i965: initial attempt at fixing the aperture overflow
[mesa.git] / src / mesa / drivers / dri / intel / intel_pixel_draw.c
1 /**************************************************************************
2 *
3 * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portionsalloc
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "glheader.h"
29 #include "enums.h"
30 #include "image.h"
31 #include "mtypes.h"
32 #include "macros.h"
33 #include "bufferobj.h"
34 #include "swrast/swrast.h"
35
36 #include "intel_screen.h"
37 #include "intel_context.h"
38 #include "intel_ioctl.h"
39 #include "intel_batchbuffer.h"
40 #include "intel_blit.h"
41 #include "intel_buffers.h"
42 #include "intel_regions.h"
43 #include "intel_pixel.h"
44 #include "intel_buffer_objects.h"
45 #include "intel_tris.h"
46
47
48
49 static GLboolean
50 do_texture_drawpixels(GLcontext * ctx,
51 GLint x, GLint y,
52 GLsizei width, GLsizei height,
53 GLenum format, GLenum type,
54 const struct gl_pixelstore_attrib *unpack,
55 const GLvoid * pixels)
56 {
57 struct intel_context *intel = intel_context(ctx);
58 struct intel_region *dst = intel_drawbuf_region(intel);
59 struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
60 GLuint rowLength = unpack->RowLength ? unpack->RowLength : width;
61 GLuint src_offset;
62
63 if (INTEL_DEBUG & DEBUG_PIXEL)
64 fprintf(stderr, "%s\n", __FUNCTION__);
65
66 intelFlush(&intel->ctx);
67
68 if (!dst)
69 return GL_FALSE;
70
71 intel->vtbl.render_start(intel);
72 intel->vtbl.emit_state(intel);
73
74 if (src) {
75 if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
76 format, type, pixels)) {
77 _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
78 return GL_TRUE;
79 }
80 }
81 else {
82 /* PBO only for now:
83 */
84 /* _mesa_printf("%s - not PBO\n", __FUNCTION__); */
85 return GL_FALSE;
86 }
87
88 /* There are a couple of things we can't do yet, one of which is
89 * set the correct state for pixel operations when GL texturing is
90 * enabled. That's a pretty rare state and probably not worth the
91 * effort. A completely device-independent version of this may do
92 * more.
93 *
94 * Similarly, we make no attempt to merge metaops processing with
95 * an enabled fragment program, though it would certainly be
96 * possible.
97 */
98 if (!intel_check_meta_tex_fragment_ops(ctx)) {
99 if (INTEL_DEBUG & DEBUG_PIXEL)
100 _mesa_printf("%s - bad GL fragment state for metaops texture\n",
101 __FUNCTION__);
102 return GL_FALSE;
103 }
104
105 intel->vtbl.install_meta_state(intel);
106
107
108 /* Is this true? Also will need to turn depth testing on according
109 * to state:
110 */
111 intel->vtbl.meta_no_stencil_write(intel);
112 intel->vtbl.meta_no_depth_write(intel);
113
114 /* Set the 3d engine to draw into the destination region:
115 */
116 intel->vtbl.meta_draw_region(intel, dst, intel->depth_region);
117
118 intel->vtbl.meta_import_pixel_state(intel);
119
120 src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
121 format, type, 0, 0, 0);
122
123
124 /* Setup the pbo up as a rectangular texture, if possible.
125 *
126 * TODO: This is almost always possible if the i915 fragment
127 * program is adjusted to correctly swizzle the sampled colors.
128 * The major exception is any 24bit texture, like RGB888, for which
129 * there is no hardware support.
