_1_0 0x21E0
206# define R300_INPUT_ROUTE_SELECT_X 0
207# define R300_INPUT_ROUTE_SELECT_Y 1
208# define R300_INPUT_ROUTE_SELECT_Z 2
209# define R300_INPUT_ROUTE_SELECT_W 3
210# define R300_INPUT_ROUTE_SELECT_ZERO 4
211# define R300_INPUT_ROUTE_SELECT_ONE 5
212# define R300_INPUT_ROUTE_SELECT_MASK 7
213# define R300_INPUT_ROUTE_X_SHIFT 0
214# define R300_INPUT_ROUTE_Y_SHIFT 3
215# define R300_INPUT_ROUTE_Z_SHIFT 6
216# define R300_INPUT_ROUTE_W_SHIFT 9
217# define R300_INPUT_ROUTE_ENABLE (15 << 12)
218#define R300_VAP_INPUT_ROUTE_1_1 0x21E4
219#define R300_VAP_INPUT_ROUTE_1_2 0x21E8
220#define R300_VAP_INPUT_ROUTE_1_3 0x21EC
221#define R300_VAP_INPUT_ROUTE_1_4 0x21F0
222#define R300_VAP_INPUT_ROUTE_1_5 0x21F4
223#define R300_VAP_INPUT_ROUTE_1_6 0x21F8
224#define R300_VAP_INPUT_ROUTE_1_7 0x21FC
225
226/* END */
227
228/* gap */
229/* BEGIN: Upload vertex program and data
230// The programmable vertex shader unit has a memory bank of unknown size
231// that can be written to in 16 byte units by writing the address into
232// UPLOAD_ADDRESS, followed by data in UPLOAD_DATA (multiples of 4 DWORDs).
233//
234// Pointers into the memory bank are always in multiples of 16 bytes.
235//
236// The memory bank is divided into areas with fixed meaning.
237//
238// Starting at address UPLOAD_PROGRAM: Vertex program instructions.
239// Native limits reported by drivers from ATI suggest size 256 (i.e. 4KB),
240// whereas the difference between known addresses suggests size 512.
241//
242// Starting at address UPLOAD_PARAMETERS: Vertex program parameters.
243// Native reported limits and the VPI layout suggest size 256, whereas
244// difference between known addresses suggests size 512.
245//
246// At address UPLOAD_POINTSIZE is a vector (0, 0, ps, 0), where ps is the
247// floating point pointsize. The exact purpose of this state is uncertain,
248// as there is also the R300_RE_POINTSIZE register.
249//
250// Multiple vertex programs and parameter sets can be loaded at once,
251// which could explain the size discrepancy. */
252#define R300_VAP_PVS_UPLOAD_ADDRESS 0x2200
253# define R300_PVS_UPLOAD_PROGRAM 0x00000000
254# define R300_PVS_UPLOAD_PARAMETERS 0x00000200
255# define R300_PVS_UPLOAD_POINTSIZE 0x00000406
256/* gap */
257#define R300_VAP_PVS_UPLOAD_DATA 0x2208
258/* END */
259
260/* gap */
261/* I do not know the purpose of this register. However, I do know that
262// it is set to 221C_CLEAR for clear operations and to 221C_NORMAL
263// for normal rendering. */
264#define R300_VAP_UNKNOWN_221C 0x221C
265# define R300_221C_NORMAL 0x00000000
266# define R300_221C_CLEAR 0x0001C000
267
268/* gap */
269/* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between
270// rendering commands and overwriting vertex program parameters.
271// Therefore, I suspect writing zero to 0x2284 synchronizes the engine and
272// avoids bugs caused by still running shaders reading bad data from memory. */
273#define R300_VAP_PVS_WAITIDLE 0x2284 /* GUESS */
274
275/* Absolutely no clue what this register is about. */
276#define R300_VAP_UNKNOWN_2288 0x2288
277# define R300_2288_R300 0x00750000 /* -- nh */
278# define R300_2288_RV350 0x0000FFFF /* -- Vladimir */
279
280/* gap */
281/* Addresses are relative to the vertex program instruction area of the
282// memory bank. PROGRAM_END points to the last instruction of the active
283// program
284//
285// The meaning of the two UNKNOWN fields is obviously not known. However,
286// experiments so far have shown that both *must* point to an instruction
287// inside the vertex program, otherwise the GPU locks up.
288// fglrx usually sets CNTL_3_UNKNOWN to the end of the program and
289// CNTL_1_UNKNOWN points to instruction where last write to position takes place.
290// Most likely this is used to ignore rest of the program in cases where group of verts arent visible.
291// For some reason this "section" is sometimes accepted other instruction that have
292// no relationship with position calculations.
293*/
294#define R300_VAP_PVS_CNTL_1 0x22D0
295# define R300_PVS_CNTL_1_PROGRAM_START_SHIFT 0
296# define R300_PVS_CNTL_1_POS_END_SHIFT 10
297# define R300_PVS_CNTL_1_PROGRAM_END_SHIFT 20
298/* Addresses are relative the the vertex program parameters area. */
299#define R300_VAP_PVS_CNTL_2 0x22D4
300# define R300_PVS_CNTL_2_PARAM_OFFSET_SHIFT 0
301# define R300_PVS_CNTL_2_PARAM_COUNT_SHIFT 16
302#define R300_VAP_PVS_CNTL_3 0x22D8
303# define R300_PVS_CNTL_3_PROGRAM_UNKNOWN_SHIFT 10
304# define R300_PVS_CNTL_3_PROGRAM_UNKNOWN2_SHIFT 0
305
306/* The entire range from 0x2300 to 0x2AC inclusive seems to be used for
307// immediate vertices */
308#define R300_VAP_VTX_COLOR_R 0x2464
309#define R300_VAP_VTX_COLOR_G 0x2468
310#define R300_VAP_VTX_COLOR_B 0x246C
311#define R300_VAP_VTX_POS_0_X_1 0x2490 /* used for glVertex2*() */
312#define R300_VAP_VTX_POS_0_Y_1 0x2494
313#define R300_VAP_VTX_COLOR_PKD 0x249C /* RGBA */
314#define R300_VAP_VTX_POS_0_X_2 0x24A0 /* used for glVertex3*() */
315#define R300_VAP_VTX_POS_0_Y_2 0x24A4
316#define R300_VAP_VTX_POS_0_Z_2 0x24A8
317#define R300_VAP_VTX_END_OF_PKT 0x24AC /* write 0 to indicate end of packet? */
318
319/* gap */
320
321/* These are values from r300_reg/r300_reg.h - they are known to be correct
322 and are here so we can use one register file instead of several
323 - Vladimir */
324#define R300_GB_VAP_RASTER_VTX_FMT_0 0x4000
325# define R300_GB_VAP_RASTER_VTX_FMT_0__POS_PRESENT (1<<0)
326# define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_0_PRESENT (1<<1)
327# define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_1_PRESENT (1<<2)
328# define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_2_PRESENT (1<<3)
329# define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_3_PRESENT (1<<4)
330# define R300_GB_VAP_RASTER_VTX_FMT_0__COLOR_SPACE (0xf<<5)
331# define R300_GB_VAP_RASTER_VTX_FMT_0__PT_SIZE_PRESENT (0x1<<16)
332
333#define R300_GB_VAP_RASTER_VTX_FMT_1 0x4004
334 /* each of the following is 3 bits wide, specifies number
335 of components */
336# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_0_COMP_CNT_SHIFT 0
337# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_1_COMP_CNT_SHIFT 3
338# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_2_COMP_CNT_SHIFT 6
339# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_3_COMP_CNT_SHIFT 9
340# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_4_COMP_CNT_SHIFT 12
341# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_5_COMP_CNT_SHIFT 15
342# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_6_COMP_CNT_SHIFT 18
343# define R300_GB_VAP_RASTER_VTX_FMT_1__TEX_7_COMP_CNT_SHIFT 21
344
345/* UNK30 seems to enables point to quad transformation on textures
346 (or something closely related to that).
