bumpmap sample is correct now
[mesa.git] / src / mesa / drivers / dri / i915tex / 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 intel->vtbl.render_start(intel);
68 intel->vtbl.emit_state(intel);
69
70 if (!dst)
71 return GL_FALSE;
72
73 if (src) {
74 if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
75 format, type, pixels)) {
76 _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
77 return GL_TRUE;
78 }
79 }
80 else {
81 /* PBO only for now:
82 */
83 /* _mesa_printf("%s - not PBO\n", __FUNCTION__); */
84 return GL_FALSE;
85 }
86
87 /* There are a couple of things we can't do yet, one of which is
88 * set the correct state for pixel operations when GL texturing is
89 * enabled. That's a pretty rare state and probably not worth the
90 * effort. A completely device-independent version of this may do
91 * more.
92 *
93 * Similarly, we make no attempt to merge metaops processing with
94 * an enabled fragment program, though it would certainly be
95 * possible.
96 */
97 if (!intel_check_meta_tex_fragment_ops(ctx)) {
98 if (INTEL_DEBUG & DEBUG_PIXEL)
99 _mesa_printf("%s - bad GL fragment state for metaops texture\n",
100 __FUNCTION__);
101 return GL_FALSE;
102 }
103
104 intel->vtbl.install_meta_state(intel);
105
106
107 /* Is this true? Also will need to turn depth testing on according
108 * to state:
109 */
110 intel->vtbl.meta_no_stencil_write(intel);
111 intel->vtbl.meta_no_depth_write(intel);
112
113 /* Set the 3d engine to draw into the destination region:
114 */
115 intel->vtbl.meta_draw_region(intel, dst, intel->intelScreen->depth_region);
116
117 intel->vtbl.meta_import_pixel_state(intel);
118
119 src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
120 format, type, 0, 0, 0);
121
122
123 /* Setup the pbo up as a rectangular texture, if possible.
124 *
125 * TODO: This is almost always possible if the i915 fragment
126 * program is adjusted to correctly swizzle the sampled colors.
127 * The major exception is any 24bit texture, like RGB888, for which
128 * there is no hardware support.
129 */
130 if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, src_offset,
131 rowLength, height, format, type)) {
132 intel->vtbl.leave_meta_state(intel);
133 return GL_FALSE;
134 }
135
136 intel->vtbl.meta_texture_blend_replace(intel);
137
138
139 LOCK_HARDWARE(intel);
140
141 if (intel->driDrawable->numClipRects) {
142 __DRIdrawablePrivate *dPriv = intel->driDrawable;
143 GLint srcx, srcy;
144 GLint dstx, dsty;
145
146 dstx = x;
147 dsty = dPriv->h - (y + height);
148
149 srcx = 0; /* skiprows/pixels already done */
150 srcy = 0;
151
152 if (0) {
153 const GLint orig_x = dstx;
154 const GLint orig_y = dsty;
155
156 if (!_mesa_clip_to_region(0, 0, dst->pitch, dst->height,
157 &dstx, &dsty, &width, &height))
158 goto out;
159
160 srcx += dstx - orig_x;
161 srcy += dsty - orig_y;
162 }
163
164
165 if (INTEL_DEBUG & DEBUG_PIXEL)
166 _mesa_printf("draw %d,%d %dx%d\n", dstx, dsty, width, height);
167
168 /* Must use the regular cliprect mechanism in order to get the
169 * drawing origin set correctly. Otherwise scissor state is in
170 * incorrect coordinate space. Does this even need to hold the
171 * lock???
172 */
173 intel_meta_draw_quad(intel,
174 dstx, dstx + width * ctx->Pixel.ZoomX,
175 dPriv->h - (y + height * ctx->Pixel.ZoomY),
176 dPriv->h - (y),
177 -ctx->Current.RasterPos[2] * .5,
178 0x00ff00ff,
179 srcx, srcx + width, srcy + height, srcy);
180 out:
181 intel->vtbl.leave_meta_state(intel);
182 intel_batchbuffer_flush(intel->batch);
183 }
184 UNLOCK_HARDWARE(intel);
185 return GL_TRUE;
186 }
187
188
189
190
191
192 /* Pros:
193 * - no waiting for idle before updating framebuffer.
194 *
195 * Cons:
196 * - if upload is by memcpy, this may actually be slower than fallback path.
197 * - uploads the whole image even if destination is clipped
198 *
199 * Need to benchmark.
200 *
201 * Given the questions about performance, implement for pbo's only.
202 * This path is definitely a win if the pbo is already in agp. If it
203 * turns out otherwise, we can add the code necessary to upload client
204 * data to agp space before performing the blit. (Though it may turn
205 * out to be better/simpler just to use the texture engine).
