diff options
Diffstat (limited to 'src/mesa/drivers/dri/intel')
44 files changed, 15280 insertions, 12 deletions
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c new file mode 100644 index 0000000000..7dbc646370 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c @@ -0,0 +1,297 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_decode.h" +#include "intel_reg.h" +#include "intel_bufmgr.h" + +/* Relocations in kernel space: + * - pass dma buffer seperately + * - memory manager knows how to patch + * - pass list of dependent buffers + * - pass relocation list + * + * Either: + * - get back an offset for buffer to fire + * - memory manager knows how to fire buffer + * + * Really want the buffer to be AGP and pinned. + * + */ + +/* Cliprect fence: The highest fence protecting a dma buffer + * containing explicit cliprect information. Like the old drawable + * lock but irq-driven. X server must wait for this fence to expire + * before changing cliprects [and then doing sw rendering?]. For + * other dma buffers, the scheduler will grab current cliprect info + * and mix into buffer. X server must hold the lock while changing + * cliprects??? Make per-drawable. Need cliprects in shared memory + * -- beats storing them with every cmd buffer in the queue. + * + * ==> X server must wait for this fence to expire before touching the + * framebuffer with new cliprects. + * + * ==> Cliprect-dependent buffers associated with a + * cliprect-timestamp. All of the buffers associated with a timestamp + * must go to hardware before any buffer with a newer timestamp. + * + * ==> Dma should be queued per-drawable for correct X/GL + * synchronization. Or can fences be used for this? + * + * Applies to: Blit operations, metaops, X server operations -- X + * server automatically waits on its own dma to complete before + * modifying cliprects ??? + */ + +void +intel_batchbuffer_reset(struct intel_batchbuffer *batch) +{ + struct intel_context *intel = batch->intel; + + if (batch->buf != NULL) { + dri_bo_unreference(batch->buf); + batch->buf = NULL; + } + + if (!batch->buffer && intel->ttm == GL_TRUE) + batch->buffer = malloc (intel->maxBatchSize); + + batch->buf = dri_bo_alloc(intel->bufmgr, "batchbuffer", + intel->maxBatchSize, 4096); + if (batch->buffer) + batch->map = batch->buffer; + else { + dri_bo_map(batch->buf, GL_TRUE); + batch->map = batch->buf->virtual; + } + batch->size = intel->maxBatchSize; + batch->ptr = batch->map; + batch->dirty_state = ~0; + batch->cliprect_mode = IGNORE_CLIPRECTS; +} + +struct intel_batchbuffer * +intel_batchbuffer_alloc(struct intel_context *intel) +{ + struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1); + + batch->intel = intel; + intel_batchbuffer_reset(batch); + + return batch; +} + +void +intel_batchbuffer_free(struct intel_batchbuffer *batch) +{ + if (batch->buffer) + free (batch->buffer); + else { + if (batch->map) { + dri_bo_unmap(batch->buf); + batch->map = NULL; + } + } + dri_bo_unreference(batch->buf); + batch->buf = NULL; + free(batch); +} + + + +/* TODO: Push this whole function into bufmgr. + */ +static void +do_flush_locked(struct intel_batchbuffer *batch, + GLuint used, GLboolean allow_unlock) +{ + struct intel_context *intel = batch->intel; + int ret = 0; + + if (batch->buffer) + dri_bo_subdata (batch->buf, 0, used, batch->buffer); + else + dri_bo_unmap(batch->buf); + + batch->map = NULL; + batch->ptr = NULL; + + /* Throw away non-effective packets. Won't work once we have + * hardware contexts which would preserve statechanges beyond a + * single buffer. + */ + + if (!(intel->numClipRects == 0 && + batch->cliprect_mode == LOOP_CLIPRECTS) || intel->no_hw) { + dri_bo_exec(batch->buf, used, + intel->pClipRects, + batch->cliprect_mode != LOOP_CLIPRECTS ? + 0 : intel->numClipRects, + (((GLuint) intel->drawX) & 0xffff) | + (((GLuint) intel->drawY) << 16)); + } + + if (intel->numClipRects == 0 && + batch->cliprect_mode == LOOP_CLIPRECTS) { + if (allow_unlock) { + /* If we are not doing any actual user-visible rendering, + * do a sched_yield to keep the app from pegging the cpu while + * achieving nothing. + */ + UNLOCK_HARDWARE(intel); + sched_yield(); + LOCK_HARDWARE(intel); + } + } + + if (INTEL_DEBUG & DEBUG_BATCH) { + dri_bo_map(batch->buf, GL_FALSE); + intel_decode(batch->buf->virtual, used / 4, batch->buf->offset, + intel->intelScreen->deviceID); + dri_bo_unmap(batch->buf); + + if (intel->vtbl.debug_batch != NULL) + intel->vtbl.debug_batch(intel); + } + + if (ret != 0) { + UNLOCK_HARDWARE(intel); + exit(1); + } + intel->vtbl.new_batch(intel); +} + +void +_intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file, + int line) +{ + struct intel_context *intel = batch->intel; + GLuint used = batch->ptr - batch->map; + GLboolean was_locked = intel->locked; + + if (used == 0) + return; + + if (INTEL_DEBUG & DEBUG_BATCH) + fprintf(stderr, "%s:%d: Batchbuffer flush with %db used\n", file, line, + used); + + /* Emit a flush if the bufmgr doesn't do it for us. */ + if (!intel->ttm) { + *(GLuint *) (batch->ptr) = intel->vtbl.flush_cmd(); + batch->ptr += 4; + used = batch->ptr - batch->map; + } + + /* Round batchbuffer usage to 2 DWORDs. */ + + if ((used & 4) == 0) { + *(GLuint *) (batch->ptr) = 0; /* noop */ + batch->ptr += 4; + used = batch->ptr - batch->map; + } + + /* Mark the end of the buffer. */ + *(GLuint *) (batch->ptr) = MI_BATCH_BUFFER_END; /* noop */ + batch->ptr += 4; + used = batch->ptr - batch->map; + + /* Workaround for recursive batchbuffer flushing: If the window is + * moved, we can get into a case where we try to flush during a + * flush. What happens is that when we try to grab the lock for + * the first flush, we detect that the window moved which then + * causes another flush (from the intel_draw_buffer() call in + * intelUpdatePageFlipping()). To work around this we reset the + * batchbuffer tail pointer before trying to get the lock. This + * prevent the nested buffer flush, but a better fix would be to + * avoid that in the first place. */ + batch->ptr = batch->map; + + if (intel->vtbl.finish_batch) + intel->vtbl.finish_batch(intel); + + /* TODO: Just pass the relocation list and dma buffer up to the + * kernel. + */ + if (!was_locked) + LOCK_HARDWARE(intel); + + do_flush_locked(batch, used, GL_FALSE); + + if (!was_locked) + UNLOCK_HARDWARE(intel); + + if (INTEL_DEBUG & DEBUG_SYNC) { + fprintf(stderr, "waiting for idle\n"); + dri_bo_map(batch->buf, GL_TRUE); + dri_bo_unmap(batch->buf); + } + + /* Reset the buffer: + */ + intel_batchbuffer_reset(batch); +} + + +/* This is the only way buffers get added to the validate list. + */ +GLboolean +intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch, + dri_bo *buffer, + uint32_t read_domains, uint32_t write_domain, + uint32_t delta) +{ + int ret; + + if (batch->ptr - batch->map > batch->buf->size) + _mesa_printf ("bad relocation ptr %p map %p offset %d size %d\n", + batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size); + ret = dri_bo_emit_reloc(batch->buf, read_domains, write_domain, + delta, batch->ptr - batch->map, buffer); + + /* + * Using the old buffer offset, write in what the right data would be, in case + * the buffer doesn't move and we can short-circuit the relocation processing + * in the kernel + */ + intel_batchbuffer_emit_dword (batch, buffer->offset + delta); + + return GL_TRUE; +} + +void +intel_batchbuffer_data(struct intel_batchbuffer *batch, + const void *data, GLuint bytes, + enum cliprect_mode cliprect_mode) +{ + assert((bytes & 3) == 0); + intel_batchbuffer_require_space(batch, bytes, cliprect_mode); + __memcpy(batch->ptr, data, bytes); + batch->ptr += bytes; +} diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.h b/src/mesa/drivers/dri/intel/intel_batchbuffer.h new file mode 100644 index 0000000000..cf7d4a0633 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.h @@ -0,0 +1,156 @@ +#ifndef INTEL_BATCHBUFFER_H +#define INTEL_BATCHBUFFER_H + +#include "mtypes.h" + +#include "intel_context.h" +#include "intel_bufmgr.h" +#include "intel_reg.h" + +#define BATCH_SZ 16384 +#define BATCH_RESERVED 16 + +enum cliprect_mode { + /** + * Batchbuffer contents may be looped over per cliprect, but do not + * require it. + */ + IGNORE_CLIPRECTS, + /** + * Batchbuffer contents require looping over per cliprect at batch submit + * time. + */ + LOOP_CLIPRECTS, + /** + * Batchbuffer contents contain drawing that should not be executed multiple + * times. + */ + NO_LOOP_CLIPRECTS, + /** + * Batchbuffer contents contain drawing that already handles cliprects, such + * as 2D drawing to front/back/depth that doesn't respect DRAWING_RECTANGLE. + * Equivalent behavior to NO_LOOP_CLIPRECTS, but may not persist in batch + * outside of LOCK/UNLOCK. + */ + REFERENCES_CLIPRECTS +}; + +struct intel_batchbuffer +{ + struct intel_context *intel; + + dri_bo *buf; + + GLubyte *buffer; + + GLubyte *map; + GLubyte *ptr; + + enum cliprect_mode cliprect_mode; + + GLuint size; + + GLuint dirty_state; +}; + +struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context + *intel); + +void intel_batchbuffer_free(struct intel_batchbuffer *batch); + + +void _intel_batchbuffer_flush(struct intel_batchbuffer *batch, + const char *file, int line); + +#define intel_batchbuffer_flush(batch) \ + _intel_batchbuffer_flush(batch, __FILE__, __LINE__) + +void intel_batchbuffer_reset(struct intel_batchbuffer *batch); + + +/* Unlike bmBufferData, this currently requires the buffer be mapped. + * Consider it a convenience function wrapping multple + * intel_buffer_dword() calls. + */ +void intel_batchbuffer_data(struct intel_batchbuffer *batch, + const void *data, GLuint bytes, + enum cliprect_mode cliprect_mode); + +void intel_batchbuffer_release_space(struct intel_batchbuffer *batch, + GLuint bytes); + +GLboolean intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch, + dri_bo *buffer, + uint32_t read_domains, + uint32_t write_domain, + uint32_t offset); + +/* Inline functions - might actually be better off with these + * non-inlined. Certainly better off switching all command packets to + * be passed as structs rather than dwords, but that's a little bit of + * work... + */ +static INLINE GLint +intel_batchbuffer_space(struct intel_batchbuffer *batch) +{ + return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map); +} + + +static INLINE void +intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, GLuint dword) +{ + assert(batch->map); + assert(intel_batchbuffer_space(batch) >= 4); + *(GLuint *) (batch->ptr) = dword; + batch->ptr += 4; +} + +static INLINE void +intel_batchbuffer_require_space(struct intel_batchbuffer *batch, + GLuint sz, + enum cliprect_mode cliprect_mode) +{ + assert(sz < batch->size - 8); + if (intel_batchbuffer_space(batch) < sz) + intel_batchbuffer_flush(batch); + + if (cliprect_mode != IGNORE_CLIPRECTS) { + if (batch->cliprect_mode == IGNORE_CLIPRECTS) { + batch->cliprect_mode = cliprect_mode; + } else { + if (batch->cliprect_mode != cliprect_mode) { + intel_batchbuffer_flush(batch); + batch->cliprect_mode = cliprect_mode; + } + } + } +} + +/* Here are the crusty old macros, to be removed: + */ +#define BATCH_LOCALS + +#define BEGIN_BATCH(n, cliprect_mode) do { \ + intel_batchbuffer_require_space(intel->batch, (n)*4, cliprect_mode); \ +} while (0) + +#define OUT_BATCH(d) intel_batchbuffer_emit_dword(intel->batch, d) + +#define OUT_RELOC(buf, read_domains, write_domain, delta) do { \ + assert((delta) >= 0); \ + intel_batchbuffer_emit_reloc(intel->batch, buf, \ + read_domains, write_domain, delta); \ +} while (0) + +#define ADVANCE_BATCH() do { } while(0) + + +static INLINE void +intel_batchbuffer_emit_mi_flush(struct intel_batchbuffer *batch) +{ + intel_batchbuffer_require_space(batch, 4, IGNORE_CLIPRECTS); + intel_batchbuffer_emit_dword(batch, MI_FLUSH); +} + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c new file mode 100644 index 0000000000..653f485dd0 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_blit.c @@ -0,0 +1,642 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <stdio.h> +#include <errno.h> + +#include "mtypes.h" +#include "context.h" +#include "enums.h" + +#include "intel_blit.h" +#include "intel_buffers.h" +#include "intel_context.h" +#include "intel_fbo.h" +#include "intel_reg.h" +#include "intel_regions.h" +#include "intel_batchbuffer.h" + +#define FILE_DEBUG_FLAG DEBUG_BLIT + +/** + * Copy the back color buffer to the front color buffer. + * Used for SwapBuffers(). + */ +void +intelCopyBuffer(const __DRIdrawablePrivate * dPriv, + const drm_clip_rect_t * rect) +{ + + struct intel_context *intel; + const intelScreenPrivate *intelScreen; + + DBG("%s\n", __FUNCTION__); + + assert(dPriv); + + intel = intelScreenContext(dPriv->driScreenPriv->private); + if (!intel) + return; + + intelScreen = intel->intelScreen; + + /* The LOCK_HARDWARE is required for the cliprects. Buffer offsets + * should work regardless. + */ + LOCK_HARDWARE(intel); + + if (dPriv && dPriv->numClipRects) { + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + struct intel_region *src, *dst; + int nbox = dPriv->numClipRects; + drm_clip_rect_t *pbox = dPriv->pClipRects; + int cpp; + int src_pitch, dst_pitch; + unsigned short src_x, src_y; + int BR13, CMD; + int i; + dri_bo *aper_array[3]; + + src = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT); + dst = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT); + + src_pitch = src->pitch * src->cpp; + dst_pitch = dst->pitch * dst->cpp; + + cpp = src->cpp; + + ASSERT(intel_fb); + ASSERT(intel_fb->Base.Name == 0); /* Not a user-created FBO */ + ASSERT(src); + ASSERT(dst); + ASSERT(src->cpp == dst->cpp); + + if (cpp == 2) { + BR13 = (0xCC << 16) | (1 << 24); + CMD = XY_SRC_COPY_BLT_CMD; + } + else { + BR13 = (0xCC << 16) | (1 << 24) | (1 << 25); + CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + } + +#ifndef I915 + if (src->tiling != I915_TILING_NONE) { + CMD |= XY_SRC_TILED; + src_pitch /= 4; + } + if (dst->tiling != I915_TILING_NONE) { + CMD |= XY_DST_TILED; + dst_pitch /= 4; + } +#endif + /* do space/cliprects check before going any further */ + intel_batchbuffer_require_space(intel->batch, 8 * 4, + REFERENCES_CLIPRECTS); + again: + aper_array[0] = intel->batch->buf; + aper_array[1] = dst->buffer; + aper_array[2] = src->buffer; + + if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) { + intel_batchbuffer_flush(intel->batch); + goto again; + } + + for (i = 0; i < nbox; i++, pbox++) { + drm_clip_rect_t box = *pbox; + + if (rect) { + if (!intel_intersect_cliprects(&box, &box, rect)) + continue; + } + + if (box.x1 >= box.x2 || + box.y1 >= box.y2) + continue; + + assert(box.x1 < box.x2); + assert(box.y1 < box.y2); + src_x = box.x1 - dPriv->x + dPriv->backX; + src_y = box.y1 - dPriv->y + dPriv->backY; + + BEGIN_BATCH(8, REFERENCES_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13 | dst_pitch); + OUT_BATCH((box.y1 << 16) | box.x1); + OUT_BATCH((box.y2 << 16) | box.x2); + + OUT_RELOC(dst->buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + 0); + OUT_BATCH((src_y << 16) | src_x); + OUT_BATCH(src_pitch); + OUT_RELOC(src->buffer, + I915_GEM_DOMAIN_RENDER, 0, + 0); + ADVANCE_BATCH(); + } + + /* Flush the rendering and the batch so that the results all land on the + * screen in a timely fashion. + */ + intel_batchbuffer_emit_mi_flush(intel->batch); + intel_batchbuffer_flush(intel->batch); + } + + UNLOCK_HARDWARE(intel); +} + + + + +void +intelEmitFillBlit(struct intel_context *intel, + GLuint cpp, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort x, GLshort y, + GLshort w, GLshort h, + GLuint color) +{ + GLuint BR13, CMD; + BATCH_LOCALS; + + dst_pitch *= cpp; + + switch (cpp) { + case 