130 */
131 if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, src_offset,
132 rowLength, height, format, type)) {
133 intel->vtbl.leave_meta_state(intel);
134 return GL_FALSE;
135 }
136
137 intel->vtbl.meta_texture_blend_replace(intel);
138
139
140 LOCK_HARDWARE(intel);
141
142 if (intel->driDrawable->numClipRects) {
143 __DRIdrawablePrivate *dPriv = intel->driDrawable;
144 GLint srcx, srcy;
145 GLint dstx, dsty;
146
147 dstx = x;
148 dsty = dPriv->h - (y + height);
149
150 srcx = 0; /* skiprows/pixels already done */
151 srcy = 0;
152
153 if (0) {
154 const GLint orig_x = dstx;
155 const GLint orig_y = dsty;
156
157 if (!_mesa_clip_to_region(0, 0, dst->pitch, dst->height,
158 &dstx, &dsty, &width, &height))
159 goto out;
160
161 srcx += dstx - orig_x;
162 srcy += dsty - orig_y;
163 }
164
165
166 if (INTEL_DEBUG & DEBUG_PIXEL)
167 _mesa_printf("draw %d,%d %dx%d\n", dstx, dsty, width, height);
168
169 /* Must use the regular cliprect mechanism in order to get the
170 * drawing origin set correctly. Otherwise scissor state is in
171 * incorrect coordinate space. Does this even need to hold the
172 * lock???
173 */
174 intel->vtbl.meta_draw_quad(intel,
175 dstx, dstx + width * ctx->Pixel.ZoomX,
176 dPriv->h - (y + height * ctx->Pixel.ZoomY),
177 dPriv->h - (y),
178 -ctx->Current.RasterPos[2] * .5,
179 0x00ff00ff,
180 srcx, srcx + width, srcy + height, srcy);
181 out:
182 intel->vtbl.leave_meta_state(intel);
183 intel_batchbuffer_flush(intel->batch);
184 }
185 UNLOCK_HARDWARE(intel);
186 return GL_TRUE;
187 }
188
189
190
191
192
193 /* Pros:
194 * - no waiting for idle before updating framebuffer.
195 *
196 * Cons:
197 * - if upload is by memcpy, this may actually be slower than fallback path.
198 * - uploads the whole image even if destination is clipped
199 *
200 * Need to benchmark.
201 *
202 * Given the questions about performance, implement for pbo's only.
203 * This path is definitely a win if the pbo is already in agp. If it
204 * turns out otherwise, we can add the code necessary to upload client
205 * data to agp space before performing the blit. (Though it may turn
206 * out to be better/simpler just to use the texture engine).
207 */
208 static GLboolean
209 do_blit_drawpixels(GLcontext * ctx,
210 GLint x, GLint y,
211 GLsizei width, GLsizei height,
212 GLenum format, GLenum type,
213 const struct gl_pixelstore_attrib *unpack,
214 const GLvoid * pixels)
215 {
216 struct intel_context *intel = intel_context(ctx);
217 struct intel_region *dest = intel_drawbuf_region(intel);
218 struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
219 GLuint src_offset;
220 GLuint rowLength;
221 dri_fence *fence = NULL;
222
223 if (INTEL_DEBUG & DEBUG_PIXEL)
224 _mesa_printf("%s\n", __FUNCTION__);
225
226
227 if (!dest) {
228 if (INTEL_DEBUG & DEBUG_PIXEL)
229 _mesa_printf("%s - no dest\n", __FUNCTION__);
230 return GL_FALSE;
231 }
232
233 if (src) {
234 /* This validation should be done by core mesa:
235 */
236 if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
237 format, type, pixels)) {
238 _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
239 return GL_TRUE;
240 }
241 }
242 else {
243 /* PBO only for now:
244 */
245 if (INTEL_DEBUG & DEBUG_PIXEL)
246 _mesa_printf("%s - not PBO\n", __FUNCTION__);
247 return GL_FALSE;
248 }
249
250 if (!intel_check_blit_format(dest, format, type)) {
251 if (INTEL_DEBUG & DEBUG_PIXEL)
252 _mesa_printf("%s - bad format for blit\n", __FUNCTION__);
253 return GL_FALSE;
254 }
255
256 if (!intel_check_blit_fragment_ops(ctx)) {
257 if (INTEL_DEBUG & DEBUG_PIXEL)
258 _mesa_printf("%s - bad GL fragment state for blitter\n",
259 __FUNCTION__);
260 return GL_FALSE;
261 }
262
263 if (ctx->Pixel.ZoomX != 1.0F) {
