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
);
178 struct r300_dma_region
*rvb
= &rmesa
->state
.elt_dma
;
181 if (r300IsGartMemory(rmesa
, elts
, n_elts
* 4)) {
182 rvb
->address
= rmesa
->radeon
.radeonScreen
->gartTextures
.map
;
183 rvb
->start
= ((char *)elts
) - rvb
->address
;
185 rmesa
->radeon
.radeonScreen
->gart_texture_offset
+
188 } else if (r300IsGartMemory(rmesa
, elts
, 1)) {
189 WARN_ONCE("Pointer not within GART memory!\n");
193 r300AllocDmaRegion(rmesa
, rvb
, n_elts
* 4, 4);
194 rvb
->aos_offset
= GET_START(rvb
);
196 out
= rvb
->address
+ rvb
->start
;
197 memcpy(out
, elts
, n_elts
* 4);
200 static void r300FireEB(r300ContextPtr rmesa
, unsigned long addr
,
201 int vertex_count
, int type
)
203 int cmd_reserved
= 0;
205 drm_radeon_cmd_header_t
*cmd
= NULL
;
207 start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_INDX_2
, 0), 0);
208 e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES
| (vertex_count
<< 16) | type
| R300_VAP_VF_CNTL__INDEX_SIZE_32bit
);
210 start_packet3(CP_PACKET3(R300_PACKET3_INDX_BUFFER
, 2), 2);
211 e32(R300_INDX_BUFFER_ONE_REG_WR
| (0 << R300_INDX_BUFFER_SKIP_SHIFT
) |
212 (R300_VAP_PORT_IDX0
>> 2));
217 static void r300EmitAOS(r300ContextPtr rmesa
, GLuint nr
, GLuint offset
)
219 int sz
= 1 + (nr
>> 1) * 3 + (nr
& 1) * 2;
221 int cmd_reserved
= 0;
223 drm_radeon_cmd_header_t
*cmd
= NULL
;
225 if (RADEON_DEBUG
& DEBUG_VERTS
)
226 fprintf(stderr
, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__
, nr
,
229 start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR
, sz
- 1), sz
- 1);
232 for (i
= 0; i
+ 1 < nr
; i
+= 2) {
233 e32((rmesa
->state
.aos
[i
].aos_size
<< 0) |
234 (rmesa
->state
.aos
[i
].aos_stride
<< 8) |
235 (rmesa
->state
.aos
[i
+ 1].aos_size
<< 16) |
236 (rmesa
->state
.aos
[i
+ 1].aos_stride
<< 24));
238 e32(rmesa
->state
.aos
[i
].aos_offset
+ offset
* 4 * rmesa
->state
.aos
[i
].aos_stride
);
239 e32(rmesa
->state
.aos
[i
+ 1].aos_offset
+ offset
* 4 * rmesa
->state
.aos
[i
+ 1].aos_stride
);
243 e32((rmesa
->state
.aos
[nr
- 1].aos_size
<< 0) |
244 (rmesa
->state
.aos
[nr
- 1].aos_stride
<< 8));
245 e32(rmesa
->state
.aos
[nr
- 1].aos_offset
+ offset
* 4 * rmesa
->state
.aos
[nr
- 1].aos_stride
);
249 static void r300FireAOS(r300ContextPtr rmesa
, int vertex_count
, int type
)
251 int cmd_reserved
= 0;
253 drm_radeon_cmd_header_t
*cmd
= NULL
;
255 start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2
, 0), 0);
256 e32(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
)
263 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
264 struct vertex_buffer
*vb
= &tnl
->vb
;
266 type
= r300PrimitiveType(rmesa
, prim
);
267 num_verts
= r300NumVerts(rmesa
, end
- start
, prim
);
269 if (type
< 0 || num_verts
<= 0)
273 if (num_verts
> 65535) {
274 /* not implemented yet */
275 WARN_ONCE("Too many elts\n");
278 /* Note: The following is incorrect, but it's the best I can do
279 * without a major refactoring of how DMA memory is handled.