1: + case 2: + case 3: + BR13 = (0xF0 << 16) | (1 << 24); + CMD = XY_COLOR_BLT_CMD; + break; + case 4: + BR13 = (0xF0 << 16) | (1 << 24) | (1 << 25); + CMD = XY_COLOR_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + break; + default: + return; + } +#ifndef I915 + if (dst_tiling != I915_TILING_NONE) { + CMD |= XY_DST_TILED; + dst_pitch /= 4; + } +#endif + + DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h); + + assert(w > 0); + assert(h > 0); + + BEGIN_BATCH(6, NO_LOOP_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13 | dst_pitch); + OUT_BATCH((y << 16) | x); + OUT_BATCH(((y + h) << 16) | (x + w)); + OUT_RELOC(dst_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + dst_offset); + OUT_BATCH(color); + ADVANCE_BATCH(); +} + +static GLuint translate_raster_op(GLenum logicop) +{ + switch(logicop) { + case GL_CLEAR: return 0x00; + case GL_AND: return 0x88; + case GL_AND_REVERSE: return 0x44; + case GL_COPY: return 0xCC; + case GL_AND_INVERTED: return 0x22; + case GL_NOOP: return 0xAA; + case GL_XOR: return 0x66; + case GL_OR: return 0xEE; + case GL_NOR: return 0x11; + case GL_EQUIV: return 0x99; + case GL_INVERT: return 0x55; + case GL_OR_REVERSE: return 0xDD; + case GL_COPY_INVERTED: return 0x33; + case GL_OR_INVERTED: return 0xBB; + case GL_NAND: return 0x77; + case GL_SET: return 0xFF; + default: return 0; + } +} + + +/* Copy BitBlt + */ +void +intelEmitCopyBlit(struct intel_context *intel, + GLuint cpp, + GLshort src_pitch, + dri_bo *src_buffer, + GLuint src_offset, + uint32_t src_tiling, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort src_x, GLshort src_y, + GLshort dst_x, GLshort dst_y, + GLshort w, GLshort h, + GLenum logic_op) +{ + GLuint CMD, BR13; + int dst_y2 = dst_y + h; + int dst_x2 = dst_x + w; + dri_bo *aper_array[3]; + BATCH_LOCALS; + + /* do space/cliprects check before going any further */ + intel_batchbuffer_require_space(intel->batch, 8 * 4, NO_LOOP_CLIPRECTS); + again: + aper_array[0] = intel->batch->buf; + aper_array[1] = dst_buffer; + aper_array[2] = src_buffer; + + if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) { + intel_batchbuffer_flush(intel->batch); + goto again; + } + + DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, + src_buffer, src_pitch, src_offset, src_x, src_y, + dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h); + + src_pitch *= cpp; + dst_pitch *= cpp; + + BR13 = translate_raster_op(logic_op) << 16; + + switch (cpp) { + case 1: + case 2: + case 3: + BR13 |= (1 << 24); + CMD = XY_SRC_COPY_BLT_CMD; + break; + case 4: + BR13 |= (1 << 24) | (1 << 25); + CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + break; + default: + return; + } + +#ifndef I915 + if (dst_tiling != I915_TILING_NONE) { + CMD |= XY_DST_TILED; + dst_pitch /= 4; + } + if (src_tiling != I915_TILING_NONE) { + CMD |= XY_SRC_TILED; + src_pitch /= 4; + } +#endif + + if (dst_y2 <= dst_y || dst_x2 <= dst_x) { + return; + } + + /* Initial y values don't seem to work with negative pitches. If + * we adjust the offsets manually (below), it seems to work fine. + * + * On the other hand, if we always adjust, the hardware doesn't + * know which blit directions to use, so overlapping copypixels get + * the wrong result. + */ + if (dst_pitch > 0 && src_pitch > 0) { + assert(dst_x < dst_x2); + assert(dst_y < dst_y2); + + BEGIN_BATCH(8, NO_LOOP_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13 | dst_pitch); + OUT_BATCH((dst_y << 16) | dst_x); + OUT_BATCH((dst_y2 << 16) | dst_x2); + OUT_RELOC(dst_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + dst_offset); + OUT_BATCH((src_y << 16) | src_x); + OUT_BATCH(src_pitch); + OUT_RELOC(src_buffer, + I915_GEM_DOMAIN_RENDER, 0, + src_offset); + ADVANCE_BATCH(); + } + else { + assert(dst_x < dst_x2); + assert(h > 0); + + BEGIN_BATCH(8, NO_LOOP_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13 | ((uint16_t)dst_pitch)); + OUT_BATCH((0 << 16) | dst_x); + OUT_BATCH((h << 16) | dst_x2); + OUT_RELOC(dst_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + dst_offset + dst_y * dst_pitch); + OUT_BATCH((0 << 16) | src_x); + OUT_BATCH(src_pitch); + OUT_RELOC(src_buffer, + I915_GEM_DOMAIN_RENDER, 0, + src_offset + src_y * src_pitch); + ADVANCE_BATCH(); + } + intel_batchbuffer_emit_mi_flush(intel->batch); +} + + +/** + * Use blitting to clear the renderbuffers named by 'flags'. + * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field + * since that might include software renderbuffers or renderbuffers + * which we're clearing with triangles. + * \param mask bitmask of BUFFER_BIT_* values indicating buffers to clear + */ +void +intelClearWithBlit(GLcontext *ctx, GLbitfield mask) +{ + struct intel_context *intel = intel_context(ctx); + struct gl_framebuffer *fb = ctx->DrawBuffer; + GLuint clear_depth; + GLbitfield skipBuffers = 0; + BATCH_LOCALS; + + /* + * Compute values for clearing the buffers. + */ + clear_depth = 0; + if (mask & BUFFER_BIT_DEPTH) { + clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear); + } + if (mask & BUFFER_BIT_STENCIL) { + clear_depth |= (ctx->Stencil.Clear & 0xff) << 24; + } + + /* If clearing both depth and stencil, skip BUFFER_BIT_STENCIL in + * the loop below. + */ + if ((mask & BUFFER_BIT_DEPTH) && (mask & BUFFER_BIT_STENCIL)) { + skipBuffers = BUFFER_BIT_STENCIL; + } + + /* XXX Move this flush/lock into the following conditional? */ + intelFlush(&intel->ctx); + LOCK_HARDWARE(intel); + + if (intel->numClipRects) { + GLint cx, cy, cw, ch; + drm_clip_rect_t clear; + int i; + + /* Get clear bounds after locking */ + cx = fb->_Xmin; + cy = fb->_Ymin; + cw = fb->_Xmax - cx; + ch = fb->_Ymax - cy; + + if (fb->Name == 0) { + /* clearing a window */ + + /* flip top to bottom */ + clear.x1 = cx + intel->drawX; + clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch; + clear.x2 = clear.x1 + cw; + clear.y2 = clear.y1 + ch; + } + else { + /* clearing FBO */ + assert(intel->numClipRects == 1); + assert(intel->pClipRects == &intel->fboRect); + clear.x1 = cx; + clear.y1 = cy; + clear.x2 = clear.x1 + cw; + clear.y2 = clear.y1 + ch; + /* no change to mask */ + } + + for (i = 0; i < intel->numClipRects; i++) { + const drm_clip_rect_t *box = &intel->pClipRects[i]; + drm_clip_rect_t b; + GLuint buf; + GLuint clearMask = mask; /* use copy, since we modify it below */ + GLboolean all = (cw == fb->Width && ch == fb->Height); + + if (!all) { + intel_intersect_cliprects(&b, &clear, box); + } + else { + b = *box; + } + + if (b.x1 >= b.x2 || b.y1 >= b.y2) + continue; + + if (0) + _mesa_printf("clear %d,%d..%d,%d, mask %x\n", + b.x1, b.y1, b.x2, b.y2, mask); + + /* Loop over all renderbuffers */ + for (buf = 0; buf < BUFFER_COUNT && clearMask; buf++) { + const GLbitfield bufBit = 1 << buf; + if ((clearMask & bufBit) && !(bufBit & skipBuffers)) { + /* OK, clear this renderbuffer */ + struct intel_region *irb_region = + intel_get_rb_region(fb, buf); + dri_bo *write_buffer = + intel_region_buffer(intel, irb_region, + all ? INTEL_WRITE_FULL : + INTEL_WRITE_PART); + + GLuint clearVal; + GLint pitch, cpp; + GLuint BR13, CMD; + + ASSERT(irb_region); + + pitch = irb_region->pitch; + cpp = irb_region->cpp; + + DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, + irb_region->buffer, (pitch * cpp), + irb_region->draw_offset, + b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1); + + BR13 = 0xf0 << 16; + CMD = XY_COLOR_BLT_CMD; + + /* Setup the blit command */ + if (cpp == 4) { + BR13 |= (1 << 24) | (1 << 25); + if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) { + if (clearMask & BUFFER_BIT_DEPTH) + CMD |= XY_BLT_WRITE_RGB; + if (clearMask & BUFFER_BIT_STENCIL) + CMD |= XY_BLT_WRITE_ALPHA; + } + else { + /* clearing RGBA */ + CMD |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + } + } + else { + ASSERT(cpp == 2 || cpp == 0); + BR13 |= (1 << 24); + } + +#ifndef I915 + if (irb_region->tiling != I915_TILING_NONE) { + CMD |= XY_DST_TILED; + pitch /= 4; + } +#endif + BR13 |= (pitch * cpp); + + if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) { + clearVal = clear_depth; + } + else { + clearVal = (cpp == 4) + ? intel->ClearColor8888 : intel->ClearColor565; + } + /* + _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n", + buf, irb->Base.Name); + */ + intel_wait_flips(intel); + + assert(b.x1 < b.x2); + assert(b.y1 < b.y2); + + BEGIN_BATCH(6, REFERENCES_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13); + OUT_BATCH((b.y1 << 16) | b.x1); + OUT_BATCH((b.y2 << 16) | b.x2); + OUT_RELOC(write_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + irb_region->draw_offset); + OUT_BATCH(clearVal); + ADVANCE_BATCH(); + clearMask &= ~bufBit; /* turn off bit, for faster loop exit */ + } + } + } + intel_batchbuffer_emit_mi_flush(intel->batch); + } + + UNLOCK_HARDWARE(intel); +} + +void +intelEmitImmediateColorExpandBlit(struct intel_context *intel, + GLuint cpp, + GLubyte *src_bits, GLuint src_size, + GLuint fg_color, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort x, GLshort y, + GLshort w, GLshort h, + GLenum logic_op) +{ + int dwords = ALIGN(src_size, 8) / 4; + uint32_t opcode, br13, blit_cmd; + + assert( logic_op - GL_CLEAR >= 0 ); + assert( logic_op - GL_CLEAR < 0x10 ); + + if (w < 0 || h < 0) + return; + + dst_pitch *= cpp; + + DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n", + __FUNCTION__, + dst_buffer, dst_pitch, dst_offset, x, y, w, h, src_size, dwords); + + intel_batchbuffer_require_space( intel->batch, + (8 * 4) + + (3 * 4) + + dwords, + REFERENCES_CLIPRECTS ); + + opcode = XY_SETUP_BLT_CMD; + if (cpp == 4) + opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; +#ifndef I915 + if (dst_tiling != I915_TILING_NONE) { + opcode |= XY_DST_TILED; + dst_pitch /= 4; + } +#endif + + br13 = dst_pitch | (translate_raster_op(logic_op) << 16) | (1 << 29); + if (cpp == 2) + br13 |= BR13_565; + else + br13 |= BR13_8888; + + blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */ + if (dst_tiling != I915_TILING_NONE) + blit_cmd |= XY_DST_TILED; + + BEGIN_BATCH(8 + 3, REFERENCES_CLIPRECTS); + OUT_BATCH(opcode); + OUT_BATCH(br13); + OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */ + OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */ + OUT_RELOC(dst_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + dst_offset); + OUT_BATCH(0); /* bg */ + OUT_BATCH(fg_color); /* fg */ + OUT_BATCH(0); /* pattern base addr */ + + OUT_BATCH(blit_cmd | ((3 - 2) + dwords)); + OUT_BATCH((y << 16) | x); + OUT_BATCH(((y + h) << 16) | (x + w)); + ADVANCE_BATCH(); + + intel_batchbuffer_data( intel->batch, + src_bits, + dwords * 4, + REFERENCES_CLIPRECTS ); + + intel_batchbuffer_emit_mi_flush(intel->batch); +} diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h new file mode 100644 index 0000000000..52065b13ed --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_blit.h @@ -0,0 +1,75 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_BLIT_H +#define INTEL_BLIT_H + +#include "intel_context.h" + +extern void intelCopyBuffer(const __DRIdrawablePrivate * dpriv, + const drm_clip_rect_t * rect); + +extern void intelClearWithBlit(GLcontext * ctx, GLbitfield mask); + +extern void intelEmitCopyBlit(struct intel_context *intel, + GLuint cpp, + GLshort src_pitch, + dri_bo *src_buffer, + GLuint src_offset, + uint32_t src_tiling, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort srcx, GLshort srcy, + GLshort dstx, GLshort dsty, + GLshort w, GLshort h, + GLenum logicop ); + +extern void intelEmitFillBlit(struct intel_context *intel, + GLuint cpp, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort x, GLshort y, + GLshort w, GLshort h, GLuint color); + +void +intelEmitImmediateColorExpandBlit(struct intel_context *intel, + GLuint cpp, + GLubyte *src_bits, GLuint src_size, + GLuint fg_color, + GLshort dst_pitch, + dri_bo *dst_buffer, + GLuint dst_offset, + uint32_t dst_tiling, + GLshort x, GLshort y, + GLshort w, GLshort h, + GLenum logic_op); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c new file mode 100644 index 0000000000..88484a07af --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c @@ -0,0 +1,284 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "imports.h" +#include "mtypes.h" +#include "bufferobj.h" + +#include "intel_context.h" +#include "intel_buffer_objects.h" +#include "intel_batchbuffer.h" +#include "intel_regions.h" + +static GLboolean intel_bufferobj_unmap(GLcontext * ctx, + GLenum target, + struct gl_buffer_object *obj); + +/** Allocates a new dri_bo to store the data for the buffer object. */ +static void +intel_bufferobj_alloc_buffer(struct intel_context *intel, + struct intel_buffer_object *intel_obj) +{ + intel_obj->buffer = dri_bo_alloc(intel->bufmgr, "bufferobj", + intel_obj->Base.Size, 64); +} + +/** + * There is some duplication between mesa's bufferobjects and our + * bufmgr buffers. Both have an integer handle and a hashtable to + * lookup an opaque structure. It would be nice if the handles and + * internal structure where somehow shared. + */ +static struct gl_buffer_object * +intel_bufferobj_alloc(GLcontext * ctx, GLuint name, GLenum target) +{ + struct intel_buffer_object *obj = CALLOC_STRUCT(intel_buffer_object); + + _mesa_initialize_buffer_object(&obj->Base, name, target); + + obj->buffer = NULL; + + return &obj->Base; +} + +/* Break the COW tie to the region. The region gets to keep the data. + */ +void +intel_bufferobj_release_region(struct intel_context *intel, + struct intel_buffer_object *intel_obj) +{ + assert(intel_obj->region->buffer == intel_obj->buffer); + intel_obj->region->pbo = NULL; + intel_obj->region = NULL; + + dri_bo_unreference(intel_obj->buffer); + intel_obj->buffer = NULL; +} + +/* Break the COW tie to the region. Both the pbo and the region end + * up with a copy of the data. + */ +void +intel_bufferobj_cow(struct intel_context *intel, + struct intel_buffer_object *intel_obj) +{ + assert(intel_obj->region); + intel_region_cow(intel, intel_obj->region); +} + + +/** + * Deallocate/free a vertex/pixel buffer object. + * Called via glDeleteBuffersARB(). + */ +static void +intel_bufferobj_free(GLcontext * ctx, struct gl_buffer_object *obj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + assert(intel_obj); + + /* Buffer objects are automatically unmapped when deleting according + * to the spec. + */ + if (obj->Pointer) + intel_bufferobj_unmap(ctx, 0, obj); + + if (intel_obj->region) { + intel_bufferobj_release_region(intel, intel_obj); + } + else if (intel_obj->buffer) { + dri_bo_unreference(intel_obj->buffer); + } + + _mesa_free(intel_obj); +} + + + +/** + * Allocate space for and store data in a buffer object. Any data that was + * previously stored in the buffer object is lost. If data is NULL, + * memory will be allocated, but no copy will occur. + * Called via glBufferDataARB(). + */ +static void +intel_bufferobj_data(GLcontext * ctx, + GLenum target, + GLsizeiptrARB size, + const GLvoid * data, + GLenum usage, struct gl_buffer_object *obj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + intel_obj->Base.Size = size; + intel_obj->Base.Usage = usage; + + /* Buffer objects are automatically unmapped when creating new data buffers + * according to the spec. + */ + if (obj->Pointer) + intel_bufferobj_unmap(ctx, 0, obj); + + if (intel_obj->region) + intel_bufferobj_release_region(intel, intel_obj); + + if (intel_obj->buffer != NULL) { + dri_bo_unreference(intel_obj->buffer); + intel_obj->buffer = NULL; + } + if (size != 0) { + intel_bufferobj_alloc_buffer(intel, intel_obj); + + if (data != NULL) + dri_bo_subdata(intel_obj->buffer, 0, size, data); + } +} + + +/** + * Replace data in a subrange of buffer object. If the data range + * specified by size + offset extends beyond the end of