347 This bit is rather fatal at the time being due to lackings at pixel shader side */
348#define R300_GB_ENABLE 0x4008
349# define R300_GB_POINT_STUFF_ENABLE (1<<0)
350# define R300_GB_LINE_STUFF_ENABLE (1<<1)
351# define R300_GB_TRIANGLE_STUFF_ENABLE (1<<2)
352# define R300_GB_STENCIL_AUTO_ENABLE (1<<4)
353# define R300_GB_UNK30 (1<<30)
354 /* each of the following is 2 bits wide */
355#define R300_GB_TEX_REPLICATE 0
356#define R300_GB_TEX_ST 1
357#define R300_GB_TEX_STR 2
358# define R300_GB_TEX0_SOURCE_SHIFT 16
359# define R300_GB_TEX1_SOURCE_SHIFT 18
360# define R300_GB_TEX2_SOURCE_SHIFT 20
361# define R300_GB_TEX3_SOURCE_SHIFT 22
362# define R300_GB_TEX4_SOURCE_SHIFT 24
363# define R300_GB_TEX5_SOURCE_SHIFT 26
364# define R300_GB_TEX6_SOURCE_SHIFT 28
365# define R300_GB_TEX7_SOURCE_SHIFT 30
366
367/* MSPOS - positions for multisample antialiasing (?) */
368#define R300_GB_MSPOS0 0x4010
369 /* shifts - each of the fields is 4 bits */
370# define R300_GB_MSPOS0__MS_X0_SHIFT 0
371# define R300_GB_MSPOS0__MS_Y0_SHIFT 4
372# define R300_GB_MSPOS0__MS_X1_SHIFT 8
373# define R300_GB_MSPOS0__MS_Y1_SHIFT 12
374# define R300_GB_MSPOS0__MS_X2_SHIFT 16
375# define R300_GB_MSPOS0__MS_Y2_SHIFT 20
376# define R300_GB_MSPOS0__MSBD0_Y 24
377# define R300_GB_MSPOS0__MSBD0_X 28
378
379#define R300_GB_MSPOS1 0x4014
380# define R300_GB_MSPOS1__MS_X3_SHIFT 0
381# define R300_GB_MSPOS1__MS_Y3_SHIFT 4
382# define R300_GB_MSPOS1__MS_X4_SHIFT 8
383# define R300_GB_MSPOS1__MS_Y4_SHIFT 12
384# define R300_GB_MSPOS1__MS_X5_SHIFT 16
385# define R300_GB_MSPOS1__MS_Y5_SHIFT 20
386# define R300_GB_MSPOS1__MSBD1 24
387
388
389#define R300_GB_TILE_CONFIG 0x4018
390# define R300_GB_TILE_ENABLE (1<<0)
391# define R300_GB_TILE_PIPE_COUNT_RV300 0
392# define R300_GB_TILE_PIPE_COUNT_R300 (3<<1)
393# define R300_GB_TILE_PIPE_COUNT_R420 (7<<1)
394# define R300_GB_TILE_SIZE_8 0
395# define R300_GB_TILE_SIZE_16 (1<<4)
396# define R300_GB_TILE_SIZE_32 (2<<4)
397# define R300_GB_SUPER_SIZE_1 (0<<6)
398# define R300_GB_SUPER_SIZE_2 (1<<6)
399# define R300_GB_SUPER_SIZE_4 (2<<6)
400# define R300_GB_SUPER_SIZE_8 (3<<6)
401# define R300_GB_SUPER_SIZE_16 (4<<6)
402# define R300_GB_SUPER_SIZE_32 (5<<6)
403# define R300_GB_SUPER_SIZE_64 (6<<6)
404# define R300_GB_SUPER_SIZE_128 (7<<6)
405# define R300_GB_SUPER_X_SHIFT 9 /* 3 bits wide */
406# define R300_GB_SUPER_Y_SHIFT 12 /* 3 bits wide */
407# define R300_GB_SUPER_TILE_A 0
408# define R300_GB_SUPER_TILE_B (1<<15)
409# define R300_GB_SUBPIXEL_1_12 0
410# define R300_GB_SUBPIXEL_1_16 (1<<16)
411
412#define R300_GB_FIFO_SIZE 0x4024
413 /* each of the following is 2 bits wide */
414#define R300_GB_FIFO_SIZE_32 0
415#define R300_GB_FIFO_SIZE_64 1
416#define R300_GB_FIFO_SIZE_128 2
417#define R300_GB_FIFO_SIZE_256 3
418# define R300_SC_IFIFO_SIZE_SHIFT 0
419# define R300_SC_TZFIFO_SIZE_SHIFT 2
420# define R300_SC_BFIFO_SIZE_SHIFT 4
421
422# define R300_US_OFIFO_SIZE_SHIFT 12
423# define R300_US_WFIFO_SIZE_SHIFT 14
424 /* the following use the same constants as above, but meaning is
425 is times 2 (i.e. instead of 32 words it means 64 */
426# define R300_RS_TFIFO_SIZE_SHIFT 6
427# define R300_RS_CFIFO_SIZE_SHIFT 8
428# define R300_US_RAM_SIZE_SHIFT 10
429 /* watermarks, 3 bits wide */
430# define R300_RS_HIGHWATER_COL_SHIFT 16
431# define R300_RS_HIGHWATER_TEX_SHIFT 19
432# define R300_OFIFO_HIGHWATER_SHIFT 22 /* two bits only */
433# define R300_CUBE_FIFO_HIGHWATER_COL_SHIFT 24
434
435#define R300_GB_SELECT 0x401C
436# define R300_GB_FOG_SELECT_C0A 0
437# define R300_GB_FOG_SELECT_C1A 1
438# define R300_GB_FOG_SELECT_C2A 2
439# define R300_GB_FOG_SELECT_C3A 3
440# define R300_GB_FOG_SELECT_1_1_W 4
441# define R300_GB_FOG_SELECT_Z 5
442# define R300_GB_DEPTH_SELECT_Z 0
443# define R300_GB_DEPTH_SELECT_1_1_W (1<<3)
444# define R300_GB_W_SELECT_1_W 0
445# define R300_GB_W_SELECT_1 (1<<4)
446
447#define R300_GB_AA_CONFIG 0x4020
448# define R300_AA_ENABLE 0x01
449# define R300_AA_SUBSAMPLES_2 0
450# define R300_AA_SUBSAMPLES_3 (1<<1)
451# define R300_AA_SUBSAMPLES_4 (2<<1)
452# define R300_AA_SUBSAMPLES_6 (3<<1)
453
454/* END */
455
456/* gap */
457/* The upper enable bits are guessed, based on fglrx reported limits. */
458#define R300_TX_ENABLE 0x4104
459# define R300_TX_ENABLE_0 (1 << 0)
460# define R300_TX_ENABLE_1 (1 << 1)
461# define R300_TX_ENABLE_2 (1 << 2)
462# define R300_TX_ENABLE_3 (1 << 3)
463# define R300_TX_ENABLE_4 (1 << 4)
464# define R300_TX_ENABLE_5 (1 << 5)
465# define R300_TX_ENABLE_6 (1 << 6)
466# define R300_TX_ENABLE_7 (1 << 7)
467# define R300_TX_ENABLE_8 (1 << 8)
468# define R300_TX_ENABLE_9 (1 << 9)
469# define R300_TX_ENABLE_10 (1 << 10)
470# define R300_TX_ENABLE_11 (1 << 11)
471# define R300_TX_ENABLE_12 (1 << 12)
472# define R300_TX_ENABLE_13 (1 << 13)
473# define R300_TX_ENABLE_14 (1 << 14)
474# define R300_TX_ENABLE_15 (1 << 15)
475
476/* The pointsize is given in multiples of 6. The pointsize can be
477// enormous: Clear() renders a single point that fills the entire
478// framebuffer. */
479#define R300_RE_POINTSIZE 0x421C
480# define R300_POINTSIZE_Y_SHIFT 0
481# define R300_POINTSIZE_Y_MASK (0xFFFF << 0) /* GUESS */
482# define R300_POINTSIZE_X_SHIFT 16
483# define R300_POINTSIZE_X_MASK (0xFFFF << 16) /* GUESS */
484# define R300_POINTSIZE_MAX (R300_POINTSIZE_Y_MASK / 6)
485
486/* The line width is given in multiples of 6.
487 In default mode lines are classified as vertical lines.
488 HO: horizontal
489 VE: vertical or horizontal
490 HO & VE: no classification
491*/
492#define R300_RE_LINE_CNT 0x4234
493# define R300_LINESIZE_SHIFT 0
494# define R300_LINESIZE_MASK (0xFFFF << 0) /* GUESS */
495# define R300_LINESIZE_MAX (R300_LINESIZE_MASK / 6)
496# define R300_LINE_CNT_HO (1 << 16)
497# define R300_LINE_CNT_VE (1 << 17)
498
499/* Some sort of scale or clamp value for texcoordless textures. */
500#define R300_RE_UNK4238 0x4238
501
502#define R300_RE_SHADE_MODEL 0x4278
503# define R300_RE_SHADE_MODEL_SMOOTH 0x3aaaa
504# define R300_RE_SHADE_MODEL_FLAT 0x39595
505
506/* Dangerous */
507#define R300_RE_POLYGON_MODE 0x4288
508# define R300_PM_ENABLED (1 << 0)
509# define R300_PM_FRONT_POINT (0 << 0)
510# define R300_PM_BACK_POINT (0 << 0)
511# define R300_PM_FRONT_LINE (1 << 4)
512# define R300_PM_FRONT_FILL (1 << 5)
513# define R300_PM_BACK_LINE (1 << 7)
514# define R300_PM_BACK_FILL (1 << 8)
515
516/* Not sure why there are duplicate of factor and constant values.
517 My best guess so far is that there are seperate zbiases for test and write.
518 Ordering might be wrong.
519 Some of the tests indicate that fgl has a fallback implementation of zbias
520 via pixel shaders. */
521#define R300_RE_ZBIAS_T_FACTOR 0x42A4
522#define R300_RE_ZBIAS_T_CONSTANT 0x42A8
523#define R300_RE_ZBIAS_W_FACTOR 0x42AC
524#define R300_RE_ZBIAS_W_CONSTANT 0x42B0
525
526/* This register needs to be set to (1<<1) for RV350 to correctly
527 perform depth test (see --vb-triangles in r300_demo)
528 Don't know about other chips. - Vladimir
529 This is set to 3 when GL_POLYGON_OFFSET_FILL is on.
530 My guess is that there are two bits for each zbias primitive (FILL, LINE, POINT).
531 One to enable depth test and one for depth write.
532 Yet this doesnt explain why depth writes work ...
533 */
534#define R300_RE_OCCLUSION_CNTL 0x42B4
535# define R300_OCCLUSION_ON (1<<1)
536
537#define R300_RE_CULL_CNTL 0x42B8
538# define R300_CULL_FRONT (1 << 0)
539# define R300_CULL_BACK (1 << 1)
540# define R300_FRONT_FACE_CCW (0 << 2)
541# define R300_FRONT_FACE_CW (1 << 2)
542
543
544/* BEGIN: Rasterization / Interpolators - many guesses
545// 0_UNKNOWN_18 has always been set except for clear operations.
546// TC_CNT is the number of incoming texture coordinate sets (i.e. it depends
547// on the vertex program, *not* the fragment program) */
548#define R300_RS_CNTL_0 0x4300
549# define R300_RS_CNTL_TC_CNT_SHIFT 2
550# define R300_RS_CNTL_TC_CNT_MASK (7 << 2)
551# define R300_RS_CNTL_CI_CNT_SHIFT 7 /* number of color interpolators used */
552# define R300_RS_CNTL_0_UNKNOWN_18 (1 << 18)
553/* Guess: RS_CNTL_1 holds the index of the highest used RS_ROUTE_n register. */
554#define R300_RS_CNTL_1 0x4304
555
556/* gap */
557/* Only used for texture coordinates.
558// Use the source field to route texture coordinate input from the vertex program
559// to the desired interpolator. Note that the source field is relative to the
560// outputs the vertex program *actually* writes. If a vertex program only writes
561// texcoord[1], this will be source index 0.
562// Set INTERP_USED on all interpolators that produce data used by the
563// fragment program. INTERP_USED looks like a swizzling mask, but
564// I haven't seen it used that way.
565//
566// Note: The _UNKNOWN constants are always set in their respective register.
