1 /**************************************************************************
3 Copyright (C) 2004 Nicolai Haehnle.
7 Permission is hereby granted, free of charge, to any person obtaining a
8 copy of this software and associated documentation files (the "Software"),
9 to deal in the Software without restriction, including without limitation
10 on the rights to use, copy, modify, merge, publish, distribute, sub
11 license, and/or sell copies of the Software, and to permit persons to whom
12 the Software is furnished to do so, subject to the following conditions:
14 The above copyright notice and this permission notice (including the next
15 paragraph) shall be included in all copies or substantial portions of the
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
31 * \brief R300 Render (Vertex Buffer Implementation)
33 * The immediate implementation has been removed from CVS in favor of the vertex
34 * buffer implementation.
36 * The render functions are called by the pipeline manager to render a batch of
37 * primitives. They return TRUE to pass on to the next stage (i.e. software
38 * rasterization) or FALSE to indicate that the pipeline has finished after
39 * rendering something.
41 * When falling back to software TCL still attempt to use hardware
44 * I am not sure that the cache related registers are setup correctly, but
45 * obviously this does work... Further investigation is needed.
47 * \author Nicolai Haehnle <prefect_@gmx.net>
49 * \todo Add immediate implementation back? Perhaps this is useful if there are
53 #include "main/glheader.h"
54 #include "main/state.h"
55 #include "main/imports.h"
56 #include "main/enums.h"
57 #include "main/macros.h"
58 #include "main/context.h"
60 #include "main/simple_list.h"
61 #include "main/api_arrayelt.h"
62 #include "swrast/swrast.h"
63 #include "swrast_setup/swrast_setup.h"
66 #include "tnl/t_vp_build.h"
67 #include "radeon_reg.h"
68 #include "radeon_macros.h"
69 #include "radeon_ioctl.h"
70 #include "radeon_state.h"
71 #include "r300_context.h"
72 #include "r300_ioctl.h"
73 #include "r300_state.h"
76 #include "r300_emit.h"
77 #include "r300_fragprog.h"
78 extern int future_hw_tcl_on
;
81 * \brief Convert a OpenGL primitive type into a R300 primitive type.
83 int r300PrimitiveType(r300ContextPtr rmesa
, int prim
)
85 switch (prim
& PRIM_MODE_MASK
) {
87 return R300_VAP_VF_CNTL__PRIM_POINTS
;
90 return R300_VAP_VF_CNTL__PRIM_LINES
;
93 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP
;
96 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP
;
99 return R300_VAP_VF_CNTL__PRIM_TRIANGLES
;
101 case GL_TRIANGLE_STRIP
:
102 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP
;
104 case GL_TRIANGLE_FAN
:
105 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN
;
108 return R300_VAP_VF_CNTL__PRIM_QUADS
;
111 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP
;
114 return R300_VAP_VF_CNTL__PRIM_POLYGON
;
123 int r300NumVerts(r300ContextPtr rmesa
, int num_verts
, int prim
)
127 switch (prim
& PRIM_MODE_MASK
) {
132 verts_off
= num_verts
% 2;
136 verts_off
= num_verts
;
140 verts_off
= num_verts
;
143 verts_off
= num_verts
% 3;
145 case GL_TRIANGLE_STRIP
:
147 verts_off
= num_verts
;
149 case GL_TRIANGLE_FAN
:
151 verts_off
= num_verts
;
154 verts_off
= num_verts
% 4;
158 verts_off
= num_verts
;
160 verts_off
= num_verts
% 2;
164 verts_off
= num_verts
;
172 return num_verts
- verts_off
;
175 static void r300EmitElts(GLcontext
* ctx
, void *elts
, unsigned long n_elts
)
177 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
180 rmesa
->state
.elt_dma_bo
= radeon_bo_open(rmesa
->radeon
.radeonScreen
->bom
,
182 RADEON_GEM_DOMAIN_GTT
);
183 rmesa
->state
.elt_dma_offset
= 0;
184 radeon_bo_map(rmesa
->state
.elt_dma_bo
, 1);
185 out
= rmesa
->state
.elt_dma_bo
->ptr
+ rmesa
->state
.elt_dma_offset
;
186 memcpy(out
, elts
, n_elts
* 4);
187 radeon_bo_unmap(rmesa
->state
.elt_dma_bo
);
190 static void r300FireEB(r300ContextPtr rmesa
, int vertex_count
, int type
)
194 if (vertex_count
> 0) {
196 OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2
, 0);
197 OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
|
198 ((vertex_count
+ 0) << 16) |