the buffer or + * if data is NULL, no copy is performed. + * Called via glBufferSubDataARB(). + */ +static void +intel_bufferobj_subdata(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + const GLvoid * data, struct gl_buffer_object *obj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + assert(intel_obj); + + if (intel_obj->region) + intel_bufferobj_cow(intel, intel_obj); + + dri_bo_subdata(intel_obj->buffer, offset, size, data); +} + + +/** + * Called via glGetBufferSubDataARB(). + */ +static void +intel_bufferobj_get_subdata(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + GLvoid * data, struct gl_buffer_object *obj) +{ + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + assert(intel_obj); + dri_bo_get_subdata(intel_obj->buffer, offset, size, data); +} + + + +/** + * Called via glMapBufferARB(). + */ +static void * +intel_bufferobj_map(GLcontext * ctx, + GLenum target, + GLenum access, struct gl_buffer_object *obj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + /* XXX: Translate access to flags arg below: + */ + assert(intel_obj); + + if (intel_obj->region) + intel_bufferobj_cow(intel, intel_obj); + + if (intel_obj->buffer == NULL) { + obj->Pointer = NULL; + return NULL; + } + + dri_bo_map(intel_obj->buffer, GL_TRUE); + obj->Pointer = intel_obj->buffer->virtual; + return obj->Pointer; +} + + +/** + * Called via glMapBufferARB(). + */ +static GLboolean +intel_bufferobj_unmap(GLcontext * ctx, + GLenum target, struct gl_buffer_object *obj) +{ + struct intel_buffer_object *intel_obj = intel_buffer_object(obj); + + assert(intel_obj); + if (intel_obj->buffer != NULL) { + assert(obj->Pointer); + dri_bo_unmap(intel_obj->buffer); + obj->Pointer = NULL; + } + return GL_TRUE; +} + +dri_bo * +intel_bufferobj_buffer(struct intel_context *intel, + struct intel_buffer_object *intel_obj, GLuint flag) +{ + if (intel_obj->region) { + if (flag == INTEL_WRITE_PART) + intel_bufferobj_cow(intel, intel_obj); + else if (flag == INTEL_WRITE_FULL) { + intel_bufferobj_release_region(intel, intel_obj); + intel_bufferobj_alloc_buffer(intel, intel_obj); + } + } + + return intel_obj->buffer; +} + +void +intel_bufferobj_init(struct intel_context *intel) +{ + GLcontext *ctx = &intel->ctx; + + ctx->Driver.NewBufferObject = intel_bufferobj_alloc; + ctx->Driver.DeleteBuffer = intel_bufferobj_free; + ctx->Driver.BufferData = intel_bufferobj_data; + ctx->Driver.BufferSubData = intel_bufferobj_subdata; + ctx->Driver.GetBufferSubData = intel_bufferobj_get_subdata; + ctx->Driver.MapBuffer = intel_bufferobj_map; + ctx->Driver.UnmapBuffer = intel_bufferobj_unmap; +} diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h new file mode 100644 index 0000000000..7cecc3232d --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h @@ -0,0 +1,86 @@ +/************************************************************************** + * + * Copyright 2005 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_BUFFEROBJ_H +#define INTEL_BUFFEROBJ_H + +#include "mtypes.h" + +struct intel_context; +struct intel_region; +struct gl_buffer_object; + + +/** + * Intel vertex/pixel buffer object, derived from Mesa's gl_buffer_object. + */ +struct intel_buffer_object +{ + struct gl_buffer_object Base; + dri_bo *buffer; /* the low-level buffer manager's buffer handle */ + + struct intel_region *region; /* Is there a zero-copy texture + associated with this (pixel) + buffer object? */ +}; + + +/* Get the bm buffer associated with a GL bufferobject: + */ +dri_bo *intel_bufferobj_buffer(struct intel_context *intel, + struct intel_buffer_object + *obj, GLuint flag); + +/* Hook the bufferobject implementation into mesa: + */ +void intel_bufferobj_init(struct intel_context *intel); + + + +/* Are the obj->Name tests necessary? Unfortunately yes, mesa + * allocates a couple of gl_buffer_object structs statically, and + * the Name == 0 test is the only way to identify them and avoid + * casting them erroneously to our structs. + */ +static INLINE struct intel_buffer_object * +intel_buffer_object(struct gl_buffer_object *obj) +{ + if (obj->Name) + return (struct intel_buffer_object *) obj; + else + return NULL; +} + +/* Helpers for zerocopy image uploads. See also intel_regions.h: + */ +void intel_bufferobj_cow(struct intel_context *intel, + struct intel_buffer_object *intel_obj); +void intel_bufferobj_release_region(struct intel_context *intel, + struct intel_buffer_object *intel_obj); + + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_buffers.c b/src/mesa/drivers/dri/intel/intel_buffers.c new file mode 100644 index 0000000000..3fe67462d5 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_buffers.c @@ -0,0 +1,1099 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_blit.h" +#include "intel_buffers.h" +#include "intel_chipset.h" +#include "intel_depthstencil.h" +#include "intel_fbo.h" +#include "intel_regions.h" +#include "intel_batchbuffer.h" +#include "intel_reg.h" +#include "context.h" +#include "utils.h" +#include "drirenderbuffer.h" +#include "framebuffer.h" +#include "swrast/swrast.h" +#include "vblank.h" +#include "i915_drm.h" + +/* This block can be removed when libdrm >= 2.3.1 is required */ + +#ifndef DRM_IOCTL_I915_FLIP + +#define DRM_VBLANK_FLIP 0x8000000 + +typedef struct drm_i915_flip { + int pipes; +} drm_i915_flip_t; + +#undef DRM_IOCTL_I915_FLIP +#define DRM_IOCTL_I915_FLIP DRM_IOW(DRM_COMMAND_BASE + DRM_I915_FLIP, \ + drm_i915_flip_t) + +#endif + +#define FILE_DEBUG_FLAG DEBUG_BLIT + +/** + * XXX move this into a new dri/common/cliprects.c file. + */ +GLboolean +intel_intersect_cliprects(drm_clip_rect_t * dst, + const drm_clip_rect_t * a, + const drm_clip_rect_t * b) +{ + GLint bx = b->x1; + GLint by = b->y1; + GLint bw = b->x2 - bx; + GLint bh = b->y2 - by; + + if (bx < a->x1) + bw -= a->x1 - bx, bx = a->x1; + if (by < a->y1) + bh -= a->y1 - by, by = a->y1; + if (bx + bw > a->x2) + bw = a->x2 - bx; + if (by + bh > a->y2) + bh = a->y2 - by; + if (bw <= 0) + return GL_FALSE; + if (bh <= 0) + return GL_FALSE; + + dst->x1 = bx; + dst->y1 = by; + dst->x2 = bx + bw; + dst->y2 = by + bh; + + return GL_TRUE; +} + +/** + * Return pointer to current color drawing region, or NULL. + */ +struct intel_region * +intel_drawbuf_region(struct intel_context *intel) +{ + struct intel_renderbuffer *irbColor = + intel_renderbuffer(intel->ctx.DrawBuffer->_ColorDrawBuffers[0]); + if (irbColor) + return irbColor->region; + else + return NULL; +} + +/** + * Return pointer to current color reading region, or NULL. + */ +struct intel_region * +intel_readbuf_region(struct intel_context *intel) +{ + struct intel_renderbuffer *irb + = intel_renderbuffer(intel->ctx.ReadBuffer->_ColorReadBuffer); + if (irb) + return irb->region; + else + return NULL; +} + + + +/** + * Update the following fields for rendering to a user-created FBO: + * intel->numClipRects + * intel->pClipRects + * intel->drawX + * intel->drawY + */ +static void +intelSetRenderbufferClipRects(struct intel_context *intel) +{ + assert(intel->ctx.DrawBuffer->Width > 0); + assert(intel->ctx.DrawBuffer->Height > 0); + intel->fboRect.x1 = 0; + intel->fboRect.y1 = 0; + intel->fboRect.x2 = intel->ctx.DrawBuffer->Width; + intel->fboRect.y2 = intel->ctx.DrawBuffer->Height; + intel->numClipRects = 1; + intel->pClipRects = &intel->fboRect; + intel->drawX = 0; + intel->drawY = 0; +} + + +/** + * As above, but for rendering to front buffer of a window. + * \sa intelSetRenderbufferClipRects + */ +static void +intelSetFrontClipRects(struct intel_context *intel) +{ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + + if (!dPriv) + return; + + intel->numClipRects = dPriv->numClipRects; + intel->pClipRects = dPriv->pClipRects; + intel->drawX = dPriv->x; + intel->drawY = dPriv->y; +} + + +/** + * As above, but for rendering to back buffer of a window. + */ +static void +intelSetBackClipRects(struct intel_context *intel) +{ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + struct intel_framebuffer *intel_fb; + + if (!dPriv) + return; + + intel_fb = dPriv->driverPrivate; + + if (intel_fb->pf_active || dPriv->numBackClipRects == 0) { + /* use the front clip rects */ + intel->numClipRects = dPriv->numClipRects; + intel->pClipRects = dPriv->pClipRects; + intel->drawX = dPriv->x; + intel->drawY = dPriv->y; + } + else { + /* use the back clip rects */ + intel->numClipRects = dPriv->numBackClipRects; + intel->pClipRects = dPriv->pBackClipRects; + intel->drawX = dPriv->backX; + intel->drawY = dPriv->backY; + } +} + +static void +intelUpdatePageFlipping(struct intel_context *intel, + GLint areaA, GLint areaB) +{ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + GLboolean pf_active; + GLint pf_planes; + + /* Update page flipping info */ + pf_planes = 0; + + if (areaA > 0) + pf_planes |= 1; + + if (areaB > 0) + pf_planes |= 2; + + intel_fb->pf_current_page = (intel->sarea->pf_current_page >> + (intel_fb->pf_planes & 0x2)) & 0x3; + + intel_fb->pf_num_pages = intel->intelScreen->third.handle ? 3 : 2; + + pf_active = pf_planes && (pf_planes & intel->sarea->pf_active) == pf_planes; + + if (INTEL_DEBUG & DEBUG_LOCK) + if (pf_active != intel_fb->pf_active) + _mesa_printf("%s - Page flipping %sactive\n", __progname, + pf_active ? "" : "in"); + + if (pf_active) { + /* Sync pages between planes if flipping on both at the same time */ + if (pf_planes == 0x3 && pf_planes != intel_fb->pf_planes && + (intel->sarea->pf_current_page & 0x3) != + (((intel->sarea->pf_current_page) >> 2) & 0x3)) { + drm_i915_flip_t flip; + + if (intel_fb->pf_current_page == + (intel->sarea->pf_current_page & 0x3)) { + /* XXX: This is ugly, but emitting two flips 'in a row' can cause + * lockups for unknown reasons. + */ + intel->sarea->pf_current_page = + intel->sarea->pf_current_page & 0x3; + intel->sarea->pf_current_page |= + ((intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) % + intel_fb->pf_num_pages) << 2; + + flip.pipes = 0x2; + } else { + intel->sarea->pf_current_page = + intel->sarea->pf_current_page & (0x3 << 2); + intel->sarea->pf_current_page |= + (intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) % + intel_fb->pf_num_pages; + + flip.pipes = 0x1; + } + + drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip)); + } + + intel_fb->pf_planes = pf_planes; + } + + intel_fb->pf_active = pf_active; + intel_flip_renderbuffers(intel_fb); + intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer); +} + +/** + * This will be called whenever the currently bound window is moved/resized. + * XXX: actually, it seems to NOT be called when the window is only moved (BP). + */ +void +intelWindowMoved(struct intel_context *intel) +{ + GLcontext *ctx = &intel->ctx; + __DRIdrawablePrivate *dPriv = intel->driDrawable; + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + + if (!intel->ctx.DrawBuffer) { + /* when would this happen? -BP */ + intelSetFrontClipRects(intel); + } + else if (intel->ctx.DrawBuffer->Name != 0) { + /* drawing to user-created FBO - do nothing */ + /* Cliprects would be set from intelDrawBuffer() */ + } + else { + /* drawing to a window */ + switch (intel_fb->Base._ColorDrawBufferIndexes[0]) { + case BUFFER_FRONT_LEFT: + intelSetFrontClipRects(intel); + break; + case BUFFER_BACK_LEFT: + intelSetBackClipRects(intel); + break; + default: + intelSetFrontClipRects(intel); + } + + } + + if (!intel->intelScreen->driScrnPriv->dri2.enabled && + intel->intelScreen->driScrnPriv->ddx_version.minor >= 7) { + volatile struct drm_i915_sarea *sarea = intel->sarea; + drm_clip_rect_t drw_rect = { .x1 = dPriv->x, .x2 = dPriv->x + dPriv->w, + .y1 = dPriv->y, .y2 = dPriv->y + dPriv->h }; + drm_clip_rect_t planeA_rect = { .x1 = sarea->planeA_x, .y1 = sarea->planeA_y, + .x2 = sarea->planeA_x + sarea->planeA_w, + .y2 = sarea->planeA_y + sarea->planeA_h }; + drm_clip_rect_t planeB_rect = { .x1 = sarea->planeB_x, .y1 = sarea->planeB_y, + .x2 = sarea->planeB_x + sarea->planeB_w, + .y2 = sarea->planeB_y + sarea->planeB_h }; + GLint areaA = driIntersectArea( drw_rect, planeA_rect ); + GLint areaB = driIntersectArea( drw_rect, planeB_rect ); + GLuint flags = dPriv->vblFlags; + + intelUpdatePageFlipping(intel, areaA, areaB); + + /* Update vblank info + */ + if (areaB > areaA || (areaA == areaB && areaB > 0)) { + flags = dPriv->vblFlags | VBLANK_FLAG_SECONDARY; + } else { + flags = dPriv->vblFlags & ~VBLANK_FLAG_SECONDARY; + } + + /* Check to see if we changed pipes */ + if (flags != dPriv->vblFlags && dPriv->vblFlags && + !(dPriv->vblFlags & VBLANK_FLAG_NO_IRQ)) { + int64_t count; + drmVBlank vbl; + int i; + + /* + * Deal with page flipping + */ + vbl.request.type = DRM_VBLANK_ABSOLUTE; + + if ( dPriv->vblFlags & VBLANK_FLAG_SECONDARY ) { + vbl.request.type |= DRM_VBLANK_SECONDARY; + } + + for (i = 0; i < intel_fb->pf_num_pages; i++) { + if (!intel_fb->color_rb[i] || + (intel_fb->vbl_waited - intel_fb->color_rb[i]->vbl_pending) <= + (1<<23)) + continue; + + vbl.request.sequence = intel_fb->color_rb[i]->vbl_pending; + drmWaitVBlank(intel->driFd, &vbl); + } + + /* + * Update msc_base from old pipe + */ + driDrawableGetMSC32(dPriv->driScreenPriv, dPriv, &count); + dPriv->msc_base = count; + /* + * Then get new vblank_base and vblSeq values + */ + dPriv->vblFlags = flags; + driGetCurrentVBlank(dPriv); + dPriv->vblank_base = dPriv->vblSeq; + + intel_fb->vbl_waited = dPriv->vblSeq; + + for (i = 0; i < intel_fb->pf_num_pages; i++) { + if (intel_fb->color_rb[i]) + intel_fb->color_rb[i]->vbl_pending = intel_fb->vbl_waited; + } + } + } else { + dPriv->vblFlags &= ~VBLANK_FLAG_SECONDARY; + } + + /* Update Mesa's notion of window size */ + driUpdateFramebufferSize(ctx, dPriv); + intel_fb->Base.Initialized = GL_TRUE; /* XXX remove someday */ + + /* Update hardware scissor */ + if (ctx->Driver.Scissor != NULL) { + ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y, + ctx->Scissor.Width, ctx->Scissor.Height); + } + + /* Re-calculate viewport related state */ + if (ctx->Driver.DepthRange != NULL) + ctx->Driver.DepthRange( ctx, ctx->Viewport.Near, ctx->Viewport.Far ); +} + + + +/* A true meta version of this would be very simple and additionally + * machine independent. Maybe we'll get there one day. + */ +static void +intelClearWithTris(struct intel_context *intel, GLbitfield mask) +{ + GLcontext *ctx = &intel->ctx; + struct gl_framebuffer *fb = ctx->DrawBuffer; + GLuint buf; + + intel->vtbl.install_meta_state(intel); + + /* Back and stencil cliprects are the same. Try and do both + * buffers at once: + */ + if (mask & (BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH)) { + struct intel_region *backRegion = + intel_get_rb_region(fb, BUFFER_BACK_LEFT); + struct intel_region *depthRegion = + intel_get_rb_region(fb, BUFFER_DEPTH); + + intel->vtbl.meta_draw_region(intel, backRegion, depthRegion); + + if (mask & BUFFER_BIT_BACK_LEFT) + intel->vtbl.meta_color_mask(intel, GL_TRUE); + else + intel->vtbl.meta_color_mask(intel, GL_FALSE); + + if (mask & BUFFER_BIT_STENCIL) + intel->vtbl.meta_stencil_replace(intel, + intel->ctx.Stencil.WriteMask[0], + intel->ctx.Stencil.Clear); + else + intel->vtbl.meta_no_stencil_write(intel); + + if (mask & BUFFER_BIT_DEPTH) + intel->vtbl.meta_depth_replace(intel); + else + intel->vtbl.meta_no_depth_write(intel); + + intel->vtbl.meta_draw_quad(intel, + fb->_Xmin, + fb->_Xmax, + fb->_Ymin, + fb->_Ymax, + intel->ctx.Depth.Clear, + intel->ClearColor8888, + 0, 0, 0, 0); /* texcoords */ + + mask &= ~(BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH); + } + + /* clear the remaining (color) renderbuffers */ + for (buf = 0; buf < BUFFER_COUNT && mask; buf++) { + const GLuint bufBit = 1 << buf; + if (mask & bufBit) { + struct intel_renderbuffer *irbColor = + intel_renderbuffer(fb->Attachment[buf].Renderbuffer); + + ASSERT(irbColor); + + intel->vtbl.meta_no_depth_write(intel); + intel->vtbl.meta_no_stencil_write(intel); + intel->vtbl.meta_color_mask(intel, GL_TRUE); + intel->vtbl.meta_draw_region(intel, irbColor->region, NULL); + + intel->vtbl.meta_draw_quad(intel, + fb->_Xmin, + fb->_Xmax, + fb->_Ymin, + fb->_Ymax, + 0, intel->ClearColor8888, + 0, 0, 0, 0); /* texcoords */ + + mask &= ~bufBit; + } + } + + intel->vtbl.leave_meta_state(intel); +} + +static const char *buffer_names[] = { + [BUFFER_FRONT_LEFT] = "front", + [BUFFER_BACK_LEFT] = "back", + [BUFFER_FRONT_RIGHT] = "front right", + [BUFFER_BACK_RIGHT] = "back right", + [BUFFER_AUX0] = "aux0", + [BUFFER_AUX1] = "aux1", + [BUFFER_AUX2] = "aux2", + [BUFFER_AUX3] = "aux3", + [BUFFER_DEPTH] = "depth", + [BUFFER_STENCIL] = "stencil", + [BUFFER_ACCUM] = "accum", + [BUFFER_COLOR0] = "color0", + [BUFFER_COLOR1] = "color1", + [BUFFER_COLOR2] = "color2", + [BUFFER_COLOR3] = "color3", + [BUFFER_COLOR4] = "color4", + [BUFFER_COLOR5] = "color5", + [BUFFER_COLOR6] = "color6", + [BUFFER_COLOR7] = "color7", +}; + +/** + * Called by ctx->Driver.Clear. + */ +static void +intelClear(GLcontext *ctx, GLbitfield mask) +{ + struct intel_context *intel = intel_context(ctx); + const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask); + GLbitfield tri_mask = 0; + GLbitfield blit_mask = 0; + GLbitfield swrast_mask = 0; + struct gl_framebuffer *fb = ctx->DrawBuffer; + GLuint i; + + if (0) + fprintf(stderr, "%s\n", __FUNCTION__); + + /* HW color buffers (front, back, aux, generic FBO, etc) */ + if (colorMask == ~0) { + /* clear all R,G,B,A */ + /* XXX FBO: need to check if colorbuffers are software RBOs! */ + blit_mask |= (mask & BUFFER_BITS_COLOR); + } + else { + /* glColorMask in effect */ + tri_mask |= (mask & BUFFER_BITS_COLOR); + } + + /* HW stencil */ + if (mask & BUFFER_BIT_STENCIL) { + const struct intel_region *stencilRegion + = intel_get_rb_region(fb, BUFFER_STENCIL); + if (stencilRegion) { + /* have hw stencil */ + if (IS_965(intel->intelScreen->deviceID) || + (ctx->Stencil.WriteMask[0] & 0xff) != 0xff) { + /* We have to use the 3D engine if we're clearing a partial mask + * of the stencil buffer, or if we're on a 965 which has a tiled + * depth/stencil buffer in a layout we can't blit to. + */ + tri_mask |= BUFFER_BIT_STENCIL; + } + else { + /* clearing all stencil bits, use blitting */ + blit_mask |= BUFFER_BIT_STENCIL; + } + } + } + + /* HW depth */ + if (mask & BUFFER_BIT_DEPTH) { + /* clear depth with whatever method is used for stencil (see above) */ + if (IS_965(intel->intelScreen->deviceID) || + tri_mask & BUFFER_BIT_STENCIL) + tri_mask |= BUFFER_BIT_DEPTH; + else + blit_mask |= BUFFER_BIT_DEPTH; + } + + /* SW fallback clearing */ + swrast_mask = mask & ~tri_mask & ~blit_mask; + + for (i = 0; i < BUFFER_COUNT; i++) { + GLuint bufBit = 1 << i; + if ((blit_mask | tri_mask) & bufBit) { + if (!fb->Attachment[i].Renderbuffer->ClassID) { + blit_mask &= ~bufBit; + tri_mask &= ~bufBit; + swrast_mask |= bufBit; + } + } + } + + if (blit_mask) { + if (INTEL_DEBUG & DEBUG_BLIT) { + DBG("blit clear:"); + for (i = 0; i < BUFFER_COUNT; i++) { + if (blit_mask & (1 << i)) + DBG(" %s", buffer_names[i]); + } + DBG("\n"); + } + intelClearWithBlit(ctx, blit_mask); + } + + if (tri_mask) { + if (INTEL_DEBUG & DEBUG_BLIT) { + DBG("tri clear:"); + for (i = 0; i < BUFFER_COUNT; i++) { + if (tri_mask & (1 << i)) + DBG(" %s", buffer_names[i]); + } + DBG("\n"); + } + intelClearWithTris(intel, tri_mask); + } + + if (swrast_mask) { + if (INTEL_DEBUG & DEBUG_BLIT) { + DBG("swrast clear:"); + for (i = 0; i < BUFFER_COUNT; i++) { + if (swrast_mask & (1 << i)) + DBG(" %s", buffer_names[i]); + } + DBG("\n"); + } + _swrast_Clear(ctx, swrast_mask); + } +} + + +/* Emit wait for pending flips */ +void +intel_wait_flips(struct intel_context *intel) +{ + struct intel_framebuffer *intel_fb = + (struct intel_framebuffer *) intel->ctx.DrawBuffer; + struct intel_renderbuffer *intel_rb = + intel_get_renderbuffer(&intel_fb->Base, + intel_fb->Base._ColorDrawBufferIndexes[0] == + BUFFER_FRONT_LEFT ? BUFFER_FRONT_LEFT : + BUFFER_BACK_LEFT); + + if (intel->intelScreen->driScrnPriv->dri2.enabled) + return; + + if (intel_fb->Base.Name == 0 && intel_rb && + intel_rb->pf_pending == intel_fb->pf_seq) { + GLint pf_planes = intel_fb->pf_planes; + BATCH_LOCALS; + + /* Wait for pending flips to take effect */ + BEGIN_BATCH(2, NO_LOOP_CLIPRECTS); + OUT_BATCH(pf_planes & 0x1 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP) + : 0); + OUT_BATCH(pf_planes & 0x2 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_B_FLIP) + : 0); + ADVANCE_BATCH(); + + intel_rb->pf_pending--; + } +} + + +/* Flip the front & back buffers + */ +static GLboolean +intelPageFlip(const __DRIdrawablePrivate * dPriv) +{ + struct intel_context *intel; + int ret; + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + + if (INTEL_DEBUG & DEBUG_IOCTL) + fprintf(stderr, "%s\n", __FUNCTION__); + + assert(dPriv); + assert(dPriv->driContextPriv); + assert(dPriv->driContextPriv->driverPrivate); + + intel = (struct intel_context *) dPriv->driContextPriv->driverPrivate; + + if (intel->intelScreen->drmMinor < 9) + return GL_FALSE; + + intelFlush(&intel->ctx); + + ret = 0; + + LOCK_HARDWARE(intel); + + if (dPriv->numClipRects && intel_fb->pf_active) { + drm_i915_flip_t flip; + + flip.pipes = intel_fb->pf_planes; + + ret = drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip)); + } + + UNLOCK_HARDWARE(intel); + + if (ret || !intel_fb->pf_active) + return GL_FALSE; + + if (!dPriv->numClipRects) { + usleep(10000); /* throttle invisible client 10ms */ + } + + intel_fb->pf_current_page = (intel->sarea->pf_current_page >> + (intel_fb->pf_planes & 0x2)) & 0x3; + + if (dPriv->numClipRects != 0) { + intel_get_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT)->pf_pending = + intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->pf_pending = + ++intel_fb->pf_seq; + } + + intel_flip_renderbuffers(intel_fb); + intel_draw_buffer(&intel->ctx, &intel_fb->Base); + + return GL_TRUE; +} + +static GLboolean +intelScheduleSwap(__DRIdrawablePrivate * dPriv, GLboolean *missed_target) +{ + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + unsigned int interval; + struct intel_context *intel = + intelScreenContext(dPriv->driScreenPriv->private); + const intelScreenPrivate *intelScreen = intel->intelScreen; + unsigned int target; + drm_i915_vblank_swap_t swap; + GLboolean ret; + + if (!dPriv->vblFlags || + (dPriv->vblFlags & VBLANK_FLAG_NO_IRQ) || + intelScreen->drmMinor < (intel_fb->pf_active ? 9 : 6)) + return GL_FALSE; + + interval = driGetVBlankInterval(dPriv); + + swap.seqtype = DRM_VBLANK_ABSOLUTE; + + if (dPriv->vblFlags & VBLANK_FLAG_SYNC) { + swap.seqtype |= DRM_VBLANK_NEXTONMISS; + } else if (interval == 0) + return GL_FALSE; + + swap.drawable = dPriv->hHWDrawable; + target = swap.sequence = dPriv->vblSeq + interval; + + if ( dPriv->vblFlags & VBLANK_FLAG_SECONDARY ) { + swap.seqtype |= DRM_VBLANK_SECONDARY; + } + + LOCK_HARDWARE(intel); + + intel_batchbuffer_flush(intel->batch); + + if ( intel_fb->pf_active ) { + swap.seqtype |= DRM_VBLANK_FLIP; + + intel_fb->pf_current_page = (((intel->sarea->pf_current_page >> + (intel_fb->pf_planes & 0x2)) & 0x3) + 1) % + intel_fb->pf_num_pages; + } + + if (!drmCommandWriteRead(intel->driFd, DRM_I915_VBLANK_SWAP, &swap, + sizeof(swap))) { + dPriv->vblSeq = swap.sequence; + swap.sequence -= target; + *missed_target = swap.sequence > 0 && swap.sequence <= (1 << 23); + + intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->vbl_pending = + intel_get_renderbuffer(&intel_fb->Base, + BUFFER_FRONT_LEFT)->vbl_pending = + dPriv->vblSeq; + + if (swap.seqtype & DRM_VBLANK_FLIP) { + intel_flip_renderbuffers(intel_fb); + intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer); + } + + ret = GL_TRUE; + } else { + if (swap.seqtype & DRM_VBLANK_FLIP) { + intel_fb->pf_current_page = ((intel->sarea->pf_current_page >> + (intel_fb->pf_planes & 0x2)) & 0x3) % + intel_fb->pf_num_pages; + } + + ret = GL_FALSE; + } + + UNLOCK_HARDWARE(intel); + + return ret; +} + +void +intelSwapBuffers(__DRIdrawablePrivate * dPriv) +{ + __DRIscreenPrivate *psp = dPriv->driScreenPriv; + + if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { + GET_CURRENT_CONTEXT(ctx); + struct intel_context *intel; + + if (ctx == NULL) + return; + + intel = intel_context(ctx); + + if (ctx->Visual.doubleBufferMode) { + GLboolean missed_target; + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + int64_t ust; + + _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ + + if (!intelScheduleSwap(dPriv, &missed_target)) { + driWaitForVBlank(dPriv, &missed_target); + + /* + * Update each buffer's vbl_pending so we don't get too out of + * sync + */ + intel_get_renderbuffer(&intel_fb->Base, + BUFFER_BACK_LEFT)->vbl_pending = + intel_get_renderbuffer(&intel_fb->Base, + BUFFER_FRONT_LEFT)->vbl_pending = + dPriv->vblSeq; + if (!intelPageFlip(dPriv)) { + intelCopyBuffer(dPriv, NULL); + } + } + + intel_fb->swap_count++; + (*psp->systemTime->getUST) (&ust); + if (missed_target) { + intel_fb->swap_missed_count++; + intel_fb->swap_missed_ust = ust - intel_fb->swap_ust; + } + + intel_fb->swap_ust = ust; + } + drmCommandNone(intel->driFd, DRM_I915_GEM_THROTTLE); + + } + else { + /* XXX this shouldn't be an error but we can't handle it for now */ + fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__); + } +} + +void +intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h) +{ + if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { + struct intel_context *intel = + (struct intel_context *) dPriv->driContextPriv->driverPrivate; + GLcontext *ctx = &intel->ctx; + + if (ctx->Visual.doubleBufferMode) { + drm_clip_rect_t rect; + rect.x1 = x + dPriv->x; + rect.y1 = (dPriv->h - y - h) + dPriv->y; + rect.x2 = rect.x1 + w; + rect.y2 = rect.y1 + h; + _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ + intelCopyBuffer(dPriv, &rect); + } + } + else { + /* XXX this shouldn't be an error but we can't handle it for now */ + fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__); + } +} + + +/** + * Update the hardware state for drawing into a window or framebuffer object. + * + * Called by glDrawBuffer, glBindFramebufferEXT, MakeCurrent, and other + * places within the driver. + * + * Basically, this needs to be called any time the current framebuffer + * changes, the renderbuffers change, or we need to draw into different + * color buffers. + */ +void +intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *colorRegions[MAX_DRAW_BUFFERS], *depthRegion = NULL; + struct intel_renderbuffer *irbDepth = NULL, *irbStencil = NULL; + int front = 0; /* drawing to front color buffer? */ + + if (!fb) { + /* this can happen during the initial context initialization */ + return; + } + + /* Do this here, note core Mesa, since this function is called from + * many places within the driver. + */ + if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */ + _mesa_update_framebuffer(ctx); + /* this updates the DrawBuffer's Width/Height if it's a FBO */ + _mesa_update_draw_buffer_bounds(ctx); + } + + if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) { + /* this may occur when we're called by glBindFrameBuffer() during + * the process of someone setting up renderbuffers, etc. + */ + /*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/ + return; + } + + if (fb->Name) + intel_validate_paired_depth_stencil(ctx, fb); + + /* If the batch contents require looping over cliprects, flush them before + * we go changing which cliprects get referenced when that happens. + */ + if (intel->batch->cliprect_mode == LOOP_CLIPRECTS) + intel_batchbuffer_flush(intel->batch); + + /* + * How many color buffers are we drawing into? + */ + if (fb->_NumColorDrawBuffers == 0) { + /* writing to 0 */ + FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE); + colorRegions[0] = NULL; + + if (fb->Name != 0) + intelSetRenderbufferClipRects(intel); + } else if (fb->_NumColorDrawBuffers > 1) { + int i; + struct intel_renderbuffer *irb; + FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE); + + if (fb->Name != 0) + intelSetRenderbufferClipRects(intel); + for (i = 0; i < fb->_NumColorDrawBuffers; i++) { + irb = intel_renderbuffer(fb->_ColorDrawBuffers[i]); + colorRegions[i] = (irb && irb->region) ? irb->region : NULL; + } + } + else { + /* draw to exactly one color buffer */ + /*_mesa_debug(ctx, "Hardware rendering\n");*/ + FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE); + if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) { + front = 1; + } + + /* + * Get the intel_renderbuffer for the colorbuffer we're drawing into. + * And set up cliprects. + */ + if (fb->Name == 0) { + /* drawing to window system buffer */ + if (front) { + intelSetFrontClipRects(intel); + colorRegions[0] = intel_get_rb_region(fb, BUFFER_FRONT_LEFT); + } + else { + intelSetBackClipRects(intel); + colorRegions[0]= intel_get_rb_region(fb, BUFFER_BACK_LEFT); + } + } + else { + /* drawing to user-created FBO */ + struct intel_renderbuffer *irb; + intelSetRenderbufferClipRects(intel); + irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]); + colorRegions[0] = (irb && irb->region) ? irb->region : NULL; + } + } + + /* Update culling direction which changes depending on the + * orientation of the buffer: + */ + if (ctx->Driver.FrontFace) + ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace); + else + ctx->NewState |= _NEW_POLYGON; + + if (!colorRegions[0]) { + FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE); + } + else { + FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE); + } + + /*** + *** Get depth buffer region and check if we need a software fallback. + *** Note that the depth buffer is usually a DEPTH_STENCIL buffer. + ***/ + if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) { + irbDepth = intel_renderbuffer(fb->_DepthBuffer->Wrapped); + if (irbDepth && irbDepth->region) { + FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE); + depthRegion = irbDepth->region; + } + else { + FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_TRUE); + depthRegion = NULL; + } + } + else { + /* not using depth buffer */ + FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE); + depthRegion = NULL; + } + + /*** + *** Stencil buffer + *** This can only be hardware accelerated if we're using a + *** combined DEPTH_STENCIL buffer (for now anyway). + ***/ + if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) { + irbStencil = intel_renderbuffer(fb->_StencilBuffer->Wrapped); + if (irbStencil && irbStencil->region) { + ASSERT(irbStencil->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE); + /* need to re-compute stencil hw state */ + if (ctx->Driver.Enable != NULL) + ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled); + else + ctx->NewState |= _NEW_STENCIL; + if (!depthRegion) + depthRegion = irbStencil->region; + } + else { + FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_TRUE); + } + } + else { + /* XXX FBO: instead of FALSE, pass ctx->Stencil.Enabled ??? */ + FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE); + /* need to re-compute stencil hw state */ + if (ctx->Driver.Enable != NULL) + ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled); + else + ctx->NewState |= _NEW_STENCIL; + } + + /* + * Update depth test state + */ + if (ctx->Driver.Enable) { + if (ctx->Depth.Test && fb->Visual.depthBits > 0) { + ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_TRUE); + } else { + ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_FALSE); + } + } else { + ctx->NewState |= _NEW_DEPTH; + } + + intel->vtbl.set_draw_region(intel, colorRegions, depthRegion, + fb->_NumColorDrawBuffers); + + /* update viewport since it depends on window size */ + if (ctx->Driver.Viewport) { + ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y, + ctx->Viewport.Width, ctx->Viewport.Height); + } else { + ctx->NewState |= _NEW_VIEWPORT; + } + + /* Set state we know depends on drawable parameters: + */ + if (ctx->Driver.Scissor) + ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y, + ctx->Scissor.Width, ctx->Scissor.Height); + intel->NewGLState |= _NEW_SCISSOR; + + if (ctx->Driver.DepthRange) + ctx->Driver.DepthRange(ctx, + ctx->Viewport.Near, + ctx->Viewport.Far); +} + + +static void +intelDrawBuffer(GLcontext * ctx, GLenum mode) +{ + intel_draw_buffer(ctx, ctx->DrawBuffer); +} + + +static void +intelReadBuffer(GLcontext * ctx, GLenum mode) +{ + if (ctx->ReadBuffer == ctx->DrawBuffer) { + /* This will update FBO completeness status. + * A framebuffer will be incomplete if the GL_READ_BUFFER setting + * refers to a missing renderbuffer. Calling glReadBuffer can set + * that straight and can make the drawing buffer complete. + */ + intel_draw_buffer(ctx, ctx->DrawBuffer); + } + /* Generally, functions which read pixels (glReadPixels, glCopyPixels, etc) + * reference ctx->ReadBuffer and do appropriate state checks. + */ +} + + +void +intelInitBufferFuncs(struct dd_function_table *functions) +{ + functions->Clear = intelClear; + functions->DrawBuffer = intelDrawBuffer; + functions->ReadBuffer = intelReadBuffer; +} diff --git a/src/mesa/drivers/dri/intel/intel_buffers.h b/src/mesa/drivers/dri/intel/intel_buffers.h new file mode 100644 index 0000000000..a669a85431 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_buffers.h @@ -0,0 +1,56 @@ + +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_BUFFERS_H +#define INTEL_BUFFERS_H + + +struct intel_context; +struct intel_framebuffer; + + +extern GLboolean +intel_intersect_cliprects(drm_clip_rect_t * dest, + const drm_clip_rect_t * a, + const drm_clip_rect_t * b); + +extern struct intel_region *intel_readbuf_region(struct intel_context *intel); + +extern struct intel_region *intel_drawbuf_region(struct intel_context *intel); + +extern void intel_wait_flips(struct intel_context *intel); + +extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv); + +extern void intelWindowMoved(struct intel_context *intel); + +extern void intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb); + +extern void intelInitBufferFuncs(struct dd_function_table *functions); + +#endif /* INTEL_BUFFERS_H */ diff --git a/src/mesa/drivers/dri/intel/intel_chipset.h b/src/mesa/drivers/dri/intel/intel_chipset.h new file mode 100644 index 0000000000..15b9ef4312 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_chipset.h @@ -0,0 +1,97 @@ +/* + * Copyright © 2007 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#define PCI_CHIP_I810 0x7121 +#define PCI_CHIP_I810_DC100 0x7123 +#define PCI_CHIP_I810_E 0x7125 +#define PCI_CHIP_I815 0x1132 + +#define PCI_CHIP_I830_M 0x3577 +#define PCI_CHIP_845_G 0x2562 +#define PCI_CHIP_I855_GM 0x3582 +#define PCI_CHIP_I865_G 0x2572 + +#define PCI_CHIP_I915_G 0x2582 +#define PCI_CHIP_E7221_G 0x258A +#define PCI_CHIP_I915_GM 0x2592 +#define PCI_CHIP_I945_G 0x2772 +#define PCI_CHIP_I945_GM 0x27A2 +#define PCI_CHIP_I945_GME 0x27AE + +#define PCI_CHIP_Q35_G 0x29B2 +#define PCI_CHIP_G33_G 0x29C2 +#define PCI_CHIP_Q33_G 0x29D2 + +#define PCI_CHIP_I965_G 0x29A2 +#define PCI_CHIP_I965_Q 0x2992 +#define PCI_CHIP_I965_G_1 0x2982 +#define PCI_CHIP_I946_GZ 0x2972 +#define PCI_CHIP_I965_GM 0x2A02 +#define PCI_CHIP_I965_GME 0x2A12 + +#define PCI_CHIP_GM45_GM 0x2A42 + +#define PCI_CHIP_IGD_E_G 0x2E02 +#define PCI_CHIP_Q45_G 0x2E12 +#define PCI_CHIP_G45_G 0x2E22 + +#define IS_MOBILE(devid) (devid == PCI_CHIP_I855_GM || \ + devid == PCI_CHIP_I915_GM || \ + devid == PCI_CHIP_I945_GM || \ + devid == PCI_CHIP_I945_GME || \ + devid == PCI_CHIP_I965_GM || \ + devid == PCI_CHIP_I965_GME || \ + devid == PCI_CHIP_GM45_GM) + +#define IS_GM45_GM(devid) (devid == PCI_CHIP_GM45_GM) +#define IS_G4X(devid) (devid == PCI_CHIP_IGD_E_G || \ + devid == PCI_CHIP_Q45_G || \ + devid == PCI_CHIP_G45_G) + +#define IS_915(devid) (devid == PCI_CHIP_I915_G || \ + devid == PCI_CHIP_E7221_G || \ + devid == PCI_CHIP_I915_GM) + +#define IS_945(devid) (devid == PCI_CHIP_I945_G || \ + devid == PCI_CHIP_I945_GM || \ + devid == PCI_CHIP_I945_GME || \ + devid == PCI_CHIP_G33_G || \ + devid == PCI_CHIP_Q33_G || \ + devid == PCI_CHIP_Q35_G) + +#define IS_965(devid) (devid == PCI_CHIP_I965_G || \ + devid == PCI_CHIP_I965_Q || \ + devid == PCI_CHIP_I965_G_1 || \ + devid == PCI_CHIP_I965_GM || \ + devid == PCI_CHIP_I965_GME || \ + devid == PCI_CHIP_I946_GZ || \ + IS_GM45_GM(devid) || \ + IS_G4X(devid)) + +#define IS_9XX(devid) (IS_915(devid) || \ + IS_945(devid) || \ + IS_965(devid)) diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c new file mode 100644 index 0000000000..4d1a742698 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_context.c @@ -0,0 +1,1089 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "glheader.h" +#include "context.h" +#include "matrix.h" +#include "simple_list.h" +#include "extensions.h" +#include "framebuffer.h" +#include "imports.h" +#include "points.h" + +#include "swrast/swrast.h" +#include "swrast_setup/swrast_setup.h" +#include "tnl/tnl.h" + +#include "tnl/t_pipeline.h" +#include "tnl/t_vertex.h" + +#include "drivers/common/driverfuncs.h" + +#include "intel_screen.h" + +#include "i830_dri.h" + +#include "intel_chipset.h" +#include "intel_buffers.h" +#include "intel_tex.h" +#include "intel_batchbuffer.h" +#include "intel_blit.h" +#include "intel_pixel.h" +#include "intel_regions.h" +#include "intel_buffer_objects.h" +#include "intel_fbo.h" +#include "intel_decode.h" +#include "intel_bufmgr.h" + +#include "drirenderbuffer.h" +#include "vblank.h" +#include "utils.h" +#include "xmlpool.h" /* for symbolic values of enum-type options */ +#ifndef INTEL_DEBUG +int INTEL_DEBUG = (0); +#endif + +#define need_GL_NV_point_sprite +#define need_GL_ARB_multisample +#define need_GL_ARB_point_parameters +#define need_GL_ARB_texture_compression +#define need_GL_ARB_vertex_buffer_object +#define need_GL_ARB_vertex_program +#define need_GL_ARB_window_pos +#define need_GL_ARB_occlusion_query +#define need_GL_EXT_blend_color +#define need_GL_EXT_blend_equation_separate +#define need_GL_EXT_blend_func_separate +#define need_GL_EXT_blend_minmax +#define need_GL_EXT_cull_vertex +#define need_GL_EXT_fog_coord +#define need_GL_EXT_framebuffer_object +#define need_GL_EXT_multi_draw_arrays +#define need_GL_EXT_secondary_color +#define need_GL_NV_vertex_program +#define need_GL_ATI_separate_stencil +#define need_GL_EXT_point_parameters +#define need_GL_VERSION_2_0 +#define need_GL_VERSION_2_1 +#define need_GL_ARB_shader_objects +#define need_GL_ARB_vertex_shader + +#include "extension_helper.h" + +#define DRIVER_DATE "20080716" +#define DRIVER_DATE_GEM "GEM " DRIVER_DATE + +static const GLubyte * +intelGetString(GLcontext * ctx, GLenum name) +{ + const struct intel_context *const intel = intel_context(ctx); + const char *chipset; + static char buffer[128]; + + switch (name) { + case GL_VENDOR: + return (GLubyte *) "Tungsten Graphics, Inc"; + break; + + case GL_RENDERER: + switch (intel->intelScreen->deviceID) { + case PCI_CHIP_845_G: + chipset = "Intel(R) 845G"; + break; + case PCI_CHIP_I830_M: + chipset = "Intel(R) 830M"; + break; + case PCI_CHIP_I855_GM: + chipset = "Intel(R) 852GM/855GM"; + break; + case PCI_CHIP_I865_G: + chipset = "Intel(R) 865G"; + break; + case PCI_CHIP_I915_G: + chipset = "Intel(R) 915G"; + break; + case PCI_CHIP_E7221_G: + chipset = "Intel (R) E7221G (i915)"; + break; + case PCI_CHIP_I915_GM: + chipset = "Intel(R) 915GM"; + break; + case PCI_CHIP_I945_G: + chipset = "Intel(R) 945G"; + break; + case PCI_CHIP_I945_GM: + chipset = "Intel(R) 945GM"; + break; + case PCI_CHIP_I945_GME: + chipset = "Intel(R) 945GME"; + break; + case PCI_CHIP_G33_G: + chipset = "Intel(R) G33"; + break; + case PCI_CHIP_Q35_G: + chipset = "Intel(R) Q35"; + break; + case PCI_CHIP_Q33_G: + chipset = "Intel(R) Q33"; + break; + case PCI_CHIP_I965_Q: + chipset = "Intel(R) 965Q"; + break; + case PCI_CHIP_I965_G: + case PCI_CHIP_I965_G_1: + chipset = "Intel(R) 965G"; + break; + case PCI_CHIP_I946_GZ: + chipset = "Intel(R) 946GZ"; + break; + case PCI_CHIP_I965_GM: + chipset = "Intel(R) 965GM"; + break; + case PCI_CHIP_I965_GME: + chipset = "Intel(R) 965GME/GLE"; + break; + case PCI_CHIP_GM45_GM: + chipset = "Mobile Intel® GM45 Express Chipset"; + break; + case PCI_CHIP_IGD_E_G: + chipset = "Intel(R) Integrated Graphics Device"; + break; + case PCI_CHIP_G45_G: + chipset = "Intel(R) G45/G43"; + break; + case PCI_CHIP_Q45_G: + chipset = "Intel(R) Q45/Q43"; + break; + default: + chipset = "Unknown Intel Chipset"; + break; + } + + (void) driGetRendererString(buffer, chipset, + (intel->ttm) ? DRIVER_DATE_GEM : DRIVER_DATE, + 0); + return (GLubyte *) buffer; + + default: + return NULL; + } +} + +void +intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) +{ + struct intel_framebuffer *intel_fb = drawable->driverPrivate; + struct intel_renderbuffer *rb; + struct intel_region *region, *depth_region; + struct intel_context *intel = context->driverPrivate; + __DRIbuffer *buffers; + __DRIscreen *screen; + int i, count; + unsigned int attachments[10]; + uint32_t name; + const char *region_name; + + if (INTEL_DEBUG & DEBUG_DRI) + fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable); + + screen = intel->intelScreen->driScrnPriv; + + i = 0; + if (intel_fb->color_rb[0]) + attachments[i++] = __DRI_BUFFER_FRONT_LEFT; + if (intel_fb->color_rb[1]) + attachments[i++] = __DRI_BUFFER_BACK_LEFT; + if (intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH)) + attachments[i++] = __DRI_BUFFER_DEPTH; + if (intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL)) + attachments[i++] = __DRI_BUFFER_STENCIL; + + buffers = (*screen->dri2.loader->getBuffers)(drawable, + &drawable->w, + &drawable->h, + attachments, i, + &count, + drawable->loaderPrivate); + + drawable->x = 0; + drawable->y = 0; + drawable->backX = 0; + drawable->backY = 0; + drawable->numClipRects = 1; + drawable->pClipRects[0].x1 = 0; + drawable->pClipRects[0].y1 = 0; + drawable->pClipRects[0].x2 = drawable->w; + drawable->pClipRects[0].y2 = drawable->h; + drawable->numBackClipRects = 1; + drawable->pBackClipRects[0].x1 = 0; + drawable->pBackClipRects[0].y1 = 0; + drawable->pBackClipRects[0].x2 = drawable->w; + drawable->pBackClipRects[0].y2 = drawable->h; + + depth_region = NULL; + for (i = 0; i < count; i++) { + switch (buffers[i].attachment) { + case __DRI_BUFFER_FRONT_LEFT: + rb = intel_fb->color_rb[0]; + region_name = "dri2 front buffer"; + break; + + case __DRI_BUFFER_BACK_LEFT: + rb = intel_fb->color_rb[1]; + region_name = "dri2 back buffer"; + break; + + case __DRI_BUFFER_DEPTH: + rb = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH); + region_name = "dri2 depth buffer"; + break; + + case __DRI_BUFFER_STENCIL: + rb = intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL); + region_name = "dri2 stencil buffer"; + break; + + case __DRI_BUFFER_ACCUM: + default: + fprintf(stderr, + "unhandled buffer attach event, attacment type %d\n", + buffers[i].attachment); + return; + } + + if (rb->region) { + dri_bo_flink(rb->region->buffer, &name); + if (name == buffers[i].name) + continue; + } + + if (INTEL_DEBUG & DEBUG_DRI) + fprintf(stderr, + "attaching buffer %d, at %d, cpp %d, pitch %d\n", + buffers[i].name, buffers[i].attachment, + buffers[i].cpp, buffers[i].pitch); + + if (buffers[i].attachment == __DRI_BUFFER_STENCIL && depth_region) { + if (INTEL_DEBUG & DEBUG_DRI) + fprintf(stderr, "(reusing depth buffer as stencil)\n"); + intel_region_reference(®ion, depth_region); + } + else + region = intel_region_alloc_for_handle(intel, buffers[i].cpp, + buffers[i].pitch / buffers[i].cpp, + drawable->h, + buffers[i].name, + region_name); + + if (buffers[i].attachment == __DRI_BUFFER_DEPTH) + depth_region = region; + + intel_renderbuffer_set_region(rb, region); + intel_region_release(®ion); + } + + driUpdateFramebufferSize(&intel->ctx, drawable); +} + +static void +intel_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei w, GLsizei h) +{ + struct intel_context *intel = intel_context(ctx); + __DRIcontext *driContext = intel->driContext; + + if (!driContext->driScreenPriv->dri2.enabled) + return; + + intel_update_renderbuffers(driContext, driContext->driDrawablePriv); + if (driContext->driDrawablePriv != driContext->driReadablePriv) + intel_update_renderbuffers(driContext, driContext->driReadablePriv); + + ctx->Driver.Viewport = NULL; + intel->driDrawable = driContext->driDrawablePriv; + intelWindowMoved(intel); + intel_draw_buffer(ctx, intel->ctx.DrawBuffer); + ctx->Driver.Viewport = intel_viewport; +} + +/** + * Extension strings exported by the intel driver. + * + * Extensions supported by all chips supported by i830_dri, i915_dri, or + * i965_dri. + */ +static const struct dri_extension card_extensions[] = { + {"GL_ARB_multisample", GL_ARB_multisample_functions}, + {"GL_ARB_multitexture", NULL}, + {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions}, + {"GL_NV_point_sprite", GL_NV_point_sprite_functions}, + {"GL_ARB_texture_border_clamp", NULL}, + {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions}, + {"GL_ARB_texture_cube_map", NULL}, + {"GL_ARB_texture_env_add", NULL}, + {"GL_ARB_texture_env_combine", NULL}, + {"GL_ARB_texture_env_crossbar", NULL}, + {"GL_ARB_texture_env_dot3", NULL}, + {"GL_ARB_texture_mirrored_repeat", NULL}, + {"GL_ARB_texture_non_power_of_two", NULL }, + {"GL_ARB_texture_rectangle", NULL}, + {"GL_NV_texture_rectangle", NULL}, + {"GL_EXT_texture_rectangle", NULL}, + {"GL_ARB_point_parameters", NULL}, + {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions}, + {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions}, + {"GL_ARB_window_pos", GL_ARB_window_pos_functions}, + {"GL_EXT_blend_color", GL_EXT_blend_color_functions}, + {"GL_EXT_blend_equation_separate", + GL_EXT_blend_equation_separate_functions}, + {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions}, + {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions}, + {"GL_EXT_blend_logic_op", NULL}, + {"GL_EXT_blend_subtract", NULL}, + {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions}, + {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions}, + {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions}, + {"GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions}, +#if 1 /* XXX FBO temporary? */ + {"GL_EXT_packed_depth_stencil", NULL}, +#endif + {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions}, + {"GL_EXT_stencil_wrap", NULL}, + {"GL_EXT_texture_edge_clamp", NULL}, + {"GL_EXT_texture_env_combine", NULL}, + {"GL_EXT_texture_env_dot3", NULL}, + {"GL_EXT_texture_filter_anisotropic", NULL}, + {"GL_EXT_texture_lod_bias", NULL}, + {"GL_3DFX_texture_compression_FXT1", NULL}, + {"GL_APPLE_client_storage", NULL}, + {"GL_MESA_pack_invert", NULL}, + {"GL_MESA_ycbcr_texture", NULL}, + {"GL_NV_blend_square", NULL}, + {"GL_NV_vertex_program", GL_NV_vertex_program_functions}, + {"GL_NV_vertex_program1_1", NULL}, + { "GL_SGIS_generate_mipmap", NULL }, + {NULL, NULL} +}; + +static const struct dri_extension brw_extensions[] = { + { "GL_ARB_shading_language_100", GL_VERSION_2_0_functions}, + { "GL_ARB_shading_language_120", GL_VERSION_2_1_functions}, + { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions}, + { "GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions}, + { "GL_ARB_point_sprite", NULL}, + { "GL_ARB_fragment_shader", NULL }, + { "GL_ARB_draw_buffers", NULL }, + { "GL_ARB_depth_texture", NULL }, + { "GL_ARB_fragment_program", NULL }, + { "GL_ARB_shadow", NULL }, + { "GL_EXT_shadow_funcs", NULL }, + { "GL_ARB_fragment_program_shadow", NULL }, + /* ARB extn won't work if not enabled */ + { "GL_SGIX_depth_texture", NULL }, + { "GL_EXT_texture_sRGB", NULL}, + { NULL, NULL } +}; + +#ifdef I915_MMIO_READ +static const struct dri_extension arb_oc_extensions[] = { + {"GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions}, + {NULL, NULL} +}; +#endif + +static const struct dri_extension ttm_extensions[] = { + {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions}, + {"GL_ARB_pixel_buffer_object", NULL}, + {NULL, NULL} +}; + +/** + * Initializes potential list of extensions if ctx == NULL, or actually enables + * extensions for a context. + */ +void intelInitExtensions(GLcontext *ctx, GLboolean enable_imaging) +{ + struct intel_context *intel = ctx?intel_context(ctx):NULL; + + /* Disable imaging extension until convolution is working in teximage paths. + */ + enable_imaging = GL_FALSE; + + driInitExtensions(ctx, card_extensions, enable_imaging); + + if (intel == NULL || intel->ttm) + driInitExtensions(ctx, ttm_extensions, GL_FALSE); + +#ifdef I915_MMIO_READ + if (intel == NULL || + (IS_965(intel->intelScreen->deviceID) && + intel->intelScreen->drmMinor >= 8)) + driInitExtensions(ctx, arb_oc_extensions, GL_FALSE); +#endif + + if (intel == NULL || IS_965(intel->intelScreen->deviceID)) + driInitExtensions(ctx, brw_extensions, GL_FALSE); +} + +static const struct dri_debug_control debug_control[] = { + { "tex", DEBUG_TEXTURE}, + { "state", DEBUG_STATE}, + { "ioctl", DEBUG_IOCTL}, + { "blit", DEBUG_BLIT}, + { "mip", DEBUG_MIPTREE}, + { "fall", DEBUG_FALLBACKS}, + { "verb", DEBUG_VERBOSE}, + { "bat", DEBUG_BATCH}, + { "pix", DEBUG_PIXEL}, + { "buf", DEBUG_BUFMGR}, + { "reg", DEBUG_REGION}, + { "fbo", DEBUG_FBO}, + { "lock", DEBUG_LOCK}, + { "sync", DEBUG_SYNC}, + { "prim", DEBUG_PRIMS }, + { "vert", DEBUG_VERTS }, + { "dri", DEBUG_DRI }, + { "dma", DEBUG_DMA }, + { "san", DEBUG_SANITY }, + { "sleep", DEBUG_SLEEP }, + { "stats", DEBUG_STATS }, + { "tile", DEBUG_TILE }, + { "sing", DEBUG_SINGLE_THREAD }, + { "thre", DEBUG_SINGLE_THREAD }, + { "wm", DEBUG_WM }, + { "urb", DEBUG_URB }, + { "vs", DEBUG_VS }, + { NULL, 0 } +}; + + +static void +intelInvalidateState(GLcontext * ctx, GLuint new_state) +{ + struct intel_context *intel = intel_context(ctx); + + _swrast_InvalidateState(ctx, new_state); + _swsetup_InvalidateState(ctx, new_state); + _vbo_InvalidateState(ctx, new_state); + _tnl_InvalidateState(ctx, new_state); + _tnl_invalidate_vertex_state(ctx, new_state); + + intel->NewGLState |= new_state; + + if (intel->vtbl.invalidate_state) + intel->vtbl.invalidate_state( intel, new_state ); +} + + +void +intelFlush(GLcontext * ctx) +{ + struct intel_context *intel = intel_context(ctx); + + if (intel->Fallback) + _swrast_flush(ctx); + + if (!IS_965(intel->intelScreen->deviceID)) + INTEL_FIREVERTICES(intel); + + /* Emit a flush so that any frontbuffer rendering that might have occurred + * lands onscreen in a timely manner, even if the X Server doesn't trigger + * a flush for us. + */ + intel_batchbuffer_emit_mi_flush(intel->batch); + + if (intel->batch->map != intel->batch->ptr) + intel_batchbuffer_flush(intel->batch); +} + +void +intelFinish(GLcontext * ctx) +{ + struct gl_framebuffer *fb = ctx->DrawBuffer; + int i; + + intelFlush(ctx); + + for (i = 0; i < fb->_NumColorDrawBuffers; i++) { + struct intel_renderbuffer *irb; + + irb = intel_renderbuffer(fb->_ColorDrawBuffers[i]); + + if (irb->region) + dri_bo_wait_rendering(irb->region->buffer); + } + if (fb->_DepthBuffer) { + /* XXX: Wait on buffer idle */ + } +} + +#ifdef I915_MMIO_READ +static void +intelBeginQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q) +{ + struct intel_context *intel = intel_context( ctx ); + struct drm_i915_mmio io = { + .read_write = I915_MMIO_READ, + .reg = MMIO_REGS_PS_DEPTH_COUNT, + .data = &q->Result + }; + intel->stats_wm++; + intelFinish(&intel->ctx); + drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io)); +} + +static void +intelEndQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q) +{ + struct intel_context *intel = intel_context( ctx ); + GLuint64EXT tmp; + struct drm_i915_mmio io = { + .read_write = I915_MMIO_READ, + .reg = MMIO_REGS_PS_DEPTH_COUNT, + .data = &tmp + }; + intelFinish(&intel->ctx); + drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io)); + q->Result = tmp - q->Result; + q->Ready = GL_TRUE; + intel->stats_wm--; +} +#endif + +void +intelInitDriverFunctions(struct dd_function_table *functions) +{ + _mesa_init_driver_functions(functions); + + functions->Flush = intelFlush; + functions->Finish = intelFinish; + functions->GetString = intelGetString; + functions->UpdateState = intelInvalidateState; + functions->Viewport = intel_viewport; + + functions->CopyColorTable = _swrast_CopyColorTable; + functions->CopyColorSubTable = _swrast_CopyColorSubTable; + functions->CopyConvolutionFilter1D = _swrast_CopyConvolutionFilter1D; + functions->CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D; + +#ifdef I915_MMIO_READ + functions->BeginQuery = intelBeginQuery; + functions->EndQuery = intelEndQuery; +#endif + + intelInitTextureFuncs(functions); + intelInitStateFuncs(functions); + intelInitBufferFuncs(functions); + intelInitPixelFuncs(functions); +} + + +GLboolean +intelInitContext(struct intel_context *intel, + const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate, + struct dd_function_table *functions) +{ + GLcontext *ctx = &intel->ctx; + GLcontext *shareCtx = (GLcontext *) sharedContextPrivate; + __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; + intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private; + int fthrottle_mode; + + if (!_mesa_initialize_context(&intel->ctx, mesaVis, shareCtx, + functions, (void *) intel)) { + _mesa_printf("%s: failed to init mesa context\n", __FUNCTION__); + return GL_FALSE; + } + + driContextPriv->driverPrivate = intel; + intel->intelScreen = intelScreen; + intel->driScreen = sPriv; + intel->sarea = intelScreen->sarea; + intel->driContext = driContextPriv; + + /* Dri stuff */ + intel->hHWContext = driContextPriv->hHWContext; + intel->driFd = sPriv->fd; + intel->driHwLock = sPriv->lock; + + intel->width = intelScreen->width; + intel->height = intelScreen->height; + + driParseConfigFiles(&intel->optionCache, &intelScreen->optionCache, + intel->driScreen->myNum, + IS_965(intelScreen->deviceID) ? "i965" : "i915"); + if (intelScreen->deviceID == PCI_CHIP_I865_G) + intel->maxBatchSize = 4096; + else + intel->maxBatchSize = BATCH_SZ; + + intel->bufmgr = intelScreen->bufmgr; + intel->ttm = intelScreen->ttm; + if (intel->ttm) { + int bo_reuse_mode; + + bo_reuse_mode = driQueryOptioni(&intel->optionCache, "bo_reuse"); + switch (bo_reuse_mode) { + case DRI_CONF_BO_REUSE_DISABLED: + break; + case DRI_CONF_BO_REUSE_ALL: + intel_bufmgr_gem_enable_reuse(intel->bufmgr); + break; + } + } + + ctx->Const.MaxTextureMaxAnisotropy = 2.0; + + /* This doesn't yet catch all non-conformant rendering, but it's a + * start. + */ + if (getenv("INTEL_STRICT_CONFORMANCE")) { + intel->strict_conformance = 1; + } + + if (intel->strict_conformance) { + ctx->Const.MinLineWidth = 1.0; + ctx->Const.MinLineWidthAA = 1.0; + ctx->Const.MaxLineWidth = 1.0; + ctx->Const.MaxLineWidthAA = 1.0; + ctx->Const.LineWidthGranularity = 1.0; + } + else { + ctx->Const.MinLineWidth = 1.0; + ctx->Const.MinLineWidthAA = 1.0; + ctx->Const.MaxLineWidth = 5.0; + ctx->Const.MaxLineWidthAA = 5.0; + ctx->Const.LineWidthGranularity = 0.5; + } + + ctx->Const.MinPointSize = 1.0; + ctx->Const.MinPointSizeAA = 1.0; + ctx->Const.MaxPointSize = 255.0; + ctx->Const.MaxPointSizeAA = 3.0; + ctx->Const.PointSizeGranularity = 1.0; + + /* reinitialize the context point state. + * It depend on constants in __GLcontextRec::Const + */ + _mesa_init_point(ctx); + + ctx->Const.MaxColorAttachments = 4; /* XXX FBO: review this */ + + /* Initialize the software rasterizer and helper modules. */ + _swrast_CreateContext(ctx); + _vbo_CreateContext(ctx); + _tnl_CreateContext(ctx); + _swsetup_CreateContext(ctx); + + /* Configure swrast to match hardware characteristics: */ + _swrast_allow_pixel_fog(ctx, GL_FALSE); + _swrast_allow_vertex_fog(ctx, GL_TRUE); + + intel->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24; + intel->hw_stipple = 1; + + /* XXX FBO: this doesn't seem to be used anywhere */ + switch (mesaVis->depthBits) { + case 0: /* what to do in this case? */ + case 16: + intel->polygon_offset_scale = 1.0; + break; + case 24: + intel->polygon_offset_scale = 2.0; /* req'd to pass glean */ + break; + default: + assert(0); + break; + } + + if (IS_965(intelScreen->deviceID)) + intel->polygon_offset_scale /= 0xffff; + + intel->RenderIndex = ~0; + + fthrottle_mode = driQueryOptioni(&intel->optionCache, "fthrottle_mode"); + intel->irqsEmitted = 0; + + intel->do_irqs = (intel->intelScreen->irq_active && + fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS); + + intel->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS); + + _math_matrix_ctr(&intel->ViewportMatrix); + + if (IS_965(intelScreen->deviceID) && !intel->intelScreen->irq_active) { + _mesa_printf("IRQs not active. Exiting\n"); + exit(1); + } + + intelInitExtensions(ctx, GL_FALSE); + + INTEL_DEBUG = driParseDebugString(getenv("INTEL_DEBUG"), debug_control); + if (INTEL_DEBUG & DEBUG_BUFMGR) + dri_bufmgr_set_debug(intel->bufmgr, GL_TRUE); + + if (!sPriv->dri2.enabled) + intel_recreate_static_regions(intel); + + intel->batch = intel_batchbuffer_alloc(intel); + + intel_bufferobj_init(intel); + intel_fbo_init(intel); + + if (intel->ctx.Mesa_DXTn) { + _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); + _mesa_enable_extension(ctx, "GL_S3_s3tc"); + } + else if (driQueryOptionb(&intel->optionCache, "force_s3tc_enable")) { + _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); + } + + intel->prim.primitive = ~0; + + /* Force all software fallbacks */ + if (driQueryOptionb(&intel->optionCache, "no_rast")) { + fprintf(stderr, "disabling 3D rasterization\n"); + intel->no_rast = 1; + } + + /* Disable all hardware rendering (skip emitting batches and fences/waits + * to the kernel) + */ + intel->no_hw = getenv("INTEL_NO_HW") != NULL; + + return GL_TRUE; +} + +void +intelDestroyContext(__DRIcontextPrivate * driContextPriv) +{ + struct intel_context *intel = + (struct intel_context *) driContextPriv->driverPrivate; + + assert(intel); /* should never be null */ + if (intel) { + GLboolean release_texture_heaps; + + INTEL_FIREVERTICES(intel); + + intel->vtbl.destroy(intel); + + release_texture_heaps = (intel->ctx.Shared->RefCount == 1); + _swsetup_DestroyContext(&intel->ctx); + _tnl_DestroyContext(&intel->ctx); + _vbo_DestroyContext(&intel->ctx); + + _swrast_DestroyContext(&intel->ctx); + intel->Fallback = 0; /* don't call _swrast_Flush later */ + + intel_batchbuffer_free(intel->batch); + free(intel->prim.vb); + + if (release_texture_heaps) { + /* This share group is about to go away, free our private + * texture object data. + */ + if (INTEL_DEBUG & DEBUG_TEXTURE) + fprintf(stderr, "do something to free texture heaps\n"); + } + + /* free the Mesa context */ + _mesa_free_context_data(&intel->ctx); + + dri_bufmgr_destroy(intel->bufmgr); + } +} + +GLboolean +intelUnbindContext(__DRIcontextPrivate * driContextPriv) +{ + return GL_TRUE; +} + +GLboolean +intelMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv) +{ + __DRIscreenPrivate *psp = driDrawPriv->driScreenPriv; + + if (driContextPriv) { + struct intel_context *intel = + (struct intel_context *) driContextPriv->driverPrivate; + struct intel_framebuffer *intel_fb = + (struct intel_framebuffer *) driDrawPriv->driverPrivate; + GLframebuffer *readFb = (GLframebuffer *) driReadPriv->driverPrivate; + + if (driContextPriv->driScreenPriv->dri2.enabled) { + intel_update_renderbuffers(driContextPriv, driDrawPriv); + if (driDrawPriv != driReadPriv) + intel_update_renderbuffers(driContextPriv, driReadPriv); + } else { + /* XXX FBO temporary fix-ups! */ + /* if the renderbuffers don't have regions, init them from the context */ + struct intel_renderbuffer *irbDepth + = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH); + struct intel_renderbuffer *irbStencil + = intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL); + + if (intel_fb->color_rb[0]) { + intel_renderbuffer_set_region(intel_fb->color_rb[0], + intel->front_region); + } + if (intel_fb->color_rb[1]) { + intel_renderbuffer_set_region(intel_fb->color_rb[1], + intel->back_region); + } +#if 0 + if (intel_fb->color_rb[2]) { + intel_renderbuffer_set_region(intel_fb->color_rb[2], + intel->third_region); + } +#endif + if (irbDepth) { + intel_renderbuffer_set_region(irbDepth, intel->depth_region); + } + if (irbStencil) { + intel_renderbuffer_set_region(irbStencil, intel->depth_region); + } + } + + /* set GLframebuffer size to match window, if needed */ + driUpdateFramebufferSize(&intel->ctx, driDrawPriv); + + if (driReadPriv != driDrawPriv) { + driUpdateFramebufferSize(&intel->ctx, driReadPriv); + } + + _mesa_make_current(&intel->ctx, &intel_fb->Base, readFb); + + /* The drawbuffer won't always be updated by _mesa_make_current: + */ + if (intel->ctx.DrawBuffer == &intel_fb->Base) { + + if (intel->driReadDrawable != driReadPriv) + intel->driReadDrawable = driReadPriv; + + if (intel->driDrawable != driDrawPriv) { + if (driDrawPriv->swap_interval == (unsigned)-1) { + int i; + + driDrawPriv->vblFlags = (intel->intelScreen->irq_active != 0) + ? driGetDefaultVBlankFlags(&intel->optionCache) + : VBLANK_FLAG_NO_IRQ; + + (*psp->systemTime->getUST) (&intel_fb->swap_ust); + driDrawableInitVBlank(driDrawPriv); + intel_fb->vbl_waited = driDrawPriv->vblSeq; + + for (i = 0; i < (intel->intelScreen->third.handle ? 3 : 2); i++) { + if (intel_fb->color_rb[i]) + intel_fb->color_rb[i]->vbl_pending = driDrawPriv->vblSeq; + } + } + intel->driDrawable = driDrawPriv; + intelWindowMoved(intel); + } + + intel_draw_buffer(&intel->ctx, &intel_fb->Base); + } + } + else { + _mesa_make_current(NULL, NULL, NULL); + } + + return GL_TRUE; +} + +static void +intelContendedLock(struct intel_context *intel, GLuint flags) +{ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + __DRIscreenPrivate *sPriv = intel->driScreen; + volatile struct drm_i915_sarea *sarea = intel->sarea; + int me = intel->hHWContext; + + drmGetLock(intel->driFd, intel->hHWContext, flags); + intel->locked = 1; + + if (INTEL_DEBUG & DEBUG_LOCK) + _mesa_printf("%s - got contended lock\n", __progname); + + /* If the window moved, may need to set a new cliprect now. + * + * NOTE: This releases and regains the hw lock, so all state + * checking must be done *after* this call: + */ + if (dPriv) + DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv); + + if (sarea && sarea->ctxOwner != me) { + if (INTEL_DEBUG & DEBUG_BUFMGR) { + fprintf(stderr, "Lost Context: sarea->ctxOwner %x me %x\n", + sarea->ctxOwner, me); + } + sarea->ctxOwner = me; + } + + /* If the last consumer of the texture memory wasn't us, notify the fake + * bufmgr and record the new owner. We should have the memory shared + * between contexts of a single fake bufmgr, but this will at least make + * things correct for now. + */ + if (!intel->ttm && sarea->texAge != intel->hHWContext) { + sarea->texAge = intel->hHWContext; + intel_bufmgr_fake_contended_lock_take(intel->bufmgr); + if (INTEL_DEBUG & DEBUG_BATCH) + intel_decode_context_reset(); + if (INTEL_DEBUG & DEBUG_BUFMGR) + fprintf(stderr, "Lost Textures: sarea->texAge %x hw context %x\n", + sarea->ctxOwner, intel->hHWContext); + } + + if (sarea->width != intel->width || sarea->height != intel->height) { + int numClipRects = intel->numClipRects; + + /* + * FIXME: Really only need to do this when drawing to a + * common back- or front buffer. + */ + + /* + * This will essentially drop the outstanding batchbuffer on + * the floor. + */ + intel->numClipRects = 0; + + if (intel->Fallback) + _swrast_flush(&intel->ctx); + + if (!IS_965(intel->intelScreen->deviceID)) + INTEL_FIREVERTICES(intel); + + if (intel->batch->map != intel->batch->ptr) + intel_batchbuffer_flush(intel->batch); + + intel->numClipRects = numClipRects; + + /* force window update */ + intel->lastStamp = 0; + + intel->width = sarea->width; + intel->height = sarea->height; + } + + /* Drawable changed? + */ + if (dPriv && intel->lastStamp != dPriv->lastStamp) { + intelWindowMoved(intel); + intel->lastStamp = dPriv->lastStamp; + } +} + + +_glthread_DECLARE_STATIC_MUTEX(lockMutex); + +/* Lock the hardware and validate our state. + */ +void LOCK_HARDWARE( struct intel_context *intel ) +{ + __DRIdrawable *dPriv = intel->driDrawable; + __DRIscreen *sPriv = intel->driScreen; + char __ret = 0; + struct intel_framebuffer *intel_fb = NULL; + struct intel_renderbuffer *intel_rb = NULL; + + _glthread_LOCK_MUTEX(lockMutex); + assert(!intel->locked); + intel->locked = 1; + + if (intel->driDrawable) { + intel_fb = intel->driDrawable->driverPrivate; + + if (intel_fb) + intel_rb = + intel_get_renderbuffer(&intel_fb->Base, + intel_fb->Base._ColorDrawBufferIndexes[0]); + } + + if (intel_rb && dPriv->vblFlags && + !