567// I don't know if this is necessary. */
568#define R300_RS_INTERP_0 0x4310
569#define R300_RS_INTERP_1 0x4314
570# define R300_RS_INTERP_1_UNKNOWN 0x40
571#define R300_RS_INTERP_2 0x4318
572# define R300_RS_INTERP_2_UNKNOWN 0x80
573#define R300_RS_INTERP_3 0x431C
574# define R300_RS_INTERP_3_UNKNOWN 0xC0
575#define R300_RS_INTERP_4 0x4320
576#define R300_RS_INTERP_5 0x4324
577#define R300_RS_INTERP_6 0x4328
578#define R300_RS_INTERP_7 0x432C
579# define R300_RS_INTERP_SRC_SHIFT 2
580# define R300_RS_INTERP_SRC_MASK (7 << 2)
581# define R300_RS_INTERP_USED 0x00D10000
582
583/* These DWORDs control how vertex data is routed into fragment program
584// registers, after interpolators. */
585#define R300_RS_ROUTE_0 0x4330
586#define R300_RS_ROUTE_1 0x4334
587#define R300_RS_ROUTE_2 0x4338
588#define R300_RS_ROUTE_3 0x433C /* GUESS */
589#define R300_RS_ROUTE_4 0x4340 /* GUESS */
590#define R300_RS_ROUTE_5 0x4344 /* GUESS */
591#define R300_RS_ROUTE_6 0x4348 /* GUESS */
592#define R300_RS_ROUTE_7 0x434C /* GUESS */
593# define R300_RS_ROUTE_SOURCE_INTERP_0 0
594# define R300_RS_ROUTE_SOURCE_INTERP_1 1
595# define R300_RS_ROUTE_SOURCE_INTERP_2 2
596# define R300_RS_ROUTE_SOURCE_INTERP_3 3
597# define R300_RS_ROUTE_SOURCE_INTERP_4 4
598# define R300_RS_ROUTE_SOURCE_INTERP_5 5 /* GUESS */
599# define R300_RS_ROUTE_SOURCE_INTERP_6 6 /* GUESS */
600# define R300_RS_ROUTE_SOURCE_INTERP_7 7 /* GUESS */
601# define R300_RS_ROUTE_ENABLE (1 << 3) /* GUESS */
602# define R300_RS_ROUTE_DEST_SHIFT 6
603# define R300_RS_ROUTE_DEST_MASK (31 << 6) /* GUESS */
604
605/* Special handling for color: When the fragment program uses color,
606// the ROUTE_0_COLOR bit is set and ROUTE_0_COLOR_DEST contains the
607// color register index. */
608# define R300_RS_ROUTE_0_COLOR (1 << 14)
609# define R300_RS_ROUTE_0_COLOR_DEST_SHIFT 17
610# define R300_RS_ROUTE_0_COLOR_DEST_MASK (31 << 17) /* GUESS */
611/* As above, but for secondary color */
612# define R300_RS_ROUTE_1_COLOR1 (1 << 14)
613# define R300_RS_ROUTE_1_COLOR1_DEST_SHIFT 17
614# define R300_RS_ROUTE_1_COLOR1_DEST_MASK (31 << 17)
615# define R300_RS_ROUTE_1_UNKNOWN11 (1 << 11)
616/* END */
617
618/* BEGIN: Scissors and cliprects
619// There are four clipping rectangles. Their corner coordinates are inclusive.
620// Every pixel is assigned a number from 0 and 15 by setting bits 0-3 depending
621// on whether the pixel is inside cliprects 0-3, respectively. For example,
622// if a pixel is inside cliprects 0 and 1, but outside 2 and 3, it is assigned
623// the number 3 (binary 0011).
624// Iff the bit corresponding to the pixel's number in RE_CLIPRECT_CNTL is set,
625// the pixel is rasterized.
626//
627// In addition to this, there is a scissors rectangle. Only pixels inside the
628// scissors rectangle are drawn. (coordinates are inclusive)
629//
630// For some reason, the top-left corner of the framebuffer is at (1440, 1440)
631// for the purpose of clipping and scissors. */
632#define R300_RE_CLIPRECT_TL_0 0x43B0
633#define R300_RE_CLIPRECT_BR_0 0x43B4
634#define R300_RE_CLIPRECT_TL_1 0x43B8
635#define R300_RE_CLIPRECT_BR_1 0x43BC
636#define R300_RE_CLIPRECT_TL_2 0x43C0
637#define R300_RE_CLIPRECT_BR_2 0x43C4
638#define R300_RE_CLIPRECT_TL_3 0x43C8
639#define R300_RE_CLIPRECT_BR_3 0x43CC
640# define R300_CLIPRECT_OFFSET 1440
641# define R300_CLIPRECT_MASK 0x1FFF
642# define R300_CLIPRECT_X_SHIFT 0
643# define R300_CLIPRECT_X_MASK (0x1FFF << 0)
644# define R300_CLIPRECT_Y_SHIFT 13
645# define R300_CLIPRECT_Y_MASK (0x1FFF << 13)
646#define R300_RE_CLIPRECT_CNTL 0x43D0
647# define R300_CLIP_OUT (1 << 0)
648# define R300_CLIP_0 (1 << 1)
649# define R300_CLIP_1 (1 << 2)
650# define R300_CLIP_10 (1 << 3)
651# define R300_CLIP_2 (1 << 4)
652# define R300_CLIP_20 (1 << 5)
653# define R300_CLIP_21 (1 << 6)
654# define R300_CLIP_210 (1 << 7)
655# define R300_CLIP_3 (1 << 8)
656# define R300_CLIP_30 (1 << 9)
657# define R300_CLIP_31 (1 << 10)
658# define R300_CLIP_310 (1 << 11)
659# define R300_CLIP_32 (1 << 12)
660# define R300_CLIP_320 (1 << 13)
661# define R300_CLIP_321 (1 << 14)
662# define R300_CLIP_3210 (1 << 15)
663
664/* gap */
665#define R300_RE_SCISSORS_TL 0x43E0
666#define R300_RE_SCISSORS_BR 0x43E4
667# define R300_SCISSORS_OFFSET 1440
668# define R300_SCISSORS_X_SHIFT 0
669# define R300_SCISSORS_X_MASK (0x1FFF << 0)
670# define R300_SCISSORS_Y_SHIFT 13
671# define R300_SCISSORS_Y_MASK (0x1FFF << 13)
672/* END */
673
674/* BEGIN: Texture specification
675// The texture specification dwords are grouped by meaning and not by texture unit.
676// This means that e.g. the offset for texture image unit N is found in register
677// TX_OFFSET_0 + (4*N) */
678#define R300_TX_FILTER_0 0x4400
679# define R300_TX_REPEAT 0
680# define R300_TX_MIRRORED 1
681# define R300_TX_CLAMP 4
682# define R300_TX_CLAMP_TO_EDGE 2
683# define R300_TX_CLAMP_TO_BORDER 6
684# define R300_TX_WRAP_S_SHIFT 0
685# define R300_TX_WRAP_S_MASK (7 << 0)
686# define R300_TX_WRAP_T_SHIFT 3
687# define R300_TX_WRAP_T_MASK (7 << 3)
688# define R300_TX_WRAP_Q_SHIFT 6
689# define R300_TX_WRAP_Q_MASK (7 << 6)
690# define R300_TX_MAG_FILTER_NEAREST (1 << 9)
691# define R300_TX_MAG_FILTER_LINEAR (2 << 9)
692# define R300_TX_MAG_FILTER_MASK (3 << 9)
693# define R300_TX_MIN_FILTER_NEAREST (1 << 11)
694# define R300_TX_MIN_FILTER_LINEAR (2 << 11)
695# define R300_TX_MIN_FILTER_NEAREST_MIP_NEAREST (5 << 11)
696# define R300_TX_MIN_FILTER_NEAREST_MIP_LINEAR (9 << 11)
697# define R300_TX_MIN_FILTER_LINEAR_MIP_NEAREST (6 << 11)
698# define R300_TX_MIN_FILTER_LINEAR_MIP_LINEAR (10 << 11)
699
700/* NOTE: NEAREST doesnt seem to exist.
701 Im not seting MAG_FILTER_MASK and (3 << 11) on for all
702 anisotropy modes because that would void selected mag filter */
703# define R300_TX_MIN_FILTER_ANISO_NEAREST ((0 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
704# define R300_TX_MIN_FILTER_ANISO_LINEAR ((0 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
705# define R300_TX_MIN_FILTER_ANISO_NEAREST_MIP_NEAREST ((1 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
706# define R300_TX_MIN_FILTER_ANISO_NEAREST_MIP_LINEAR ((2 << 13) /*|R300_TX_MAG_FILTER_MASK|(3<<11)*/)
707# define R300_TX_MIN_FILTER_MASK ( (15 << 11) | (3 << 13) )
708# define R300_TX_MAX_ANISO_1_TO_1 (0 << 21)
709# define R300_TX_MAX_ANISO_2_TO_1 (2 << 21)
710# define R300_TX_MAX_ANISO_4_TO_1 (4 << 21)
711# define R300_TX_MAX_ANISO_8_TO_1 (6 << 21)
712# define R300_TX_MAX_ANISO_16_TO_1 (8 << 21)
713# define R300_TX_MAX_ANISO_MASK (14 << 21)
714
715#define R300_TX_UNK1_0 0x4440
716# define R300_LOD_BIAS_MASK 0x1fff
717
718#define R300_TX_SIZE_0 0x4480
719# define R300_TX_WIDTHMASK_SHIFT 0
720# define R300_TX_WIDTHMASK_MASK (2047 << 0)
721# define R300_TX_HEIGHTMASK_SHIFT 11
722# define R300_TX_HEIGHTMASK_MASK (2047 << 11)
723# define R300_TX_UNK23 (1 << 23)
724# define R300_TX_SIZE_SHIFT 26 /* largest of width, height */
725# define R300_TX_SIZE_MASK (15 << 26)
726#define R300_TX_FORMAT_0 0x44C0
727 /* The interpretation of the format word by Wladimir van der Laan */
728 /* The X, Y, Z and W refer to the layout of the components.