200 R300_VAP_VF_CNTL__INDEX_SIZE_32bit
);
202 OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER
, 2);
203 OUT_BATCH(R300_EB_UNK1
| (0 << 16) | R300_EB_UNK2
);
204 OUT_BATCH_RELOC(0, rmesa
->state
.elt_dma_bo
,
205 rmesa
->state
.elt_dma_offset
, 0);
206 OUT_BATCH(vertex_count
);
211 static void r300EmitAOS(r300ContextPtr rmesa
, GLuint nr
, GLuint offset
)
215 int sz
= 1 + (nr
>> 1) * 3 + (nr
& 1) * 2;
218 if (RADEON_DEBUG
& DEBUG_VERTS
)
219 fprintf(stderr
, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__
, nr
,
223 OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR
, sz
- 1);
226 for (i
= 0; i
+ 1 < nr
; i
+= 2) {
227 OUT_BATCH((rmesa
->state
.aos
[i
].components
<< 0) |
228 (rmesa
->state
.aos
[i
].stride
<< 8) |
229 (rmesa
->state
.aos
[i
+ 1].components
<< 16) |
230 (rmesa
->state
.aos
[i
+ 1].stride
<< 24));
232 voffset
= rmesa
->state
.aos
[i
+ 0].offset
+
233 offset
* 4 * rmesa
->state
.aos
[i
+ 0].stride
;
234 OUT_BATCH_RELOC(0, rmesa
->state
.aos
[i
].bo
, voffset
, 0);
235 voffset
= rmesa
->state
.aos
[i
+ 1].offset
+
236 offset
* 4 * rmesa
->state
.aos
[i
+ 1].stride
;
237 OUT_BATCH_RELOC(0, rmesa
->state
.aos
[i
+1].bo
, voffset
, 0);
241 OUT_BATCH((rmesa
->state
.aos
[nr
- 1].components
<< 0) |
242 (rmesa
->state
.aos
[nr
- 1].stride
<< 8));
243 OUT_BATCH_RELOC(0, rmesa
->state
.aos
[nr
- 1].bo
,
244 rmesa
->state
.aos
[nr
- 1].offset
+ offset
* 4 * rmesa
->state
.aos
[nr
- 1].stride
, 0);
249 static void r300FireAOS(r300ContextPtr rmesa
, int vertex_count
, int type
)
254 OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2
, 0);
255 OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST
| (vertex_count
<< 16) | type
);
259 static void r300RunRenderPrimitive(r300ContextPtr rmesa
, GLcontext
* ctx
,
260 int start
, int end
, int prim
)
264 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
265 struct vertex_buffer
*vb
= &tnl
->vb
;
267 type
= r300PrimitiveType(rmesa
, prim
);
268 num_verts
= r300NumVerts(rmesa
, end
- start
, prim
);
270 if (type
< 0 || num_verts
<= 0)
273 /* Make space for at least 64 dwords.
274 * This is supposed to ensure that we can get all rendering
275 * commands into a single command buffer.
277 r300EnsureCmdBufSpace(rmesa
, 64, __FUNCTION__
);
280 if (num_verts
> 65535) {
281 /* not implemented yet */
282 WARN_ONCE("Too many elts\n");
285 /* Note: The following is incorrect, but it's the best I can do
286 * without a major refactoring of how DMA memory is handled.
287 * The problem: Ensuring that both vertex arrays *and* index
288 * arrays are at the right position, and then ensuring that
289 * the LOAD_VBPNTR, DRAW_INDX and INDX_BUFFER packets are emitted
292 * So why is the following incorrect? Well, it seems like
293 * allocating the index array might actually evict the vertex
296 r300EmitElts(ctx
, vb
->Elts
, num_verts
);
297 r300EmitAOS(rmesa
, rmesa
->state
.aos_count
, start
);
298 r300FireEB(rmesa
, num_verts
, type
);
300 r300EmitAOS(rmesa
, rmesa
->state
.aos_count
, start
);
301 r300FireAOS(rmesa
, num_verts
, type
);
306 static GLboolean
r300RunRender(GLcontext
* ctx
,
307 struct tnl_pipeline_stage
*stage
)
309 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
311 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
312 struct vertex_buffer
*vb
= &tnl
->vb
;
314 if (RADEON_DEBUG
& DEBUG_PRIMS
)
315 fprintf(stderr
, "%s\n", __FUNCTION__
);
317 r300UpdateShaders(rmesa
);
318 if (r300EmitArrays(ctx
))
321 r300UpdateShaderStates(rmesa
);
323 r300EmitCacheFlush(rmesa
);
324 r300EmitState(rmesa
);
326 for (i
= 0; i
< vb
->PrimitiveCount
; i
++) {
327 GLuint prim
= _tnl_translate_prim(&vb
->Primitive
[i
]);
328 GLuint start
= vb
->Primitive
[i
].start
;
329 GLuint end
= vb
->Primitive
[i
].start
+ vb
->Primitive
[i
].count
;
330 r300RunRenderPrimitive(rmesa
, ctx
, start
, end
, prim
);
333 r300EmitCacheFlush(rmesa
);
335 r300ReleaseArrays(ctx
);
340 #define FALLBACK_IF(expr) \
343 if (1 || RADEON_DEBUG & DEBUG_FALLBACKS) \
344 WARN_ONCE("Software fallback:%s\n", \
346 return R300_FALLBACK_RAST; \
350 static int r300Fallback(GLcontext
* ctx
)
352 r300ContextPtr r300
= R300_CONTEXT(ctx
);
353 /* Do we need to use new-style shaders?