206 */
207 static GLboolean
208 do_blit_drawpixels(GLcontext * ctx,
209 GLint x, GLint y,
210 GLsizei width, GLsizei height,
211 GLenum format, GLenum type,
212 const struct gl_pixelstore_attrib *unpack,
213 const GLvoid * pixels)
214 {
215 struct intel_context *intel = intel_context(ctx);
216 struct intel_region *dest = intel_drawbuf_region(intel);
217 struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
218 GLuint src_offset;
219 GLuint rowLength;
220 struct _DriFenceObject *fence = NULL;
221
222 if (INTEL_DEBUG & DEBUG_PIXEL)
223 _mesa_printf("%s\n", __FUNCTION__);
224
225
226 if (!dest) {
227 if (INTEL_DEBUG & DEBUG_PIXEL)
228 _mesa_printf("%s - no dest\n", __FUNCTION__);
229 return GL_FALSE;
230 }
231
232 if (src) {
233 /* This validation should be done by core mesa:
234 */
235 if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
236 format, type, pixels)) {
237 _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
238 return GL_TRUE;
239 }
240 }
241 else {
242 /* PBO only for now:
243 */
244 if (INTEL_DEBUG & DEBUG_PIXEL)
245 _mesa_printf("%s - not PBO\n", __FUNCTION__);
246 return GL_FALSE;
247 }
248
249 if (!intel_check_blit_format(dest, format, type)) {
250 if (INTEL_DEBUG & DEBUG_PIXEL)
251 _mesa_printf("%s - bad format for blit\n", __FUNCTION__);
252 return GL_FALSE;
253 }
254
255 if (!intel_check_blit_fragment_ops(ctx)) {
256 if (INTEL_DEBUG & DEBUG_PIXEL)
257 _mesa_printf("%s - bad GL fragment state for blitter\n",
258 __FUNCTION__);
259 return GL_FALSE;
260 }
261
262 if (ctx->Pixel.ZoomX != 1.0F) {
263 if (INTEL_DEBUG & DEBUG_PIXEL)
264 _mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__);
265 return GL_FALSE;
266 }
267
268
269 if (unpack->RowLength > 0)
270 rowLength = unpack->RowLength;
271 else
272 rowLength = width;
273
274 if (ctx->Pixel.ZoomY == -1.0F) {
275 if (INTEL_DEBUG & DEBUG_PIXEL)
276 _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
277 return GL_FALSE; /* later */
278 y -= height;
279 }
280 else if (ctx->Pixel.ZoomY == 1.0F) {
281 rowLength = -rowLength;
282 }
283 else {
284 if (INTEL_DEBUG & DEBUG_PIXEL)
285 _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
286 return GL_FALSE;
287 }
288
289 src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
290 format, type, 0, 0, 0);
291
292 intelFlush(&intel->ctx);
293 LOCK_HARDWARE(intel);
294
295 if (intel->driDrawable->numClipRects) {
296 __DRIdrawablePrivate *dPriv = intel->driDrawable;
297 int nbox = dPriv->numClipRects;
298 drm_clip_rect_t *box = dPriv->pClipRects;
299 drm_clip_rect_t rect;
300 drm_clip_rect_t dest_rect;
301 struct _DriBufferObject *src_buffer =
302 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,
317 src_buffer, src_offset,
318 dest->pitch,
319 dest->buffer, 0,
320 rect.x1 - dest_rect.x1,
321 rect.y2 - dest_rect.y2,
322 rect.x1,
323 rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1,
324 ctx->Color.ColorLogicOpEnabled ?
325 ctx->Color.LogicOp : GL_COPY);
326 }
327 fence = intel_batchbuffer_flush(intel->batch);
328 driFenceReference(fence);
329 }
330 UNLOCK_HARDWARE(intel);
331
332 if (fence) {
333 driFenceFinish(fence, DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, GL_FALSE);
334 driFenceUnReference(fence);
335 }
336
337 if (INTEL_DEBUG & DEBUG_PIXEL)
338 _mesa_printf("%s - DONE\n", __FUNCTION__);
339
340 return GL_TRUE;
341 }
342
343
344
345 void
346 intelDrawPixels(GLcontext * ctx,
347 GLint x, GLint y,
348 GLsizei width, GLsizei height,
349 GLenum format,
350 GLenum type,
351 const struct gl_pixelstore_attrib *unpack,
352 const GLvoid * pixels)
353 {
354 if (do_blit_drawpixels(ctx, x, y, width, height, format, type,
355 unpack, pixels))
356 return;
357
358 if (do_texture_drawpixels(ctx, x, y, width, height, format, type,
359 unpack, pixels))
360 return;
361
362
363 if (INTEL_DEBUG & DEBUG_PIXEL)
364 _mesa_printf("%s: fallback to swrast\n", __FUNCTION__);
365
366 if (ctx->FragmentProgram._Current == ctx->FragmentProgram._TexEnvProgram) {
367 /*
368 * We don't want the i915 texenv program to be applied to DrawPixels.
369 * This is really just a performance optimization (mesa will other-
370 * wise happily run the fragment program on each pixel in the image).
371 */
372 struct gl_fragment_program *fpSave = ctx->FragmentProgram._Current;
373 ctx->FragmentProgram._Current = NULL;
374 _swrast_DrawPixels( ctx, x, y, width, height, format, type,
375 unpack, pixels );
376 ctx->FragmentProgram._Current = fpSave;
377 }
378 else {
379 _swrast_DrawPixels( ctx, x, y, width, height, format, type,
380 unpack, pixels );
381 }
382 }