(dPriv->vblFlags & VBLANK_FLAG_NO_IRQ) && + (intel_fb->vbl_waited - intel_rb->vbl_pending) > (1<<23)) { + drmVBlank vbl; + + vbl.request.type = DRM_VBLANK_ABSOLUTE; + + if ( dPriv->vblFlags & VBLANK_FLAG_SECONDARY ) { + vbl.request.type |= DRM_VBLANK_SECONDARY; + } + + vbl.request.sequence = intel_rb->vbl_pending; + drmWaitVBlank(intel->driFd, &vbl); + intel_fb->vbl_waited = vbl.reply.sequence; + } + + if (!sPriv->dri2.enabled) { + DRM_CAS(intel->driHwLock, intel->hHWContext, + (DRM_LOCK_HELD|intel->hHWContext), __ret); + + if (__ret) + intelContendedLock( intel, 0 ); + } + + + if (INTEL_DEBUG & DEBUG_LOCK) + _mesa_printf("%s - locked\n", __progname); +} + + +/* Unlock the hardware using the global current context + */ +void UNLOCK_HARDWARE( struct intel_context *intel ) +{ + __DRIscreen *sPriv = intel->driScreen; + + intel->vtbl.note_unlock( intel ); + intel->locked = 0; + + if (!sPriv->dri2.enabled) + DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext); + + _glthread_UNLOCK_MUTEX(lockMutex); + + if (INTEL_DEBUG & DEBUG_LOCK) + _mesa_printf("%s - unlocked\n", __progname); + + /** + * Nothing should be left in batch outside of LOCK/UNLOCK which references + * cliprects. + */ + if (intel->batch->cliprect_mode == REFERENCES_CLIPRECTS) + intel_batchbuffer_flush(intel->batch); +} + diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h new file mode 100644 index 0000000000..8bd75753ea --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_context.h @@ -0,0 +1,508 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTELCONTEXT_INC +#define INTELCONTEXT_INC + + + +#include "mtypes.h" +#include "drm.h" +#include "mm.h" +#include "texmem.h" +#include "intel_bufmgr.h" + +#include "intel_screen.h" +#include "intel_tex_obj.h" +#include "i915_drm.h" +#include "tnl/t_vertex.h" + +#define TAG(x) intel##x +#include "tnl_dd/t_dd_vertex.h" +#undef TAG + +#define DV_PF_555 (1<<8) +#define DV_PF_565 (2<<8) +#define DV_PF_8888 (3<<8) + +struct intel_region; +struct intel_context; + +typedef void (*intel_tri_func) (struct intel_context *, intelVertex *, + intelVertex *, intelVertex *); +typedef void (*intel_line_func) (struct intel_context *, intelVertex *, + intelVertex *); +typedef void (*intel_point_func) (struct intel_context *, intelVertex *); + +#define INTEL_FALLBACK_DRAW_BUFFER 0x1 +#define INTEL_FALLBACK_READ_BUFFER 0x2 +#define INTEL_FALLBACK_DEPTH_BUFFER 0x4 +#define INTEL_FALLBACK_STENCIL_BUFFER 0x8 +#define INTEL_FALLBACK_USER 0x10 +#define INTEL_FALLBACK_RENDERMODE 0x20 +#define INTEL_FALLBACK_TEXTURE 0x40 + +extern void intelFallback(struct intel_context *intel, GLuint bit, + GLboolean mode); +#define FALLBACK( intel, bit, mode ) intelFallback( intel, bit, mode ) + + +#define INTEL_WRITE_PART 0x1 +#define INTEL_WRITE_FULL 0x2 +#define INTEL_READ 0x4 + +#define INTEL_MAX_FIXUP 64 + +struct intel_context +{ + GLcontext ctx; /* the parent class */ + + struct + { + void (*destroy) (struct intel_context * intel); + void (*emit_state) (struct intel_context * intel); + void (*finish_batch) (struct intel_context * intel); + void (*new_batch) (struct intel_context * intel); + void (*emit_invarient_state) (struct intel_context * intel); + void (*note_fence) (struct intel_context *intel, GLuint fence); + void (*note_unlock) (struct intel_context *intel); + void (*update_texture_state) (struct intel_context * intel); + + void (*render_start) (struct intel_context * intel); + void (*render_prevalidate) (struct intel_context * intel); + void (*set_draw_region) (struct intel_context * intel, + struct intel_region * draw_regions[], + struct intel_region * depth_region, + GLuint num_regions); + + GLuint (*flush_cmd) (void); + void (*emit_flush) (struct intel_context *intel, GLuint unused); + + void (*reduced_primitive_state) (struct intel_context * intel, + GLenum rprim); + + GLboolean (*check_vertex_size) (struct intel_context * intel, + GLuint expected); + void (*invalidate_state) (struct intel_context *intel, + GLuint new_state); + + + /* Metaops: + */ + void (*install_meta_state) (struct intel_context * intel); + void (*leave_meta_state) (struct intel_context * intel); + + void (*meta_draw_region) (struct intel_context * intel, + struct intel_region * draw_region, + struct intel_region * depth_region); + + void (*meta_draw_quad)(struct intel_context *intel, + GLfloat x0, GLfloat x1, + GLfloat y0, GLfloat y1, + GLfloat z, + GLuint color, /* ARGB32 */ + GLfloat s0, GLfloat s1, + GLfloat t0, GLfloat t1); + + void (*meta_color_mask) (struct intel_context * intel, GLboolean); + + void (*meta_stencil_replace) (struct intel_context * intel, + GLuint mask, GLuint clear); + + void (*meta_depth_replace) (struct intel_context * intel); + + void (*meta_texture_blend_replace) (struct intel_context * intel); + + void (*meta_no_stencil_write) (struct intel_context * intel); + void (*meta_no_depth_write) (struct intel_context * intel); + void (*meta_no_texture) (struct intel_context * intel); + + void (*meta_import_pixel_state) (struct intel_context * intel); + void (*meta_frame_buffer_texture) (struct intel_context *intel, + GLint xoff, GLint yoff); + + GLboolean(*meta_tex_rect_source) (struct intel_context * intel, + dri_bo * buffer, + GLuint offset, + GLuint pitch, + GLuint height, + GLenum format, GLenum type); + + void (*assert_not_dirty) (struct intel_context *intel); + + void (*debug_batch)(struct intel_context *intel); + } vtbl; + + GLint refcount; + GLuint Fallback; + GLuint NewGLState; + + dri_bufmgr *bufmgr; + unsigned int maxBatchSize; + + struct intel_region *front_region; + struct intel_region *back_region; + struct intel_region *third_region; + struct intel_region *depth_region; + + /** + * This value indicates that the kernel memory manager is being used + * instead of the fake client-side memory manager. + */ + GLboolean ttm; + + struct intel_batchbuffer *batch; + GLboolean no_batch_wrap; + unsigned batch_id; + + struct + { + GLuint id; + uint32_t primitive; /**< Current hardware primitive type */ + void (*flush) (struct intel_context *); + dri_bo *vb_bo; + uint8_t *vb; + unsigned int start_offset; /**< Byte offset of primitive sequence */ + unsigned int current_offset; /**< Byte offset of next vertex */ + unsigned int count; /**< Number of vertices in current primitive */ + } prim; + + GLuint stats_wm; + GLboolean locked; + char *prevLockFile; + int prevLockLine; + + GLubyte clear_chan[4]; + GLuint ClearColor565; + GLuint ClearColor8888; + + /* Offsets of fields within the current vertex: + */ + GLuint coloroffset; + GLuint specoffset; + GLuint wpos_offset; + GLuint wpos_size; + + struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX]; + GLuint vertex_attr_count; + + GLfloat polygon_offset_scale; /* dependent on depth_scale, bpp */ + + GLboolean hw_stencil; + GLboolean hw_stipple; + GLboolean depth_buffer_is_float; + GLboolean no_rast; + GLboolean strict_conformance; + + /* State for intelvb.c and inteltris.c. + */ + GLuint RenderIndex; + GLmatrix ViewportMatrix; + GLenum render_primitive; + GLenum reduced_primitive; + GLuint vertex_size; + GLubyte *verts; /* points to tnl->clipspace.vertex_buf */ + struct intel_region *draw_region; + + /* Fallback rasterization functions + */ + intel_point_func draw_point; + intel_line_func draw_line; + intel_tri_func draw_tri; + + /* These refer to the current drawing buffer: + */ + int drawX, drawY; /**< origin of drawing area within region */ + GLuint numClipRects; /**< cliprects for drawing */ + drm_clip_rect_t *pClipRects; + struct gl_texture_object *frame_buffer_texobj; + drm_clip_rect_t fboRect; /**< cliprect for FBO rendering */ + + int perf_boxes; + + GLuint do_usleeps; + int do_irqs; + GLuint irqsEmitted; + + GLboolean scissor; + drm_clip_rect_t draw_rect; + drm_clip_rect_t scissor_rect; + + drm_context_t hHWContext; + drmLock *driHwLock; + int driFd; + + __DRIcontextPrivate *driContext; + __DRIdrawablePrivate *driDrawable; + __DRIdrawablePrivate *driReadDrawable; + __DRIscreenPrivate *driScreen; + intelScreenPrivate *intelScreen; + volatile struct drm_i915_sarea *sarea; + + GLuint lastStamp; + + GLboolean no_hw; + + /** + * Configuration cache + */ + driOptionCache optionCache; + + /* Last seen width/height of the screen */ + int width; + int height; + + int64_t swap_ust; + int64_t swap_missed_ust; + + GLuint swap_count; + GLuint swap_missed_count; +}; + +/* These are functions now: + */ +void LOCK_HARDWARE( struct intel_context *intel ); +void UNLOCK_HARDWARE( struct intel_context *intel ); + +extern char *__progname; + + +#define SUBPIXEL_X 0.125 +#define SUBPIXEL_Y 0.125 + +#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0])) +#define ALIGN(value, alignment) ((value + alignment - 1) & ~(alignment - 1)) + +#define INTEL_FIREVERTICES(intel) \ +do { \ + if ((intel)->prim.flush) \ + (intel)->prim.flush(intel); \ +} while (0) + +/* ================================================================ + * Color packing: + */ + +#define INTEL_PACKCOLOR4444(r,g,b,a) \ + ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4)) + +#define INTEL_PACKCOLOR1555(r,g,b,a) \ + ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \ + ((a) ? 0x8000 : 0)) + +#define INTEL_PACKCOLOR565(r,g,b) \ + ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3)) + +#define INTEL_PACKCOLOR8888(r,g,b,a) \ + ((a<<24) | (r<<16) | (g<<8) | b) + +#define INTEL_PACKCOLOR(format, r, g, b, a) \ +(format == DV_PF_555 ? INTEL_PACKCOLOR1555(r,g,b,a) : \ + (format == DV_PF_565 ? INTEL_PACKCOLOR565(r,g,b) : \ + (format == DV_PF_8888 ? INTEL_PACKCOLOR8888(r,g,b,a) : \ + 0))) + +/* ================================================================ + * From linux kernel i386 header files, copes with odd sizes better + * than COPY_DWORDS would: + * XXX Put this in src/mesa/main/imports.h ??? + */ +#if defined(i386) || defined(__i386__) +static INLINE void * __memcpy(void * to, const void * from, size_t n) +{ + int d0, d1, d2; + __asm__ __volatile__( + "rep ; movsl\n\t" + "testb $2,%b4\n\t" + "je 1f\n\t" + "movsw\n" + "1:\ttestb $1,%b4\n\t" + "je 2f\n\t" + "movsb\n" + "2:" + : "=&c" (d0), "=&D" (d1), "=&S" (d2) + :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from) + : "memory"); + return (to); +} +#else +#define __memcpy(a,b,c) memcpy(a,b,c) +#endif + + +/* ================================================================ + * Debugging: + */ +extern int INTEL_DEBUG; + +#define DEBUG_TEXTURE 0x1 +#define DEBUG_STATE 0x2 +#define DEBUG_IOCTL 0x4 +#define DEBUG_BLIT 0x8 +#define DEBUG_MIPTREE 0x10 +#define DEBUG_FALLBACKS 0x20 +#define DEBUG_VERBOSE 0x40 +#define DEBUG_BATCH 0x80 +#define DEBUG_PIXEL 0x100 +#define DEBUG_BUFMGR 0x200 +#define DEBUG_REGION 0x400 +#define DEBUG_FBO 0x800 +#define DEBUG_LOCK 0x1000 +#define DEBUG_SYNC 0x2000 +#define DEBUG_PRIMS 0x4000 +#define DEBUG_VERTS 0x8000 +#define DEBUG_DRI 0x10000 +#define DEBUG_DMA 0x20000 +#define DEBUG_SANITY 0x40000 +#define DEBUG_SLEEP 0x80000 +#define DEBUG_STATS 0x100000 +#define DEBUG_TILE 0x200000 +#define DEBUG_SINGLE_THREAD 0x400000 +#define DEBUG_WM 0x800000 +#define DEBUG_URB 0x1000000 +#define DEBUG_VS 0x2000000 + +#define DBG(...) do { \ + if (INTEL_DEBUG & FILE_DEBUG_FLAG) \ + _mesa_printf(__VA_ARGS__); \ +} while(0) + +#define PCI_CHIP_845_G 0x2562 +#define PCI_CHIP_I830_M 0x3577 +#define PCI_CHIP_I855_GM 0x3582 +#define PCI_CHIP_I865_G 0x2572 +#define PCI_CHIP_I915_G 0x2582 +#define PCI_CHIP_I915_GM 0x2592 +#define PCI_CHIP_I945_G 0x2772 +#define PCI_CHIP_I945_GM 0x27A2 +#define PCI_CHIP_I945_GME 0x27AE +#define PCI_CHIP_G33_G 0x29C2 +#define PCI_CHIP_Q35_G 0x29B2 +#define PCI_CHIP_Q33_G 0x29D2 + + +/* ================================================================ + * intel_context.c: + */ + +extern GLboolean intelInitContext(struct intel_context *intel, + const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate, + struct dd_function_table *functions); + +extern void intelGetLock(struct intel_context *intel, GLuint flags); + +extern void intelFinish(GLcontext * ctx); +extern void intelFlush(GLcontext * ctx); + +extern void intelInitDriverFunctions(struct dd_function_table *functions); +extern void intelInitExtensions(GLcontext *ctx, GLboolean enable_imaging); + + +/* ================================================================ + * intel_state.c: + */ +extern void intelInitStateFuncs(struct dd_function_table *functions); + +#define COMPAREFUNC_ALWAYS 0 +#define COMPAREFUNC_NEVER 0x1 +#define COMPAREFUNC_LESS 0x2 +#define COMPAREFUNC_EQUAL 0x3 +#define COMPAREFUNC_LEQUAL 0x4 +#define COMPAREFUNC_GREATER 0x5 +#define COMPAREFUNC_NOTEQUAL 0x6 +#define COMPAREFUNC_GEQUAL 0x7 + +#define STENCILOP_KEEP 0 +#define STENCILOP_ZERO 0x1 +#define STENCILOP_REPLACE 0x2 +#define STENCILOP_INCRSAT 0x3 +#define STENCILOP_DECRSAT 0x4 +#define STENCILOP_INCR 0x5 +#define STENCILOP_DECR 0x6 +#define STENCILOP_INVERT 0x7 + +#define LOGICOP_CLEAR 0 +#define LOGICOP_NOR 0x1 +#define LOGICOP_AND_INV 0x2 +#define LOGICOP_COPY_INV 0x3 +#define LOGICOP_AND_RVRSE 0x4 +#define LOGICOP_INV 0x5 +#define LOGICOP_XOR 0x6 +#define LOGICOP_NAND 0x7 +#define LOGICOP_AND 0x8 +#define LOGICOP_EQUIV 0x9 +#define LOGICOP_NOOP 0xa +#define LOGICOP_OR_INV 0xb +#define LOGICOP_COPY 0xc +#define LOGICOP_OR_RVRSE 0xd +#define LOGICOP_OR 0xe +#define LOGICOP_SET 0xf + +#define BLENDFACT_ZERO 0x01 +#define BLENDFACT_ONE 0x02 +#define BLENDFACT_SRC_COLR 0x03 +#define BLENDFACT_INV_SRC_COLR 0x04 +#define BLENDFACT_SRC_ALPHA 0x05 +#define BLENDFACT_INV_SRC_ALPHA 0x06 +#define BLENDFACT_DST_ALPHA 0x07 +#define BLENDFACT_INV_DST_ALPHA 0x08 +#define BLENDFACT_DST_COLR 0x09 +#define BLENDFACT_INV_DST_COLR 0x0a +#define BLENDFACT_SRC_ALPHA_SATURATE 0x0b +#define BLENDFACT_CONST_COLOR 0x0c +#define BLENDFACT_INV_CONST_COLOR 0x0d +#define BLENDFACT_CONST_ALPHA 0x0e +#define BLENDFACT_INV_CONST_ALPHA 0x0f +#define BLENDFACT_MASK 0x0f + +enum { + DRI_CONF_BO_REUSE_DISABLED, + DRI_CONF_BO_REUSE_ALL +}; + +extern int intel_translate_shadow_compare_func(GLenum func); +extern int intel_translate_compare_func(GLenum func); +extern int intel_translate_stencil_op(GLenum op); +extern int