729 They are given meanings as R, G, B and Alpha by the swizzle
730 specification */
731# define R300_TX_FORMAT_X8 0x0
732# define R300_TX_FORMAT_X16 0x1
733# define R300_TX_FORMAT_Y4X4 0x2
734# define R300_TX_FORMAT_Y8X8 0x3
735# define R300_TX_FORMAT_Y16X16 0x4
736# define R300_TX_FORMAT_Z3Y3X2 0x5
737# define R300_TX_FORMAT_Z5Y6X5 0x6
738# define R300_TX_FORMAT_Z6Y5X5 0x7
739# define R300_TX_FORMAT_Z11Y11X10 0x8
740# define R300_TX_FORMAT_Z10Y11X11 0x9
741# define R300_TX_FORMAT_W4Z4Y4X4 0xA
742# define R300_TX_FORMAT_W1Z5Y5X5 0xB
743# define R300_TX_FORMAT_W8Z8Y8X8 0xC
744# define R300_TX_FORMAT_W2Z10Y10X10 0xD
745# define R300_TX_FORMAT_W16Z16Y16X16 0xE
746# define R300_TX_FORMAT_DXT1 0xF
747# define R300_TX_FORMAT_DXT3 0x10
748# define R300_TX_FORMAT_DXT5 0x11
749# define R300_TX_FORMAT_D3DMFT_CxV8U8 0x12 /* no swizzle */
750# define R300_TX_FORMAT_A8R8G8B8 0x13 /* no swizzle */
751# define R300_TX_FORMAT_B8G8_B8G8 0x14 /* no swizzle */
752# define R300_TX_FORMAT_G8R8_G8B8 0x15 /* no swizzle */
753 /* 0x16 - some 16 bit green format.. ?? */
754# define R300_TX_FORMAT_UNK25 (1 << 25) /* no swizzle */
755
756 /* gap */
757 /* Floating point formats */
758 /* Note - hardware supports both 16 and 32 bit floating point */
759# define R300_TX_FORMAT_FL_I16 0x18
760# define R300_TX_FORMAT_FL_I16A16 0x19
761# define R300_TX_FORMAT_FL_R16G16B16A16 0x1A
762# define R300_TX_FORMAT_FL_I32 0x1B
763# define R300_TX_FORMAT_FL_I32A32 0x1C
764# define R300_TX_FORMAT_FL_R32G32B32A32 0x1D
765 /* alpha modes, convenience mostly */
766 /* if you have alpha, pick constant appropriate to the
767 number of channels (1 for I8, 2 for I8A8, 4 for R8G8B8A8, etc */
768# define R300_TX_FORMAT_ALPHA_1CH 0x000
769# define R300_TX_FORMAT_ALPHA_2CH 0x200
770# define R300_TX_FORMAT_ALPHA_4CH 0x600
771# define R300_TX_FORMAT_ALPHA_NONE 0xA00
772 /* Swizzling */
773 /* constants */
774# define R300_TX_FORMAT_X 0
775# define R300_TX_FORMAT_Y 1
776# define R300_TX_FORMAT_Z 2
777# define R300_TX_FORMAT_W 3
778# define R300_TX_FORMAT_ZERO 4
779# define R300_TX_FORMAT_ONE 5
780# define R300_TX_FORMAT_CUT_Z 6 /* 2.0*Z, everything above 1.0 is set to 0.0 */
781# define R300_TX_FORMAT_CUT_W 7 /* 2.0*W, everything above 1.0 is set to 0.0 */
782
783# define R300_TX_FORMAT_B_SHIFT 18
784# define R300_TX_FORMAT_G_SHIFT 15
785# define R300_TX_FORMAT_R_SHIFT 12
786# define R300_TX_FORMAT_A_SHIFT 9
787 /* Convenience macro to take care of layout and swizzling */
788# define R300_EASY_TX_FORMAT(B, G, R, A, FMT) (\
789 ((R300_TX_FORMAT_##B)<<R300_TX_FORMAT_B_SHIFT) \
790 | ((R300_TX_FORMAT_##G)<<R300_TX_FORMAT_G_SHIFT) \
791 | ((R300_TX_FORMAT_##R)<<R300_TX_FORMAT_R_SHIFT) \
792 | ((R300_TX_FORMAT_##A)<<R300_TX_FORMAT_A_SHIFT) \
793 | (R300_TX_FORMAT_##FMT) \
794 )
795 /* These can be ORed with result of R300_EASY_TX_FORMAT() */
796 /* We don't really know what they do. Take values from a constant color ? */
797# define R300_TX_FORMAT_CONST_X (1<<5)
798# define R300_TX_FORMAT_CONST_Y (2<<5)
799# define R300_TX_FORMAT_CONST_Z (4<<5)
800# define R300_TX_FORMAT_CONST_W (8<<5)
801
802# define R300_TX_FORMAT_YUV_MODE 0x00800000
803
804#define R300_TX_OFFSET_0 0x4540
805/* BEGIN: Guess from R200 */
806# define R300_TXO_ENDIAN_NO_SWAP (0 << 0)
807# define R300_TXO_ENDIAN_BYTE_SWAP (1 << 0)
808# define R300_TXO_ENDIAN_WORD_SWAP (2 << 0)
809# define R300_TXO_ENDIAN_HALFDW_SWAP (3 << 0)
810# define R300_TXO_OFFSET_MASK 0xffffffe0
811# define R300_TXO_OFFSET_SHIFT 5
812/* END */
813#define R300_TX_UNK4_0 0x4580
814#define R300_TX_BORDER_COLOR_0 0x45C0 //ff00ff00 == { 0, 1.0, 0, 1.0 }
815
816/* END */
817
818/* BEGIN: Fragment program instruction set
819// Fragment programs are written directly into register space.
820// There are separate instruction streams for texture instructions and ALU
821// instructions.
822// In order to synchronize these streams, the program is divided into up
823// to 4 nodes. Each node begins with a number of TEX operations, followed
824// by a number of ALU operations.
825// The first node can have zero TEX ops, all subsequent nodes must have at least
826// one TEX ops.
827// All nodes must have at least one ALU op.
828//
829// The index of the last node is stored in PFS_CNTL_0: A value of 0 means
830// 1 node, a value of 3 means 4 nodes.
831// The total amount of instructions is defined in PFS_CNTL_2. The offsets are
832// offsets into the respective instruction streams, while *_END points to the
833// last instruction relative to this offset. */
834#define R300_PFS_CNTL_0 0x4600
835# define R300_PFS_CNTL_LAST_NODES_SHIFT 0
836# define R300_PFS_CNTL_LAST_NODES_MASK (3 << 0)
837# define R300_PFS_CNTL_FIRST_NODE_HAS_TEX (1 << 3)
838#define R300_PFS_CNTL_1 0x4604
839/* There is an unshifted value here which has so far always been equal to the
840// index of the highest used temporary register. */
841#define R300_PFS_CNTL_2 0x4608
842# define R300_PFS_CNTL_ALU_OFFSET_SHIFT 0
843# define R300_PFS_CNTL_ALU_OFFSET_MASK (63 << 0)
844# define R300_PFS_CNTL_ALU_END_SHIFT 6
845# define R300_PFS_CNTL_ALU_END_MASK (63 << 0)
846# define R300_PFS_CNTL_TEX_OFFSET_SHIFT 12
847# define R300_PFS_CNTL_TEX_OFFSET_MASK (31 << 12) /* GUESS */
848# define R300_PFS_CNTL_TEX_END_SHIFT 18
849# define R300_PFS_CNTL_TEX_END_MASK (31 << 18) /* GUESS */
850
851/* gap */
852/* Nodes are stored backwards. The last active node is always stored in
853// PFS_NODE_3.
854// Example: In a 2-node program, NODE_0 and NODE_1 are set to 0. The
855// first node is stored in NODE_2, the second node is stored in NODE_3.
856//
857// Offsets are relative to the master offset from PFS_CNTL_2.
858// LAST_NODE is set for the last node, and only for the last node. */
859#define R300_PFS_NODE_0 0x4610
860#define R300_PFS_NODE_1 0x4614
861#define R300_PFS_NODE_2 0x4618
862#define R300_PFS_NODE_3 0x461C
863# define R300_PFS_NODE_ALU_OFFSET_SHIFT 0
864# define R300_PFS_NODE_ALU_OFFSET_MASK (63 << 0)
865# define R300_PFS_NODE_ALU_END_SHIFT 6
866# define R300_PFS_NODE_ALU_END_MASK (63 << 6)
867# define R300_PFS_NODE_TEX_OFFSET_SHIFT 12
868# define R300_PFS_NODE_TEX_OFFSET_MASK (31 << 12)
869# define R300_PFS_NODE_TEX_END_SHIFT 17
870# define R300_PFS_NODE_TEX_END_MASK (31 << 17)
871# define R300_PFS_NODE_LAST_NODE (1 << 22)
872
873/* TEX
874// As far as I can tell, texture instructions cannot write into output
875// registers directly. A subsequent ALU instruction is always necessary,
876// even if it's just MAD o0, r0, 1, 0 */
877#define R300_PFS_TEXI_0 0x4620
878# define R300_FPITX_SRC_SHIFT 0
879# define R300_FPITX_SRC_MASK (31 << 0)
880# define R300_FPITX_SRC_CONST (1 << 5) /* GUESS */
881# define R300_FPITX_DST_SHIFT 6
882# define R300_FPITX_DST_MASK (31 << 6)
883# define R300_FPITX_IMAGE_SHIFT 11
884# define R300_FPITX_IMAGE_MASK (15 << 11) /* GUESS based on layout and native limits */
885/* Unsure if these are opcodes, or some kind of bitfield, but this is how
886 * they were set when I checked
887 */
888# define R300_FPITX_OPCODE_SHIFT 15
889# define R300_FPITX_OP_TEX 1
890# define R300_FPITX_OP_TXP 3
891# define R300_FPITX_OP_TXB 4
892
893/* ALU
894// The ALU instructions register blocks are enumerated according to the order
895// in which fglrx. I assume there is space for 64 instructions, since
896// each block has space for a maximum of 64 DWORDs, and this matches reported
897// native limits.
898//
899// The basic functional block seems to be one MAD for each color and alpha,
900// and an adder that adds all components after the MUL.
901// - ADD, MUL, MAD etc.: use MAD with appropriate neutral operands
902// - DP4: Use OUTC_DP4, OUTA_DP4
903// - DP3: Use OUTC_DP3, OUTA_DP4, appropriate alpha operands
904// - DPH: Use OUTC_DP4, OUTA_DP4, appropriate alpha operands
905// - CMP: If ARG2 < 0, return ARG1, else return ARG0
906// - FLR: use FRC+MAD
907// - XPD: use MAD+MAD
908// - SGE, SLT: use MAD+CMP
909// - RSQ: use ABS modifier for argument
910// - Use OUTC_REPL_ALPHA to write results of an alpha-only operation (e.g. RCP)
911// into color register
912// - apparently, there's no quick DST operation
913// - fglrx set FPI2_UNKNOWN_31 on a "MAD fragment.color, tmp0, tmp1, tmp2"
914// - fglrx set FPI2_UNKNOWN_31 on a "MAX r2, r1, c0"
915// - fglrx once set FPI0_UNKNOWN_31 on a "FRC r1, r1"
916//
917// Operand selection
918// First stage selects three sources from the available registers and
919// constant parameters. This is defined in INSTR1 (color) and INSTR3 (alpha).
920// fglrx sorts the three source fields: Registers before constants,
921// lower indices before higher indices; I do not know whether this is necessary.
922// fglrx fills unused sources with "read constant 0"
923// According to specs, you cannot select more than two different constants.
924//
925// Second stage selects the operands from the sources. This is defined in
926// INSTR0 (color) and INSTR2 (alpha). You can also select the special constants
927// zero and one.
928// Swizzling and negation happens in this stage, as well.
929//
930// Important: Color and alpha seem to be mostly separate, i.e. their sources
931// selection appears to be fully independent (the register storage is probably
932// physically split into a color and an alpha section).
933// However (because of the apparent physical split), there is some interaction
934// WRT swizzling. If, for example, you want to load an R component into an
935// Alpha operand, this R component is taken from a *color* source, not from
936// an alpha source. The corresponding register doesn't even have to appear in
937// the alpha sources list. (I hope this alll makes sense to you)
938//
939// Destination selection
940// The destination register index is in FPI1 (color) and FPI3 (alpha) together
941// with enable bits.
942// There are separate enable bits for writing into temporary registers
943// (DSTC_REG_* /DSTA_REG) and and program output registers (DSTC_OUTPUT_* /DSTA_OUTPUT).
944// You can write to both at once, or not write at all (the same index
945// must be used for both).
946//
947// Note: There is a special form for LRP
948// - Argument order is the same as in ARB_fragment_program.