280 * The problem: Ensuring that both vertex arrays *and* index
281 * arrays are at the right position, and then ensuring that
282 * the LOAD_VBPNTR, DRAW_INDX and INDX_BUFFER packets are emitted
285 * So why is the following incorrect? Well, it seems like
286 * allocating the index array might actually evict the vertex
289 r300EmitElts(ctx
, vb
->Elts
, num_verts
);
290 r300EmitAOS(rmesa
, rmesa
->state
.aos_count
, start
);
291 r300FireEB(rmesa
, rmesa
->state
.elt_dma
.aos_offset
, num_verts
, type
);
293 r300EmitAOS(rmesa
, rmesa
->state
.aos_count
, start
);
294 r300FireAOS(rmesa
, num_verts
, type
);
298 static GLboolean
r300RunRender(GLcontext
* ctx
,
299 struct tnl_pipeline_stage
*stage
)
301 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
303 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
304 struct vertex_buffer
*vb
= &tnl
->vb
;
307 if (RADEON_DEBUG
& DEBUG_PRIMS
)
308 fprintf(stderr
, "%s\n", __FUNCTION__
);
310 r300UpdateShaders(rmesa
);
311 if (r300EmitArrays(ctx
))
314 r300UpdateShaderStates(rmesa
);
316 r300EmitCacheFlush(rmesa
);
317 r300EmitState(rmesa
);
319 for (i
= 0; i
< vb
->PrimitiveCount
; i
++) {
320 GLuint prim
= _tnl_translate_prim(&vb
->Primitive
[i
]);
321 GLuint start
= vb
->Primitive
[i
].start
;
322 GLuint end
= vb
->Primitive
[i
].start
+ vb
->Primitive
[i
].count
;
323 r300RunRenderPrimitive(rmesa
, ctx
, start
, end
, prim
);
326 r300EmitCacheFlush(rmesa
);
332 r300ReleaseArrays(ctx
);
337 #define FALLBACK_IF(expr) \
340 if (1 || RADEON_DEBUG & DEBUG_FALLBACKS) \
341 WARN_ONCE("Software fallback:%s\n", \
343 return R300_FALLBACK_RAST; \
347 static int r300Fallback(GLcontext
* ctx
)
349 r300ContextPtr r300
= R300_CONTEXT(ctx
);
350 const unsigned back
= ctx
->Stencil
._BackFace
;
352 /* Do we need to use new-style shaders?
353 * Also is there a better way to do this? */
354 if (r300
->radeon
.radeonScreen
->chip_family
>= CHIP_FAMILY_RV515
) {
355 struct r500_fragment_program
*fp
= (struct r500_fragment_program
*)
356 (char *)ctx
->FragmentProgram
._Current
;
358 if (!fp
->translated
) {
359 r500TranslateFragmentShader(r300
, fp
);
360 FALLBACK_IF(!fp
->translated
);
364 struct r300_fragment_program
*fp
= (struct r300_fragment_program
*)
365 (char *)ctx
->FragmentProgram
._Current
;
367 if (!fp
->translated
) {
368 r300TranslateFragmentShader(r300
, fp
);
369 FALLBACK_IF(!fp
->translated
);
374 FALLBACK_IF(ctx
->RenderMode
!= GL_RENDER
);
376 /* If GL_EXT_stencil_two_side is disabled, this fallback check can
379 FALLBACK_IF(ctx
->Stencil
.Ref
[0] != ctx
->Stencil
.Ref
[back
]
380 || ctx
->Stencil
.ValueMask
[0] !=
381 ctx
->Stencil
.ValueMask
[back
]
382 || ctx
->Stencil
.WriteMask
[0] !=
383 ctx
->Stencil
.WriteMask
[back
]);
385 if (ctx
->Extensions
.NV_point_sprite
|| ctx
->Extensions
.ARB_point_sprite
)
386 FALLBACK_IF(ctx
->Point
.PointSprite
);
388 if (!r300
->disable_lowimpact_fallback
) {
389 FALLBACK_IF(ctx
->Polygon
.StippleFlag
);
390 FALLBACK_IF(ctx
->Multisample
._Enabled
);
391 FALLBACK_IF(ctx
->Line
.StippleFlag
);
392 FALLBACK_IF(ctx
->Line
.SmoothFlag
);
393 FALLBACK_IF(ctx
->Point
.SmoothFlag
);
396 return R300_FALLBACK_NONE
;
399 static GLboolean
r300RunNonTCLRender(GLcontext
* ctx
,
400 struct tnl_pipeline_stage
*stage
)
402 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
404 if (RADEON_DEBUG
& DEBUG_PRIMS
)
405 fprintf(stderr
, "%s\n", __FUNCTION__
);
407 if (r300Fallback(ctx
) >= R300_FALLBACK_RAST
)
410 if (!(rmesa
->radeon
.radeonScreen
->chip_flags
& RADEON_CHIPSET_TCL
))
413 return r300RunRender(ctx
, stage
);
416 static GLboolean
r300RunTCLRender(GLcontext
* ctx
,
417 struct tnl_pipeline_stage
*stage
)
419 r300ContextPtr rmesa
= R300_CONTEXT(ctx
);
420 struct r300_vertex_program
*vp
;
422 hw_tcl_on
= future_hw_tcl_on
;
424 if (RADEON_DEBUG
& DEBUG_PRIMS
)
425 fprintf(stderr
, "%s\n", __FUNCTION__
);
427 if (hw_tcl_on
== GL_FALSE
)
430 if (r300Fallback(ctx
) >= R300_FALLBACK_TCL
) {
431 hw_tcl_on
= GL_FALSE
;
435 r300UpdateShaders(rmesa
);
437 vp
= (struct r300_vertex_program
*)CURRENT_VERTEX_SHADER(ctx
);
438 if (vp
->native
== GL_FALSE
) {
439 hw_tcl_on
= GL_FALSE
;
443 return r300RunRender(ctx
, stage
);
446 const struct tnl_pipeline_stage _r300_render_stage
= {
447 "r300 Hardware Rasterization",
455 const struct tnl_pipeline_stage _r300_tcl_stage
= {
456 "r300 Hardware Transform, Clipping and Lighting",