264 if (INTEL_DEBUG & DEBUG_PIXEL)
265 _mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__);
266 return GL_FALSE;
267 }
268
269
270 if (unpack->RowLength > 0)
271 rowLength = unpack->RowLength;
272 else
273 rowLength = width;
274
275 if (ctx->Pixel.ZoomY == -1.0F) {
276 if (INTEL_DEBUG & DEBUG_PIXEL)
277 _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
278 return GL_FALSE; /* later */
279 y -= height;
280 }
281 else if (ctx->Pixel.ZoomY == 1.0F) {
282 rowLength = -rowLength;
283 }
284 else {
285 if (INTEL_DEBUG & DEBUG_PIXEL)
286 _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
287 return GL_FALSE;
288 }
289
290 src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
291 format, type, 0, 0, 0);
292
293 intelFlush(&intel->ctx);
294 LOCK_HARDWARE(intel);
295
296 if (intel->driDrawable->numClipRects) {
297 __DRIdrawablePrivate *dPriv = intel->driDrawable;
298 int nbox = dPriv->numClipRects;
299 drm_clip_rect_t *box = dPriv->pClipRects;
300 drm_clip_rect_t rect;
301 drm_clip_rect_t dest_rect;
302 dri_bo *src_buffer = intel_bufferobj_buffer(intel, src, INTEL_READ);
303 int i;
304
305 dest_rect.x1 = dPriv->x + x;
306 dest_rect.y1 = dPriv->y + dPriv->h - (y + height);
307 dest_rect.x2 = dest_rect.x1 + width;
308 dest_rect.y2 = dest_rect.y1 + height;
309
310 for (i = 0; i < nbox; i++) {
311 if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i]))
312 continue;
313
314 intelEmitCopyBlit(intel,
315 dest->cpp,
316 rowLength, src_buffer, src_offset, GL_FALSE,
317 dest->pitch, dest->buffer, 0, dest->tiled,
318 rect.x1 - dest_rect.x1,
319 rect.y2 - dest_rect.y2,
320 rect.x1,
321 rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1,
322 ctx->Color.ColorLogicOpEnabled ?
323 ctx->Color.LogicOp : GL_COPY);
324 }
325 intel_batchbuffer_flush(intel->batch);
326 fence = intel->batch->last_fence;
327 dri_fence_reference(fence);
328 }
329 UNLOCK_HARDWARE(intel);
330
331 if (fence) {
332 dri_fence_wait(fence);
333 dri_fence_unreference(fence);
334 }
335
336 if (INTEL_DEBUG & DEBUG_PIXEL)
337 _mesa_printf("%s - DONE\n", __FUNCTION__);
338
339 return GL_TRUE;
340 }
341
342
343
344 void
345 intelDrawPixels(GLcontext * ctx,
346 GLint x, GLint y,
347 GLsizei width, GLsizei height,
348 GLenum format,
349 GLenum type,
350 const struct gl_pixelstore_attrib *unpack,
351 const GLvoid * pixels)
352 {
353 if (do_blit_drawpixels(ctx, x, y, width, height, format, type,
354 unpack, pixels))
355 return;
356
357 if (do_texture_drawpixels(ctx, x, y, width, height, format, type,
358 unpack, pixels))
359 return;
360
361
362 if (INTEL_DEBUG & DEBUG_PIXEL)
363 _mesa_printf("%s: fallback to swrast\n", __FUNCTION__);
364
365 if (ctx->FragmentProgram._Current == ctx->FragmentProgram._TexEnvProgram) {
366 /*
367 * We don't want the i915 texenv program to be applied to DrawPixels.
368 * This is really just a performance optimization (mesa will other-
369 * wise happily run the fragment program on each pixel in the image).
370 */
371 struct gl_fragment_program *fpSave = ctx->FragmentProgram._Current;
372 /* can't just set current frag prog to 0 here as on buffer resize
373 we'll get new state checks which will segfault. Remains a hack. */
374 ctx->FragmentProgram._Current = NULL;
375 ctx->FragmentProgram._UseTexEnvProgram = GL_FALSE;
376 ctx->FragmentProgram._Active = GL_FALSE;
377 _swrast_DrawPixels( ctx, x, y, width, height, format, type,
378 unpack, pixels );
379 ctx->FragmentProgram._Current = fpSave;
380 ctx->FragmentProgram._UseTexEnvProgram = GL_TRUE;
381 ctx->FragmentProgram._Active = GL_TRUE;
382 _swrast_InvalidateState(ctx, _NEW_PROGRAM);
383 }
384 else {
385 _swrast_DrawPixels( ctx, x, y, width, height, format, type,
386 unpack, pixels );
387 }
388 }