949// - Operation is MAD
950// - ARG1 is set to ARGC_SRC1C_LRP/ARGC_SRC1A_LRP
951// - Set FPI0/FPI2_SPECIAL_LRP
952// Arbitrary LRP (including support for swizzling) requires vanilla MAD+MAD */
953#define R300_PFS_INSTR1_0 0x46C0
954# define R300_FPI1_SRC0C_SHIFT 0
955# define R300_FPI1_SRC0C_MASK (31 << 0)
956# define R300_FPI1_SRC0C_CONST (1 << 5)
957# define R300_FPI1_SRC1C_SHIFT 6
958# define R300_FPI1_SRC1C_MASK (31 << 6)
959# define R300_FPI1_SRC1C_CONST (1 << 11)
960# define R300_FPI1_SRC2C_SHIFT 12
961# define R300_FPI1_SRC2C_MASK (31 << 12)
962# define R300_FPI1_SRC2C_CONST (1 << 17)
963# define R300_FPI1_DSTC_SHIFT 18
964# define R300_FPI1_DSTC_MASK (31 << 18)
965# define R300_FPI1_DSTC_REG_X (1 << 23)
966# define R300_FPI1_DSTC_REG_Y (1 << 24)
967# define R300_FPI1_DSTC_REG_Z (1 << 25)
968# define R300_FPI1_DSTC_OUTPUT_X (1 << 26)
969# define R300_FPI1_DSTC_OUTPUT_Y (1 << 27)
970# define R300_FPI1_DSTC_OUTPUT_Z (1 << 28)
971
972#define R300_PFS_INSTR3_0 0x47C0
973# define R300_FPI3_SRC0A_SHIFT 0
974# define R300_FPI3_SRC0A_MASK (31 << 0)
975# define R300_FPI3_SRC0A_CONST (1 << 5)
976# define R300_FPI3_SRC1A_SHIFT 6
977# define R300_FPI3_SRC1A_MASK (31 << 6)
978# define R300_FPI3_SRC1A_CONST (1 << 11)
979# define R300_FPI3_SRC2A_SHIFT 12
980# define R300_FPI3_SRC2A_MASK (31 << 12)
981# define R300_FPI3_SRC2A_CONST (1 << 17)
982# define R300_FPI3_DSTA_SHIFT 18
983# define R300_FPI3_DSTA_MASK (31 << 18)
984# define R300_FPI3_DSTA_REG (1 << 23)
985# define R300_FPI3_DSTA_OUTPUT (1 << 24)
986
987#define R300_PFS_INSTR0_0 0x48C0
988# define R300_FPI0_ARGC_SRC0C_XYZ 0
989# define R300_FPI0_ARGC_SRC0C_XXX 1
990# define R300_FPI0_ARGC_SRC0C_YYY 2
991# define R300_FPI0_ARGC_SRC0C_ZZZ 3
992# define R300_FPI0_ARGC_SRC1C_XYZ 4
993# define R300_FPI0_ARGC_SRC1C_XXX 5
994# define R300_FPI0_ARGC_SRC1C_YYY 6
995# define R300_FPI0_ARGC_SRC1C_ZZZ 7
996# define R300_FPI0_ARGC_SRC2C_XYZ 8
997# define R300_FPI0_ARGC_SRC2C_XXX 9
998# define R300_FPI0_ARGC_SRC2C_YYY 10
999# define R300_FPI0_ARGC_SRC2C_ZZZ 11
1000# define R300_FPI0_ARGC_SRC0A 12
1001# define R300_FPI0_ARGC_SRC1A 13
1002# define R300_FPI0_ARGC_SRC2A 14
1003# define R300_FPI0_ARGC_SRC1C_LRP 15
1004# define R300_FPI0_ARGC_ZERO 20
1005# define R300_FPI0_ARGC_ONE 21
1006# define R300_FPI0_ARGC_HALF 22 /* GUESS */
1007# define R300_FPI0_ARGC_SRC0C_YZX 23
1008# define R300_FPI0_ARGC_SRC1C_YZX 24
1009# define R300_FPI0_ARGC_SRC2C_YZX 25
1010# define R300_FPI0_ARGC_SRC0C_ZXY 26
1011# define R300_FPI0_ARGC_SRC1C_ZXY 27
1012# define R300_FPI0_ARGC_SRC2C_ZXY 28
1013# define R300_FPI0_ARGC_SRC0CA_WZY 29
1014# define R300_FPI0_ARGC_SRC1CA_WZY 30
1015# define R300_FPI0_ARGC_SRC2CA_WZY 31
1016
1017# define R300_FPI0_ARG0C_SHIFT 0
1018# define R300_FPI0_ARG0C_MASK (31 << 0)
1019# define R300_FPI0_ARG0C_NEG (1 << 5)
1020# define R300_FPI0_ARG0C_ABS (1 << 6)
1021# define R300_FPI0_ARG1C_SHIFT 7
1022# define R300_FPI0_ARG1C_MASK (31 << 7)
1023# define R300_FPI0_ARG1C_NEG (1 << 12)
1024# define R300_FPI0_ARG1C_ABS (1 << 13)
1025# define R300_FPI0_ARG2C_SHIFT 14
1026# define R300_FPI0_ARG2C_MASK (31 << 14)
1027# define R300_FPI0_ARG2C_NEG (1 << 19)
1028# define R300_FPI0_ARG2C_ABS (1 << 20)
1029# define R300_FPI0_SPECIAL_LRP (1 << 21)
1030# define R300_FPI0_OUTC_MAD (0 << 23)
1031# define R300_FPI0_OUTC_DP3 (1 << 23)
1032# define R300_FPI0_OUTC_DP4 (2 << 23)
1033# define R300_FPI0_OUTC_MIN (4 << 23)
1034# define R300_FPI0_OUTC_MAX (5 << 23)
1035# define R300_FPI0_OUTC_CMP (8 << 23)
1036# define R300_FPI0_OUTC_FRC (9 << 23)
1037# define R300_FPI0_OUTC_REPL_ALPHA (10 << 23)
1038# define R300_FPI0_OUTC_SAT (1 << 30)
1039# define R300_FPI0_UNKNOWN_31 (1 << 31)
1040
1041#define R300_PFS_INSTR2_0 0x49C0
1042# define R300_FPI2_ARGA_SRC0C_X 0
1043# define R300_FPI2_ARGA_SRC0C_Y 1
1044# define R300_FPI2_ARGA_SRC0C_Z 2
1045# define R300_FPI2_ARGA_SRC1C_X 3
1046# define R300_FPI2_ARGA_SRC1C_Y 4
1047# define R300_FPI2_ARGA_SRC1C_Z 5
1048# define R300_FPI2_ARGA_SRC2C_X 6
1049# define R300_FPI2_ARGA_SRC2C_Y 7
1050# define R300_FPI2_ARGA_SRC2C_Z 8
1051# define R300_FPI2_ARGA_SRC0A 9
1052# define R300_FPI2_ARGA_SRC1A 10
1053# define R300_FPI2_ARGA_SRC2A 11
1054# define R300_FPI2_ARGA_SRC1A_LRP 15
1055# define R300_FPI2_ARGA_ZERO 16
1056# define R300_FPI2_ARGA_ONE 17
1057# define R300_FPI2_ARGA_HALF 18 /* GUESS */
1058
1059# define R300_FPI2_ARG0A_SHIFT 0
1060# define R300_FPI2_ARG0A_MASK (31 << 0)
1061# define R300_FPI2_ARG0A_NEG (1 << 5)
1062# define R300_FPI2_ARG0A_ABS (1 << 6) /* GUESS */
1063# define R300_FPI2_ARG1A_SHIFT 7
1064# define R300_FPI2_ARG1A_MASK (31 << 7)
1065# define R300_FPI2_ARG1A_NEG (1 << 12)
1066# define R300_FPI2_ARG1A_ABS (1 << 13) /* GUESS */
1067# define R300_FPI2_ARG2A_SHIFT 14
1068# define R300_FPI2_ARG2A_MASK (31 << 14)
1069# define R300_FPI2_ARG2A_NEG (1 << 19)
1070# define R300_FPI2_ARG2A_ABS (1 << 20) /* GUESS */
1071# define R300_FPI2_SPECIAL_LRP (1 << 21)
1072# define R300_FPI2_OUTA_MAD (0 << 23)
1073# define R300_FPI2_OUTA_DP4 (1 << 23)
1074# define R300_FPI2_OUTA_MIN (2 << 23)
1075# define R300_FPI2_OUTA_MAX (3 << 23)
1076# define R300_FPI2_OUTA_CMP (6 << 23)
1077# define R300_FPI2_OUTA_FRC (7 << 23)
1078# define R300_FPI2_OUTA_EX2 (8 << 23)
1079# define R300_FPI2_OUTA_LG2 (9 << 23)
1080# define R300_FPI2_OUTA_RCP (10 << 23)
1081# define R300_FPI2_OUTA_RSQ (11 << 23)
1082# define R300_FPI2_OUTA_SAT (1 << 30)
1083# define R300_FPI2_UNKNOWN_31 (1 << 31)
1084/* END */
1085
1086/* gap */
1087#define R300_PP_ALPHA_TEST 0x4BD4
1088# define R300_REF_ALPHA_MASK 0x000000ff
1089# define R300_ALPHA_TEST_FAIL (0 << 8)
1090# define R300_ALPHA_TEST_LESS (1 << 8)
1091# define R300_ALPHA_TEST_LEQUAL (3 << 8)
1092# define R300_ALPHA_TEST_EQUAL (2 << 8)
1093# define R300_ALPHA_TEST_GEQUAL (6 << 8)
1094# define R300_ALPHA_TEST_GREATER (4 << 8)
1095# define R300_ALPHA_TEST_NEQUAL (5 << 8)
1096# define R300_ALPHA_TEST_PASS (7 << 8)
1097# define R300_ALPHA_TEST_OP_MASK (7 << 8)
1098# define R300_ALPHA_TEST_ENABLE (1 << 11)
1099
1100/* gap */
1101/* Fragment program parameters in 7.16 floating point */
1102#define R300_PFS_PARAM_0_X 0x4C00
1103#define R300_PFS_PARAM_0_Y 0x4C04
1104#define R300_PFS_PARAM_0_Z 0x4C08
1105#define R300_PFS_PARAM_0_W 0x4C0C
1106/* GUESS: PARAM_31 is last, based on native limits reported by fglrx */
1107#define R300_PFS_PARAM_31_X 0x4DF0
1108#define R300_PFS_PARAM_31_Y 0x4DF4
1109#define R300_PFS_PARAM_31_Z 0x4DF8
1110#define R300_PFS_PARAM_31_W 0x4DFC
1111
1112/* Notes:
1113// - AFAIK fglrx always sets BLEND_UNKNOWN when blending is used in the application
1114// - AFAIK fglrx always sets BLEND_NO_SEPARATE when CBLEND and ABLEND are set to the same
1115// function (both registers are always set up completely in any case)
1116// - Most blend flags are simply copied from R200 and not tested yet */
1117#define R300_RB3D_CBLEND 0x4E04
1118#define R300_RB3D_ABLEND 0x4E08
1119 /* the following only appear in CBLEND */
1120# define R300_BLEND_ENABLE (1 << 0)
1121# define R300_BLEND_UNKNOWN (3 << 1)
1122# define R300_BLEND_NO_SEPARATE (1 << 3)
1123 /* the following are shared between CBLEND and ABLEND */
1124# define R300_FCN_MASK (3 << 12)
1125# define R300_COMB_FCN_ADD_CLAMP (0 << 12)
1126# define R300_COMB_FCN_ADD_NOCLAMP (1 << 12)
1127# define R300_COMB_FCN_SUB_CLAMP (2 << 12)
1128# define R300_COMB_FCN_SUB_NOCLAMP (3 << 12)
1129# define R300_SRC_BLEND_GL_ZERO (32 << 16)
1130# define R300_SRC_BLEND_GL_ONE (33 << 16)
1131# define R300_SRC_BLEND_GL_SRC_COLOR (34 << 16)
1132# define R300_SRC_BLEND_GL_ONE_MINUS_SRC_COLOR (35 << 16)