354 * Also is there a better way to do this? */
355 if (r300
->radeon
.radeonScreen
->chip_family
>= CHIP_FAMILY_RV515
) {
356 struct r500_fragment_program
*fp
= (struct r500_fragment_program
*)
357 (char *)ctx
->FragmentProgram
._Current
;
359 if (!fp
->translated
) {
360 r500TranslateFragmentShader(r300
, fp
);
361 FALLBACK_IF(!fp
->translated
);
365 struct r300_fragment_program
*fp
= (struct r300_fragment_program
*)
366 (char *)ctx
->FragmentProgram
._Current
;
368 if (!fp
->translated
) {
369 r300TranslateFragmentShader(r300
, fp
);
370 FALLBACK_IF(!fp
->translated
);
375 FALLBACK_IF(ctx
->RenderMode
!= GL_RENDER
);
377 FALLBACK_IF(ctx
->Stencil
._TestTwoSide
378 && (ctx
->Stencil
.Ref
[0] != ctx
->Stencil
.Ref
[1]
379 || ctx
->Stencil
.ValueMask
[0] !=
380 ctx
->Stencil
.ValueMask
[1]
381 || ctx
->Stencil
.WriteMask
[0] !=
382 ctx
->Stencil
.WriteMask
[1]));
384 if (ctx
->Extensions
.NV_point_sprite
|| ctx
->Extensions
.ARB_point_sprite
)
385 FALLBACK_IF(ctx
->Point
.PointSprite
);
387 if (!r300
->disable_lowimpact_fallback
) {
388 FALLBACK_IF(ctx
->Polygon
.StippleFlag
);
389 FALLBACK_IF(ctx
->Multisample
._Enabled
);
390 FALLBACK_IF(ctx
->Line
.StippleFlag
);
391 FALLBACK_IF(ctx
->Line
.SmoothFlag
);
392 FALLBACK_IF(ctx
->Point
.SmoothFlag
);
395 return R300_FALLBACK_NONE
;
398 static GLboolean
r300RunNonTCLRender(GLcontext
* ctx
,
399 struct tnl_pipeline_stage
*stage
)
401 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
403 if (RADEON_DEBUG
& DEBUG_PRIMS
)
404 fprintf(stderr
, "%s\n", __FUNCTION__
);
406 if (r300Fallback(ctx
) >= R300_FALLBACK_RAST
)
409 if (!(rmesa
->radeon
.radeonScreen
->chip_flags
& RADEON_CHIPSET_TCL
))
412 return r300RunRender(ctx
, stage
);
415 static GLboolean
r300RunTCLRender(GLcontext
* ctx
,
416 struct tnl_pipeline_stage
*stage
)
418 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
419 struct r300_vertex_program
*vp
;
421 hw_tcl_on
= future_hw_tcl_on
;
423 if (RADEON_DEBUG
& DEBUG_PRIMS
)
424 fprintf(stderr
, "%s\n", __FUNCTION__
);
426 if (hw_tcl_on
== GL_FALSE
)
429 if (r300Fallback(ctx
) >= R300_FALLBACK_TCL
) {
430 hw_tcl_on
= GL_FALSE
;
434 r300UpdateShaders(rmesa
);
436 vp
= (struct r300_vertex_program
*)CURRENT_VERTEX_SHADER(ctx
);
437 if (vp
->native
== GL_FALSE
) {
438 hw_tcl_on
= GL_FALSE
;
442 return r300RunRender(ctx
, stage
);
445 const struct tnl_pipeline_stage _r300_render_stage
= {
446 "r300 Hardware Rasterization",
454 const struct tnl_pipeline_stage _r300_tcl_stage
= {
455 "r300 Hardware Transform, Clipping and Lighting",