1133# define R300_SRC_BLEND_GL_DST_COLOR (36 << 16)
1134# define R300_SRC_BLEND_GL_ONE_MINUS_DST_COLOR (37 << 16)
1135# define R300_SRC_BLEND_GL_SRC_ALPHA (38 << 16)
1136# define R300_SRC_BLEND_GL_ONE_MINUS_SRC_ALPHA (39 << 16)
1137# define R300_SRC_BLEND_GL_DST_ALPHA (40 << 16)
1138# define R300_SRC_BLEND_GL_ONE_MINUS_DST_ALPHA (41 << 16)
1139# define R300_SRC_BLEND_GL_SRC_ALPHA_SATURATE (42 << 16)
1140# define R300_SRC_BLEND_MASK (63 << 16)
1141# define R300_DST_BLEND_GL_ZERO (32 << 24)
1142# define R300_DST_BLEND_GL_ONE (33 << 24)
1143# define R300_DST_BLEND_GL_SRC_COLOR (34 << 24)
1144# define R300_DST_BLEND_GL_ONE_MINUS_SRC_COLOR (35 << 24)
1145# define R300_DST_BLEND_GL_DST_COLOR (36 << 24)
1146# define R300_DST_BLEND_GL_ONE_MINUS_DST_COLOR (37 << 24)
1147# define R300_DST_BLEND_GL_SRC_ALPHA (38 << 24)
1148# define R300_DST_BLEND_GL_ONE_MINUS_SRC_ALPHA (39 << 24)
1149# define R300_DST_BLEND_GL_DST_ALPHA (40 << 24)
1150# define R300_DST_BLEND_GL_ONE_MINUS_DST_ALPHA (41 << 24)
1151# define R300_DST_BLEND_MASK (63 << 24)
1152#define R300_RB3D_COLORMASK 0x4E0C
1153# define R300_COLORMASK0_B (1<<0)
1154# define R300_COLORMASK0_G (1<<1)
1155# define R300_COLORMASK0_R (1<<2)
1156# define R300_COLORMASK0_A (1<<3)
1157
1158/* gap */
1159#define R300_RB3D_COLOROFFSET0 0x4E28
1160# define R300_COLOROFFSET_MASK 0xFFFFFFF0 /* GUESS */
1161#define R300_RB3D_COLOROFFSET1 0x4E2C /* GUESS */
1162#define R300_RB3D_COLOROFFSET2 0x4E30 /* GUESS */
1163#define R300_RB3D_COLOROFFSET3 0x4E34 /* GUESS */
1164/* gap */
1165/* Bit 16: Larger tiles
1166// Bit 17: 4x2 tiles
1167// Bit 18: Extremely weird tile like, but some pixels duplicated? */
1168#define R300_RB3D_COLORPITCH0 0x4E38
1169# define R300_COLORPITCH_MASK 0x00001FF8 /* GUESS */
1170# define R300_COLOR_TILE_ENABLE (1 << 16) /* GUESS */
1171# define R300_COLOR_MICROTILE_ENABLE (1 << 17) /* GUESS */
1172# define R300_COLOR_ENDIAN_NO_SWAP (0 << 18) /* GUESS */
1173# define R300_COLOR_ENDIAN_WORD_SWAP (1 << 18) /* GUESS */
1174# define R300_COLOR_ENDIAN_DWORD_SWAP (2 << 18) /* GUESS */
1175# define R300_COLOR_FORMAT_RGB565 (2 << 22)
1176# define R300_COLOR_FORMAT_ARGB8888 (3 << 22)
1177#define R300_RB3D_COLORPITCH1 0x4E3C /* GUESS */
1178#define R300_RB3D_COLORPITCH2 0x4E40 /* GUESS */
1179#define R300_RB3D_COLORPITCH3 0x4E44 /* GUESS */
1180
1181/* gap */
1182/* Guess by Vladimir.
1183// Set to 0A before 3D operations, set to 02 afterwards. */
1184#define R300_RB3D_DSTCACHE_CTLSTAT 0x4E4C
1185# define R300_RB3D_DSTCACHE_02 0x00000002
1186# define R300_RB3D_DSTCACHE_0A 0x0000000A
1187
1188/* gap */
1189/* There seems to be no "write only" setting, so use Z-test = ALWAYS for this. */
1190/* Bit (1<<8) is the "test" bit. so plain write is 6 - vd */
1191#define R300_RB3D_ZSTENCIL_CNTL_0 0x4F00
1192# define R300_RB3D_Z_DISABLED_1 0x00000010 /* GUESS */
1193# define R300_RB3D_Z_DISABLED_2 0x00000014 /* GUESS */
1194# define R300_RB3D_Z_TEST 0x00000012
1195# define R300_RB3D_Z_TEST_AND_WRITE 0x00000016
1196# define R300_RB3D_Z_WRITE_ONLY 0x00000006
1197
1198# define R300_RB3D_Z_TEST 0x00000012
1199# define R300_RB3D_Z_TEST_AND_WRITE 0x00000016
1200# define R300_RB3D_Z_WRITE_ONLY 0x00000006
1201# define R300_RB3D_STENCIL_ENABLE 0x00000001
1202
1203#define R300_RB3D_ZSTENCIL_CNTL_1 0x4F04
1204 /* functions */
1205# define R300_ZS_NEVER 0
1206# define R300_ZS_LESS 1
1207# define R300_ZS_LEQUAL 2
1208# define R300_ZS_EQUAL 3
1209# define R300_ZS_GEQUAL 4
1210# define R300_ZS_GREATER 5
1211# define R300_ZS_NOTEQUAL 6
1212# define R300_ZS_ALWAYS 7
1213# define R300_ZS_MASK 7
1214 /* operations */
1215# define R300_ZS_KEEP 0
1216# define R300_ZS_ZERO 1
1217# define R300_ZS_REPLACE 2
1218# define R300_ZS_INCR 3
1219# define R300_ZS_DECR 4
1220# define R300_ZS_INVERT 5
1221# define R300_ZS_INCR_WRAP 6
1222# define R300_ZS_DECR_WRAP 7
1223
1224 /* front and back refer to operations done for front
1225 and back faces, i.e. separate stencil function support */
1226# define R300_RB3D_ZS1_DEPTH_FUNC_SHIFT 0
1227# define R300_RB3D_ZS1_FRONT_FUNC_SHIFT 3
1228# define R300_RB3D_ZS1_FRONT_FAIL_OP_SHIFT 6
1229# define R300_RB3D_ZS1_FRONT_ZPASS_OP_SHIFT 9
1230# define R300_RB3D_ZS1_FRONT_ZFAIL_OP_SHIFT 12
1231# define R300_RB3D_ZS1_BACK_FUNC_SHIFT 15
1232# define R300_RB3D_ZS1_BACK_FAIL_OP_SHIFT 18
1233# define R300_RB3D_ZS1_BACK_ZPASS_OP_SHIFT 21
1234# define R300_RB3D_ZS1_BACK_ZFAIL_OP_SHIFT 24
1235
1236
1237
1238#define R300_RB3D_ZSTENCIL_CNTL_2 0x4F08
1239# define R300_RB3D_ZS2_STENCIL_REF_SHIFT 0
1240# define R300_RB3D_ZS2_STENCIL_MASK 0xFF
1241# define R300_RB3D_ZS2_STENCIL_MASK_SHIFT 8
1242# define R300_RB3D_ZS2_STENCIL_WRITE_MASK_SHIFT 16
1243
1244/* gap */
1245
1246#define R300_RB3D_ZSTENCIL_FORMAT 0x4F10
1247# define R300_DEPTH_FORMAT_16BIT_INT_Z (0 << 0)
1248# define R300_DEPTH_FORMAT_24BIT_INT_Z (2 << 0)
1249
1250/* gap */
1251#define R300_RB3D_DEPTHOFFSET 0x4F20
1252#define R300_RB3D_DEPTHPITCH 0x4F24
1253# define R300_DEPTHPITCH_MASK 0x00001FF8 /* GUESS */
1254# define R300_DEPTH_TILE_ENABLE (1 << 16) /* GUESS */
1255# define R300_DEPTH_MICROTILE_ENABLE (1 << 17) /* GUESS */
1256# define R300_DEPTH_ENDIAN_NO_SWAP (0 << 18) /* GUESS */
1257# define R300_DEPTH_ENDIAN_WORD_SWAP (1 << 18) /* GUESS */
1258# define R300_DEPTH_ENDIAN_DWORD_SWAP (2 << 18) /* GUESS */
1259
1260/* BEGIN: Vertex program instruction set
1261// Every instruction is four dwords long:
1262// DWORD 0: output and opcode
1263// DWORD 1: first argument
1264// DWORD 2: second argument
1265// DWORD 3: third argument
1266//
1267// Notes:
1268// - ABS r, a is implemented as MAX r, a, -a
1269// - MOV is implemented as ADD to zero
1270// - XPD is implemented as MUL + MAD
1271// - FLR is implemented as FRC + ADD
1272// - apparently, fglrx tries to schedule instructions so that there is at least
1273// one instruction between the write to a temporary and the first read
1274// from said temporary; however, violations of this scheduling are allowed
1275// - register indices seem to be unrelated with OpenGL aliasing to conventional state
1276// - only one attribute and one parameter can be loaded at a time; however, the
1277// same attribute/parameter can be used for more than one argument
1278// - the second software argument for POW is the third hardware argument (no idea why)
1279// - MAD with only temporaries as input seems to use VPI_OUT_SELECT_MAD_2
1280//
1281// There is some magic surrounding LIT:
1282// The single argument is replicated across all three inputs, but swizzled:
1283// First argument: xyzy
1284// Second argument: xyzx
1285// Third argument: xyzw
1286// Whenever the result is used later in the fragment program, fglrx forces x and w
1287// to be 1.0 in the input selection; I don't know whether this is strictly necessary */
1288#define R300_VPI_OUT_OP_DOT (1 << 0)
1289#define R300_VPI_OUT_OP_MUL (2 << 0)
1290#define R300_VPI_OUT_OP_ADD (3 << 0)
1291#define R300_VPI_OUT_OP_MAD (4 << 0)
1292#define R300_VPI_OUT_OP_DST (5 << 0)
1293#define R300_VPI_OUT_OP_FRC (6 << 0)
1294#define R300_VPI_OUT_OP_MAX (7 << 0)
1295#define R300_VPI_OUT_OP_MIN (8 << 0)
1296#define R300_VPI_OUT_OP_SGE (9 << 0)
1297#define R300_VPI_OUT_OP_SLT (10 << 0)
1298#define R300_VPI_OUT_OP_UNK12 (12 << 0) /* Used in GL_POINT_DISTANCE_ATTENUATION_ARB, vector(scalar, vector) */
1299#define R300_VPI_OUT_OP_EXP (65 << 0)
1300#define R300_VPI_OUT_OP_LOG (66 << 0)
1301#define R300_VPI_OUT_OP_UNK67 (67 << 0) /* Used in fog computations, scalar(scalar) */
1302#define R300_VPI_OUT_OP_LIT (68 << 0)
1303#define R300_VPI_OUT_OP_POW (69 << 0)
1304#define R300_VPI_OUT_OP_RCP (70 << 0)
1305#define R300_VPI_OUT_OP_RSQ (72 << 0)
1306#define R300_VPI_OUT_OP_UNK73 (73 << 0) /* Used in GL_POINT_DISTANCE_ATTENUATION_ARB, scalar(scalar) */
1307#define R300_VPI_OUT_OP_EX2 (75 << 0)
1308#define R300_VPI_OUT_OP_LG2 (76 << 0)
1309#define R300_VPI_OUT_OP_MAD_2 (128 << 0)
1310#define R300_VPI_OUT_OP_UNK129 (129 << 0) /* all temps, vector(scalar, vector, vector) */
1311
1312#define R300_VPI_OUT_REG_CLASS_TEMPORARY (0 << 8)
1313#define R300_VPI_OUT_REG_CLASS_RESULT (2 << 8)
1314#define R300_VPI_OUT_REG_CLASS_MASK (31 << 8)
1315
1316#define R300_VPI_OUT_REG_INDEX_SHIFT 13
1317#define R300_VPI_OUT_REG_INDEX_MASK (31 << 13) /* GUESS based on fglrx native limits */
1318
1319#define R300_VPI_OUT_WRITE_X (1 << 20)
1320#define R300_VPI_OUT_WRITE_Y (1 << 21)
1321#define R300_VPI_OUT_WRITE_Z (1 << 22)
1322#define R300_VPI_OUT_WRITE_W (1 << 23)
1323
1324#define R300_VPI_IN_REG_CLASS_TEMPORARY (0 << 0)
1325#define R300_VPI_IN_REG_CLASS_ATTRIBUTE (1 << 0)
1326#define R300_VPI_IN_REG_CLASS_PARAMETER (2 << 0)
1327#define R300_VPI_IN_REG_CLASS_NONE (9 << 0)
1328#define R300_VPI_IN_REG_CLASS_MASK (31 << 0) /* GUESS */
1329
1330#define R300_VPI_IN_REG_INDEX_SHIFT 5
1331#define R300_VPI_IN_REG_INDEX_MASK (255 << 5) /* GUESS based on fglrx native limits */
1332
1333/* The R300 can select components from the input register arbitrarily.
1334// Use the following constants, shifted by the component shift you
1335// want to select */
1336#define R300_VPI_IN_SELECT_X 0
1337#define R300_VPI_IN_SELECT_Y 1
1338#define R300_VPI_IN_SELECT_Z 2
1339#define R300_VPI_IN_SELECT_W 3
1340#define R300_VPI_IN_SELECT_ZERO 4
1341#define R300_VPI_IN_SELECT_ONE 5
1342#define R300_VPI_IN_SELECT_MASK 7
1343
1344#define R300_VPI_IN_X_SHIFT 13
1345#define R300_VPI_IN_Y_SHIFT 16
1346#define R300_VPI_IN_Z_SHIFT 19
1347#define R300_VPI_IN_W_SHIFT 22
1348
1349#define R300_VPI_IN_NEG_X (1 << 25)
1350#define R300_VPI_IN_NEG_Y (1 << 26)
1351#define R300_VPI_IN_NEG_Z (1 << 27)
1352#define R300_VPI_IN_NEG_W (1 << 28)
1353/* END */
1354
1355//BEGIN: Packet 3 commands
1356
1357// A primitive emission dword.
1358#define R300_PRIM_TYPE_NONE (0 << 0)
1359#define R300_PRIM_TYPE_POINT (1 << 0)
1360#define R300_PRIM_TYPE_LINE (2 << 0)
1361#define R300_PRIM_TYPE_LINE_STRIP (3 << 0)
1362#define R300_PRIM_TYPE_TRI_LIST (4 << 0)
1363#define R300_PRIM_TYPE_TRI_FAN (5 << 0)
1364#define R300_PRIM_TYPE_TRI_STRIP (6 << 0)
1365#define R300_PRIM_TYPE_TRI_TYPE2 (7 << 0)
1366#define R300_PRIM_TYPE_RECT_LIST (8 << 0)
1367#define R300_PRIM_TYPE_3VRT_POINT_LIST (9 << 0)
1368#define R300_PRIM_TYPE_3VRT_LINE_LIST (10 << 0)
1369#define R300_PRIM_TYPE_POINT_SPRITES (11 << 0) // GUESS (based on r200)
1370#define R300_PRIM_TYPE_LINE_LOOP (12 << 0)
1371#define R300_PRIM_TYPE_QUADS (13 << 0)
1372#define R300_PRIM_TYPE_QUAD_STRIP (14 << 0)
1373#define R300_PRIM_TYPE_POLYGON (15 << 0)
1374#define R300_PRIM_TYPE_MASK 0xF
1375#define R300_PRIM_WALK_IND (1 << 4)
1376#define R300_PRIM_WALK_LIST (2 << 4)
1377#define R300_PRIM_WALK_RING (3 << 4)
1378#define R300_PRIM_WALK_MASK (3 << 4)
1379#define R300_PRIM_COLOR_ORDER_BGRA (0 << 6) // GUESS (based on r200)
1380#define R300_PRIM_COLOR_ORDER_RGBA (1 << 6) // GUESS
1381#define R300_PRIM_NUM_VERTICES_SHIFT 16
1382
1383// Draw a primitive from vertex data in arrays loaded via 3D_LOAD_VBPNTR.
1384// Two parameter dwords:
1385// 0. The first parameter appears to be always 0
1386// 1. The second parameter is a standard primitive emission dword.
1387#define R300_PACKET3_3D_DRAW_VBUF 0x00002800
1388
1389// Specify the full set of vertex arrays as (address, stride).
1390// The first parameter is the number of vertex arrays specified.
1391// The rest of the command is a variable length list of blocks, where
1392// each block is three dwords long and specifies two arrays.
1393// The first dword of a block is split into two words, the lower significant
1394// word refers to the first array, the more significant word to the second
1395// array in the block.
1396// The low byte of each word contains the size of an array entry in dwords,
1397// the high byte contains the stride of the array.
1398// The second dword of a block contains the pointer to the first array,
1399// the third dword of a block contains the pointer to the second array.
1400// Note that if the total number of arrays is odd, the third dword of
1401// the last block is omitted.
1402#define R300_PACKET3_3D_LOAD_VBPNTR 0x00002F00
1403
1404#define R300_PACKET3_INDX_BUFFER 0x00003300
1405# define R300_EB_UNK1_SHIFT 24
1406# define R300_EB_UNK1 (0x80<<24)
1407# define R300_EB_UNK2 0x0810
1408#define R300_PACKET3_3D_DRAW_INDX_2 0x00003600
1409
1410//END
1411
1412#endif /* _R300_REG_H */
diff --git a/drivers/char/drm/radeon_cp.c b/drivers/char/drm/radeon_cp.c
index f24a27c4dd17..6d9080a3ca7e 100644
--- a/drivers/char/drm/radeon_cp.c
+++ b/drivers/char/drm/radeon_cp.c
@@ -32,6 +32,7 @@
32#include "drm.h" 32#include "drm.h"
33#include "radeon_drm.h" 33#include "radeon_drm.h"
34#include "radeon_drv.h" 34#include "radeon_drv.h"
35#include "r300_reg.h"
35 36
36#define RADEON_FIFO_DEBUG 0 37#define RADEON_FIFO_DEBUG 0
37 38
@@ -1151,6 +1152,8 @@ static void radeon_cp_init_ring_buffer( drm_device_t *dev,
1151 1152
1152#if __OS_HAS_AGP 1153#if __OS_HAS_AGP
1153 if ( !dev_priv->is_pci ) { 1154 if ( !dev_priv->is_pci ) {
1155 /* set RADEON_AGP_BASE here instead of relying on X from user space */
1156 RADEON_WRITE(RADEON_AGP_BASE, (unsigned int)dev->agp->base);
1154 RADEON_WRITE( RADEON_CP_RB_RPTR_ADDR, 1157 RADEON_WRITE( RADEON_CP_RB_RPTR_ADDR,
1155 dev_priv->ring_rptr->offset 1158 dev_priv->ring_rptr->offset
1156 - dev->agp->base 1159 - dev->agp->base
@@ -1626,6 +1629,9 @@ int radeon_cp_init( DRM_IOCTL_ARGS )
1626 1629
1627 DRM_COPY_FROM_USER_IOCTL( init, (drm_radeon_init_t __user *)data, sizeof(init) ); 1630 DRM_COPY_FROM_USER_IOCTL( init, (drm_radeon_init_t __user *)data, sizeof(init) );
1628 1631
1632 if(init.func == RADEON_INIT_R300_CP)
1633 r300_init_reg_flags();
1634
1629 switch ( init.func ) { 1635 switch ( init.func ) {
1630 case RADEON_INIT_CP: 1636 case RADEON_INIT_CP:
1631 case RADEON_INIT_R200_CP: 1637 case RADEON_INIT_R200_CP:
@@ -2040,12 +2046,19 @@ int radeon_driver_preinit(struct drm_device *dev, unsigned long flags)
2040 case CHIP_RV200: 2046 case CHIP_RV200:
2041 case CHIP_R200: 2047 case CHIP_R200:
2042 case CHIP_R300: 2048 case CHIP_R300:
2049 case CHIP_R420:
2043 dev_priv->flags |= CHIP_HAS_HIERZ; 2050 dev_priv->flags |= CHIP_HAS_HIERZ;
2044 break; 2051 break;
2045 default: 2052 default:
2046 /* all other chips have no hierarchical z buffer */ 2053 /* all other chips have no hierarchical z buffer */
2047 break; 2054 break;
2048 } 2055 }
2056
2057 if (drm_device_is_agp(dev))
2058 dev_priv->flags |= CHIP_IS_AGP;
2059
2060 DRM_DEBUG("%s card detected\n",
2061 ((dev_priv->flags & CHIP_IS_AGP) ? "AGP" : "PCI"));
2049 return ret; 2062 return ret;
2050} 2063}
2051 2064
diff --git a/drivers/char/drm/radeon_drm.h b/drivers/char/drm/radeon_drm.h
index c1e62d047989..3792798270a4 100644
--- a/drivers/char/drm/radeon_drm.h
+++ b/drivers/char/drm/radeon_drm.h
@@ -195,6 +195,52 @@ typedef union {
195#define RADEON_WAIT_2D 0x1 195#define RADEON_WAIT_2D 0x1
196#define RADEON_WAIT_3D 0x2 196#define RADEON_WAIT_3D 0x2
197 197
198/* Allowed parameters for R300_CMD_PACKET3
199 */
200#define R300_CMD_PACKET3_CLEAR 0
201#define R300_CMD_PACKET3_RAW 1
202
203/* Commands understood by cmd_buffer ioctl for R300.
204 * The interface has not been stabilized, so some of these may be removed
205 * and eventually reordered before stabilization.
206 */
207#define R300_CMD_PACKET0 1
208#define R300_CMD_VPU 2 /* emit vertex program upload */
209#define R300_CMD_PACKET3 3 /* emit a packet3 */
210#define R300_CMD_END3D 4 /* emit sequence ending 3d rendering */
211#define R300_CMD_CP_DELAY 5
212#define R300_CMD_DMA_DISCARD 6
213#define R300_CMD_WAIT 7
214# define R300_WAIT_2D 0x1
215# define R300_WAIT_3D 0x2
216# define R300_WAIT_2D_CLEAN 0x3
217# define R300_WAIT_3D_CLEAN 0x4
218
219typedef union {
220 unsigned int u;
221 struct {
222 unsigned char cmd_type, pad0, pad1, pad2;
223 } header;
224 struct {
225 unsigned char cmd_type, count, reglo, reghi;
226 } packet0;
227 struct {
228 unsigned char cmd_type, count, adrlo, adrhi;
229 } vpu;
230 struct {
231 unsigned char cmd_type, packet, pad0, pad1;
232 } packet3;
233 struct {
234 unsigned char cmd_type, packet;
235 unsigned short count; /* amount of packet2 to emit */
236 } delay;
237 struct {
238 unsigned char cmd_type, buf_idx, pad0, pad1;
239 } dma;
240 struct {
241 unsigned char cmd_type, flags, pad0, pad1;
242 } wait;
243} drm_r300_cmd_header_t;
198 244
199#define RADEON_FRONT 0x1 245#define RADEON_FRONT 0x1
200#define RADEON_BACK 0x2 246#define RADEON_BACK 0x2
diff --git a/drivers/char/drm/radeon_drv.h b/drivers/char/drm/radeon_drv.h
index e701dffe978d..f12a963ede18 100644
--- a/drivers/char/drm/radeon_drv.h
+++ b/drivers/char/drm/radeon_drv.h
@@ -82,9 +82,10 @@
82 * - Add support for r100 cube maps 82 * - Add support for r100 cube maps
83 * 1.16- Add R200_EMIT_PP_TRI_PERF_CNTL packet to support brilinear 83 * 1.16- Add R200_EMIT_PP_TRI_PERF_CNTL packet to support brilinear
84 * texture filtering on r200 84 * texture filtering on r200
85 * 1.17- Add initial support for R300 (3D).
85 */ 86 */
86#define DRIVER_MAJOR 1 87#define DRIVER_MAJOR 1
87#define DRIVER_MINOR 16 88#define DRIVER_MINOR 17
88#define DRIVER_PATCHLEVEL 0 89#define DRIVER_PATCHLEVEL 0
89 90
90#define GET_RING_HEAD(dev_priv) DRM_READ32( (dev_priv)->ring_rptr, 0 ) 91#define GET_RING_HEAD(dev_priv) DRM_READ32( (dev_priv)->ring_rptr, 0 )
@@ -106,7 +107,9 @@ enum radeon_family {
106 CHIP_RV280, 107 CHIP_RV280,
107 CHIP_R300, 108 CHIP_R300,
108 CHIP_RS300, 109 CHIP_RS300,
110 CHIP_R350,
109 CHIP_RV350, 111 CHIP_RV350,
112 CHIP_R420,
110 CHIP_LAST, 113 CHIP_LAST,
111}; 114};
112 115
@@ -321,6 +324,14 @@ extern int radeon_postcleanup( struct drm_device *dev );
321extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd, 324extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
322 unsigned long arg); 325 unsigned long arg);
323 326
327
328/* r300_cmdbuf.c */
329extern void r300_init_reg_flags(void);
330
331extern int r300_do_cp_cmdbuf(drm_device_t* dev, DRMFILE filp,
332 drm_file_t* filp_priv,
333 drm_radeon_cmd_buffer_t* cmdbuf);
334
324/* Flags for stats.boxes 335/* Flags for stats.boxes
325 */ 336 */
326#define RADEON_BOX_DMA_IDLE 0x1 337#define RADEON_BOX_DMA_IDLE 0x1
@@ -358,6 +369,11 @@ extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
358#define RADEON_CRTC2_OFFSET 0x0324 369#define RADEON_CRTC2_OFFSET 0x0324
359#define RADEON_CRTC2_OFFSET_CNTL 0x0328 370#define RADEON_CRTC2_OFFSET_CNTL 0x0328
360 371
372#define RADEON_MPP_TB_CONFIG 0x01c0
373#define RADEON_MEM_CNTL 0x0140
374#define RADEON_MEM_SDRAM_MODE_REG 0x0158
375#define RADEON_AGP_BASE 0x0170
376
361#define RADEON_RB3D_COLOROFFSET 0x1c40 377#define RADEON_RB3D_COLOROFFSET 0x1c40
362#define RADEON_RB3D_COLORPITCH 0x1c48 378#define RADEON_RB3D_COLORPITCH 0x1c48
363 379
@@ -652,16 +668,27 @@ extern long radeon_compat_ioctl(struct file *filp, unsigned int cmd,
652#define RADEON_CP_PACKET1 0x40000000 668#define RADEON_CP_PACKET1 0x40000000
653#define RADEON_CP_PACKET2 0x80000000 669#define RADEON_CP_PACKET2 0x80000000
654#define RADEON_CP_PACKET3 0xC0000000 670#define RADEON_CP_PACKET3 0xC0000000
671# define RADEON_CP_NOP 0x00001000
672# define RADEON_CP_NEXT_CHAR 0x00001900
673# define RADEON_CP_PLY_NEXTSCAN 0x00001D00
674# define RADEON_CP_SET_SCISSORS 0x00001E00
675 /* GEN_INDX_PRIM is unsupported starting with R300 */
655# define RADEON_3D_RNDR_GEN_INDX_PRIM 0x00002300 676# define RADEON_3D_RNDR_GEN_INDX_PRIM 0x00002300
656# define RADEON_WAIT_FOR_IDLE 0x00002600 677# define RADEON_WAIT_FOR_IDLE 0x00002600
657# define RADEON_3D_DRAW_VBUF 0x00002800 678# define RADEON_3D_DRAW_VBUF 0x00002800
658# define RADEON_3D_DRAW_IMMD 0x00002900 679# define RADEON_3D_DRAW_IMMD 0x00002900
659# define RADEON_3D_DRAW_INDX 0x00002A00 680# define RADEON_3D_DRAW_INDX 0x00002A00
681# define RADEON_CP_LOAD_PALETTE 0x00002C00
660# define RADEON_3D_LOAD_VBPNTR 0x00002F00 682# define RADEON_3D_LOAD_VBPNTR 0x00002F00
661# define RADEON_MPEG_IDCT_MACROBLOCK 0x00003000 683# define RADEON_MPEG_IDCT_MACROBLOCK 0x00003000
662# define RADEON_MPEG_IDCT_MACROBLOCK_REV 0x00003100 684# define RADEON_MPEG_IDCT_MACROBLOCK_REV 0x00003100
663# define RADEON_3D_CLEAR_ZMASK 0x00003200 685# define RADEON_3D_CLEAR_ZMASK 0x00003200
686# define RADEON_CP_INDX_BUFFER 0x00003300
687# define RADEON_CP_3D_DRAW_VBUF_2 0x00003400
688# define RADEON_CP_3D_DRAW_IMMD_2 0x00003500
689# define RADEON_CP_3D_DRAW_INDX_2 0x00003600
664# define RADEON_3D_CLEAR_HIZ 0x00003700 690# define RADEON_3D_CLEAR_HIZ 0x00003700
691# define RADEON_CP_3D_CLEAR_CMASK 0x00003802
665# define RADEON_CNTL_HOSTDATA_BLT 0x00009400 692# define RADEON_CNTL_HOSTDATA_BLT 0x00009400
666# define RADEON_CNTL_PAINT_MULTI 0x00009A00 693# define RADEON_CNTL_PAINT_MULTI 0x00009A00
667# define RADEON_CNTL_BITBLT_MULTI 0x00009B00 694# define RADEON_CNTL_BITBLT_MULTI 0x00009B00
diff --git a/drivers/char/drm/radeon_state.c b/drivers/char/drm/radeon_state.c
index 1f79e249146c..d57accdd8df5 100644
--- a/drivers/char/drm/radeon_state.c
+++ b/drivers/char/drm/radeon_state.c
@@ -2797,6 +2797,17 @@ static int radeon_cp_cmdbuf( DRM_IOCTL_ARGS )
2797 2797
2798 orig_nbox = cmdbuf.nbox; 2798 orig_nbox = cmdbuf.nbox;
2799 2799
2800 if(dev_priv->microcode_version == UCODE_R300) {
2801 int temp;
2802 temp=r300_do_cp_cmdbuf(dev, filp, filp_priv, &cmdbuf);
2803
2804 if (orig_bufsz != 0)
2805 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2806
2807 return temp;
2808 }
2809
2810 /* microcode_version != r300 */
2800 while ( cmdbuf.bufsz >= sizeof(header) ) { 2811 while ( cmdbuf.bufsz >= sizeof(header) ) {
2801 2812
2802 header.i = *(int *)cmdbuf.buf; 2813 header.i = *(int *)cmdbuf.buf;