intel_translate_blend_factor(GLenum factor); +extern int intel_translate_logic_op(GLenum opcode); + +void intel_update_renderbuffers(__DRIcontext *context, + __DRIdrawable *drawable); + +/*====================================================================== + * Inline conversion functions. + * These are better-typed than the macros used previously: + */ +static INLINE struct intel_context * +intel_context(GLcontext * ctx) +{ + return (struct intel_context *) ctx; +} + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_decode.c b/src/mesa/drivers/dri/intel/intel_decode.c new file mode 100644 index 0000000000..9c105013c0 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_decode.c @@ -0,0 +1,1053 @@ +/* -*- c-basic-offset: 4 -*- */ +/* + * Copyright © 2007 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +/** @file intel_decode.c + * This file contains code to print out batchbuffer contents in a + * human-readable format. + * + * The current version only supports i915 packets, and only pretty-prints a + * subset of them. The intention is for it to make just a best attempt to + * decode, but never crash in the process. + */ + +#include <stdio.h> +#include <stdarg.h> +#include <inttypes.h> + +#include "intel_decode.h" +#include "intel_chipset.h" + +#define BUFFER_FAIL(_count, _len, _name) do { \ + fprintf(out, "Buffer size too small in %s (%d < %d)\n", \ + (_name), (_count), (_len)); \ + (*failures)++; \ + return count; \ +} while (0) + +static FILE *out; +static uint32_t saved_s2 = 0, saved_s4 = 0; +static char saved_s2_set = 0, saved_s4_set = 0; + +static float +int_as_float(uint32_t intval) +{ + union intfloat { + uint32_t i; + float f; + } uval; + + uval.i = intval; + return uval.f; +} + +static void +instr_out(uint32_t *data, uint32_t hw_offset, unsigned int index, + char *fmt, ...) +{ + va_list va; + + fprintf(out, "0x%08x: 0x%08x:%s ", hw_offset + index * 4, data[index], + index == 0 ? "" : " "); + va_start(va, fmt); + vfprintf(out, fmt, va); + va_end(va); +} + + +static int +decode_mi(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + unsigned int opcode; + + struct { + uint32_t opcode; + int min_len; + int max_len; + char *name; + } opcodes_mi[] = { + { 0x08, 1, 1, "MI_ARB_ON_OFF" }, + { 0x0a, 1, 1, "MI_BATCH_BUFFER_END" }, + { 0x31, 2, 2, "MI_BATCH_BUFFER_START" }, + { 0x14, 3, 3, "MI_DISPLAY_BUFFER_INFO" }, + { 0x04, 1, 1, "MI_FLUSH" }, + { 0x22, 3, 3, "MI_LOAD_REGISTER_IMM" }, + { 0x13, 2, 2, "MI_LOAD_SCAN_LINES_EXCL" }, + { 0x12, 2, 2, "MI_LOAD_SCAN_LINES_INCL" }, + { 0x00, 1, 1, "MI_NOOP" }, + { 0x11, 2, 2, "MI_OVERLAY_FLIP" }, + { 0x07, 1, 1, "MI_REPORT_HEAD" }, + { 0x18, 2, 2, "MI_SET_CONTEXT" }, + { 0x20, 3, 4, "MI_STORE_DATA_IMM" }, + { 0x21, 3, 4, "MI_STORE_DATA_INDEX" }, + { 0x24, 3, 3, "MI_STORE_REGISTER_MEM" }, + { 0x02, 1, 1, "MI_USER_INTERRUPT" }, + { 0x03, 1, 1, "MI_WAIT_FOR_EVENT" }, + }; + + + for (opcode = 0; opcode < sizeof(opcodes_mi) / sizeof(opcodes_mi[0]); + opcode++) { + if ((data[0] & 0x1f800000) >> 23 == opcodes_mi[opcode].opcode) { + unsigned int len = 1, i; + + instr_out(data, hw_offset, 0, "%s\n", opcodes_mi[opcode].name); + if (opcodes_mi[opcode].max_len > 1) { + len = (data[0] & 0x000000ff) + 2; + if (len < opcodes_mi[opcode].min_len || + len > opcodes_mi[opcode].max_len) + { + fprintf(out, "Bad length in %s\n", + opcodes_mi[opcode].name); + } + } + + for (i = 1; i < len; i++) { + if (i >= count) + BUFFER_FAIL(count, len, opcodes_mi[opcode].name); + instr_out(data, hw_offset, i, "dword %d\n", i); + } + + return len; + } + } + + instr_out(data, hw_offset, 0, "MI UNKNOWN\n"); + (*failures)++; + return 1; +} + +static int +decode_2d(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + unsigned int opcode, len; + char *format = NULL; + + struct { + uint32_t opcode; + int min_len; + int max_len; + char *name; + } opcodes_2d[] = { + { 0x40, 5, 5, "COLOR_BLT" }, + { 0x43, 6, 6, "SRC_COPY_BLT" }, + { 0x01, 8, 8, "XY_SETUP_BLT" }, + { 0x11, 9, 9, "XY_SETUP_MONO_PATTERN_SL_BLT" }, + { 0x03, 3, 3, "XY_SETUP_CLIP_BLT" }, + { 0x24, 2, 2, "XY_PIXEL_BLT" }, + { 0x25, 3, 3, "XY_SCANLINES_BLT" }, + { 0x26, 4, 4, "Y_TEXT_BLT" }, + { 0x31, 5, 134, "XY_TEXT_IMMEDIATE_BLT" }, + { 0x50, 6, 6, "XY_COLOR_BLT" }, + { 0x51, 6, 6, "XY_PAT_BLT" }, + { 0x76, 8, 8, "XY_PAT_CHROMA_BLT" }, + { 0x72, 7, 135, "XY_PAT_BLT_IMMEDIATE" }, + { 0x77, 9, 137, "XY_PAT_CHROMA_BLT_IMMEDIATE" }, + { 0x52, 9, 9, "XY_MONO_PAT_BLT" }, + { 0x59, 7, 7, "XY_MONO_PAT_FIXED_BLT" }, + { 0x53, 8, 8, "XY_SRC_COPY_BLT" }, + { 0x54, 8, 8, "XY_MONO_SRC_COPY_BLT" }, + { 0x71, 9, 137, "XY_MONO_SRC_COPY_IMMEDIATE_BLT" }, + { 0x55, 9, 9, "XY_FULL_BLT" }, + { 0x55, 9, 137, "XY_FULL_IMMEDIATE_PATTERN_BLT" }, + { 0x56, 9, 9, "XY_FULL_MONO_SRC_BLT" }, + { 0x75, 10, 138, "XY_FULL_MONO_SRC_IMMEDIATE_PATTERN_BLT" }, + { 0x57, 12, 12, "XY_FULL_MONO_PATTERN_BLT" }, + { 0x58, 12, 12, "XY_FULL_MONO_PATTERN_MONO_SRC_BLT" }, + }; + + switch ((data[0] & 0x1fc00000) >> 22) { + case 0x50: + instr_out(data, hw_offset, 0, + "XY_COLOR_BLT (rgb %sabled, alpha %sabled, dst tile %d)\n", + (data[0] & (1 << 20)) ? "en" : "dis", + (data[0] & (1 << 21)) ? "en" : "dis", + (data[0] >> 11) & 1); + + len = (data[0] & 0x000000ff) + 2; + if (len != 6) + fprintf(out, "Bad count in XY_COLOR_BLT\n"); + if (count < 6) + BUFFER_FAIL(count, len, "XY_COLOR_BLT"); + + switch ((data[1] >> 24) & 0x3) { + case 0: + format="8"; + break; + case 1: + format="565"; + break; + case 2: + format="1555"; + break; + case 3: + format="8888"; + break; + } + + instr_out(data, hw_offset, 1, "format %s, pitch %d, " + "clipping %sabled\n", format, + (short)(data[1] & 0xffff), + data[1] & (1 << 30) ? "en" : "dis"); + instr_out(data, hw_offset, 2, "(%d,%d)\n", + data[2] & 0xffff, data[2] >> 16); + instr_out(data, hw_offset, 3, "(%d,%d)\n", + data[3] & 0xffff, data[3] >> 16); + instr_out(data, hw_offset, 4, "offset 0x%08x\n", data[4]); + instr_out(data, hw_offset, 5, "color\n"); + return len; + case 0x53: + instr_out(data, hw_offset, 0, + "XY_SRC_COPY_BLT (rgb %sabled, alpha %sabled, " + "src tile %d, dst tile %d)\n", + (data[0] & (1 << 20)) ? "en" : "dis", + (data[0] & (1 << 21)) ? "en" : "dis", + (data[0] >> 15) & 1, + (data[0] >> 11) & 1); + + len = (data[0] & 0x000000ff) + 2; + if (len != 8) + fprintf(out, "Bad count in XY_SRC_COPY_BLT\n"); + if (count < 8) + BUFFER_FAIL(count, len, "XY_SRC_COPY_BLT"); + + switch ((data[1] >> 24) & 0x3) { + case 0: + format="8"; + break; + case 1: + format="565"; + break; + case 2: + format="1555"; + break; + case 3: + format="8888"; + break; + } + + instr_out(data, hw_offset, 1, "format %s, dst pitch %d, " + "clipping %sabled\n", format, + (short)(data[1] & 0xffff), + data[1] & (1 << 30) ? "en" : "dis"); + instr_out(data, hw_offset, 2, "dst (%d,%d)\n", + data[2] & 0xffff, data[2] >> 16); + instr_out(data, hw_offset, 3, "dst (%d,%d)\n", + data[3] & 0xffff, data[3] >> 16); + instr_out(data, hw_offset, 4, "dst offset 0x%08x\n", data[4]); + instr_out(data, hw_offset, 5, "src (%d,%d)\n", + data[5] & 0xffff, data[5] >> 16); + instr_out(data, hw_offset, 6, "src pitch %d\n", + (short)(data[6] & 0xffff)); + instr_out(data, hw_offset, 7, "src offset 0x%08x\n", data[7]); + return len; + } + + for (opcode = 0; opcode < sizeof(opcodes_2d) / sizeof(opcodes_2d[0]); + opcode++) { + if ((data[0] & 0x1fc00000) >> 22 == opcodes_2d[opcode].opcode) { + unsigned int i; + + len = 1; + instr_out(data, hw_offset, 0, "%s\n", opcodes_2d[opcode].name); + if (opcodes_2d[opcode].max_len > 1) { + len = (data[0] & 0x000000ff) + 2; + if (len < opcodes_2d[opcode].min_len || + len > opcodes_2d[opcode].max_len) + { + fprintf(out, "Bad count in %s\n", opcodes_2d[opcode].name); + } + } + + for (i = 1; i < len; i++) { + if (i >= count) + BUFFER_FAIL(count, len, opcodes_2d[opcode].name); + instr_out(data, hw_offset, i, "dword %d\n", i); + } + + return len; + } + } + + instr_out(data, hw_offset, 0, "2D UNKNOWN\n"); + (*failures)++; + return 1; +} + +static int +decode_3d_1c(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + switch ((data[0] & 0x00f80000) >> 19) { + case 0x11: + instr_out(data, hw_offset, 0, "3DSTATE_DEPTH_SUBRECTANGLE_DISALBE\n"); + return 1; + case 0x10: + instr_out(data, hw_offset, 0, "3DSTATE_SCISSOR_ENABLE\n"); + return 1; + } + + instr_out(data, hw_offset, 0, "3D UNKNOWN\n"); + (*failures)++; + return 1; +} + +static int +decode_3d_1d(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + unsigned int len, i, c, opcode, word, map, sampler, instr; + + struct { + uint32_t opcode; + int min_len; + int max_len; + char *name; + } opcodes_3d_1d[] = { + { 0x8e, 3, 3, "3DSTATE_BUFFER_INFO" }, + { 0x86, 4, 4, "3DSTATE_CHROMA_KEY" }, + { 0x9c, 1, 1, "3DSTATE_CLEAR_PARAMETERS" }, + { 0x88, 2, 2, "3DSTATE_CONSTANT_BLEND_COLOR" }, + { 0x99, 2, 2, "3DSTATE_DEFAULT_DIFFUSE" }, + { 0x9a, 2, 2, "3DSTATE_DEFAULT_SPECULAR" }, + { 0x98, 2, 2, "3DSTATE_DEFAULT_Z" }, + { 0x97, 2, 2, "3DSTATE_DEPTH_OFFSET_SCALE" }, + { 0x85, 2, 2, "3DSTATE_DEST_BUFFER_VARIABLES" }, + { 0x80, 5, 5, "3DSTATE_DRAWING_RECTANGLE" }, + { 0x8e, 3, 3, "3DSTATE_BUFFER_INFO" }, + { 0x9d, 65, 65, "3DSTATE_FILTER_COEFFICIENTS_4X4" }, + { 0x9e, 4, 4, "3DSTATE_MONO_FILTER" }, + { 0x89, 4, 4, "3DSTATE_FOG_MODE" }, + { 0x8f, 2, 16, "3DSTATE_MAP_PALLETE_LOAD_32" }, + { 0x81, 3, 3, "3DSTATE_SCISSOR_RECTANGLE" }, + { 0x83, 2, 2, "3DSTATE_SPAN_STIPPLE" }, + }; + + switch ((data[0] & 0x00ff0000) >> 16) { + case 0x07: + /* This instruction is unusual. A 0 length means just 1 DWORD instead of + * 2. The 0 length is specified in one place to be unsupported, but + * stated to be required in another, and 0 length LOAD_INDIRECTs appear + * to cause no harm at least. + */ + instr_out(data, hw_offset, 0, "3DSTATE_LOAD_INDIRECT\n"); + len = (data[0] & 0x000000ff) + 1; + i = 1; + if (data[0] & (0x01 << 8)) { + if (i + 2 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "SIS.0\n"); + instr_out(data, hw_offset, i++, "SIS.1\n"); + } + if (data[0] & (0x02 << 8)) { + if (i + 1 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "DIS.0\n"); + } + if (data[0] & (0x04 << 8)) { + if (i + 2 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "SSB.0\n"); + instr_out(data, hw_offset, i++, "SSB.1\n"); + } + if (data[0] & (0x08 << 8)) { + if (i + 2 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "MSB.0\n"); + instr_out(data, hw_offset, i++, "MSB.1\n"); + } + if (data[0] & (0x10 << 8)) { + if (i + 2 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "PSP.0\n"); + instr_out(data, hw_offset, i++, "PSP.1\n"); + } + if (data[0] & (0x20 << 8)) { + if (i + 2 >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_INDIRECT"); + instr_out(data, hw_offset, i++, "PSC.0\n"); + instr_out(data, hw_offset, i++, "PSC.1\n"); + } + if (len != i) { + fprintf(out, "Bad count in 3DSTATE_LOAD_INDIRECT\n"); + (*failures)++; + return len; + } + return len; + case 0x04: + instr_out(data, hw_offset, 0, "3DSTATE_LOAD_STATE_IMMEDIATE_1\n"); + len = (data[0] & 0x0000000f) + 2; + i = 1; + for (word = 0; word <= 7; word++) { + if (data[0] & (1 << (4 + word))) { + if (i >= count) + BUFFER_FAIL(count, len, "3DSTATE_LOAD_STATE_IMMEDIATE_1"); + + /* save vertex state for decode */ + if (word == 2) { + saved_s2_set = 1; + saved_s2 = data[i]; + } + if (word == 4) { + saved_s4_set = 1; + saved_s4 = data[i]; + } + + instr_out(data, hw_offset, i++, "S%d\n", word); + } + } + if (len != i) { + fprintf(out, "Bad count in 3DSTATE_LOAD_INDIRECT\n"); + (*failures)++; + } + return len; + case 0x00: + instr_out(data, hw_offset, 0, "3DSTATE_MAP_STATE\n"); + len = (data[0] & 0x0000003f) + 2; + + i = 1; + for (map = 0; map <= 15; map++) { + if (data[1] & (1 << map)) { + if (i + 3 >= count) + BUFFER_FAIL(count, len, "3DSTATE_MAP_STATE"); + instr_out(data, hw_offset, i++, "map %d MS2\n", map); + instr_out(data, hw_offset, i++, "map %d MS3\n", map); + instr_out(data, hw_offset, i++, "map %d MS4\n", map); + } + } + if (len != i) { + fprintf(out, "Bad count in 3DSTATE_MAP_STATE\n"); + (*failures)++; + return len; + } + return len; + case 0x06: + instr_out(data, hw_offset, 0, "3DSTATE_PIXEL_SHADER_CONSTANTS\n"); + len = (data[0] & 0x000000ff) + 2; + + i = 1; + for (c = 0; c <= 31; c++) { + if (data[1] & (1 << c)) { + if (i + 4 >= count) + BUFFER_FAIL(count, len, "3DSTATE_PIXEL_SHADER_CONSTANTS"); + instr_out(data, hw_offset, i, "C%d.X = %f\n", + c, int_as_float(data[i])); + i++; + instr_out(data, hw_offset, i, "C%d.Y = %f\n", + c, int_as_float(data[i])); + i++; + instr_out(data, hw_offset, i, "C%d.Z = %f\n", + c, int_as_float(data[i])); + i++; + instr_out(data, hw_offset, i, "C%d.W = %f\n", + c, int_as_float(data[i])); + i++; + } + } + if (len != i) { + fprintf(out, "Bad count in 3DSTATE_MAP_STATE\n"); + (*failures)++; + } + return len; + case 0x05: + instr_out(data, hw_offset, 0, "3DSTATE_PIXEL_SHADER_PROGRAM\n"); + len = (data[0] & 0x000000ff) + 2; + if ((len - 1) % 3 != 0 || len > 370) { + fprintf(out, "Bad count in 3DSTATE_PIXEL_SHADER_PROGRAM\n"); + (*failures)++; + } + i = 1; + for (instr = 0; instr < (len - 1) / 3; instr++) { + if (i + 3 >= count) + BUFFER_FAIL(count, len, "3DSTATE_MAP_STATE"); + instr_out(data, hw_offset, i++, "PS%03x\n", instr); + instr_out(data, hw_offset, i++, "PS%03x\n", instr); + instr_out(data, hw_offset, i++, "PS%03x\n", instr); + } + return len; + case 0x01: + instr_out(data, hw_offset, 0, "3DSTATE_SAMPLER_STATE\n"); + len = (data[0] & 0x0000003f) + 2; + i = 1; + for (sampler = 0; sampler <= 15; sampler++) { + if (data[1] & (1 << sampler)) { + if (i + 3 >= count) + BUFFER_FAIL(count, len, "3DSTATE_SAMPLER_STATE"); + instr_out(data, hw_offset, i++, "sampler %d SS2\n", + sampler); + instr_out(data, hw_offset, i++, "sampler %d SS3\n", + sampler); + instr_out(data, hw_offset, i++, "sampler %d SS4\n", + sampler); + } + } + if (len != i) { + fprintf(out, "Bad count in 3DSTATE_SAMPLER_STATE\n"); + (*failures)++; + } + return len; + } + + for (opcode = 0; opcode < sizeof(opcodes_3d_1d) / sizeof(opcodes_3d_1d[0]); + opcode++) + { + if (((data[0] & 0x00ff0000) >> 16) == opcodes_3d_1d[opcode].opcode) { + len = 1; + + instr_out(data, hw_offset, 0, "%s\n", opcodes_3d_1d[opcode].name); + if (opcodes_3d_1d[opcode].max_len > 1) { + len = (data[0] & 0x0000ffff) + 2; + if (len < opcodes_3d_1d[opcode].min_len || + len > opcodes_3d_1d[opcode].max_len) + { + fprintf(out, "Bad count in %s\n", + opcodes_3d_1d[opcode].name); + (*failures)++; + } + } + + for (i = 1; i < len; i++) { + if (i >= count) + BUFFER_FAIL(count, len, opcodes_3d_1d[opcode].name); + instr_out(data, hw_offset, i, "dword %d\n", i); + } + + return len; + } + } + + instr_out(data, hw_offset, 0, "3D UNKNOWN\n"); + (*failures)++; + return 1; +} + +static int +decode_3d_primitive(uint32_t *data, int count, uint32_t hw_offset, + int *failures) +{ + char immediate = (data[0] & (1 << 23)) == 0; + unsigned int len, i; + char *primtype; + + switch ((data[0] >> 18) & 0xf) { + case 0x0: primtype = "TRILIST"; break; + case 0x1: primtype = "TRISTRIP"; break; + case 0x2: primtype = "TRISTRIP_REVERSE"; break; + case 0x3: primtype = "TRIFAN"; break; + case 0x4: primtype = "POLYGON"; break; + case 0x5: primtype = "LINELIST"; break; + case 0x6: primtype = "LINESTRIP"; break; + case 0x7: primtype = "RECTLIST"; break; + case 0x8: primtype = "POINTLIST"; break; + case 0x9: primtype = "DIB"; break; + case 0xa: primtype = "CLEAR_RECT"; break; + default: primtype = "unknown"; break; + } + + /* XXX: 3DPRIM_DIB not supported */ + if (immediate) { + len = (data[0] & 0x0003ffff) + 2; + instr_out(data, hw_offset, 0, "3DPRIMITIVE inline %s\n", primtype); + if (count < len) + BUFFER_FAIL(count, len, "3DPRIMITIVE inline"); + if (!saved_s2_set || !saved_s4_set) { + fprintf(out, "unknown vertex format\n"); + for (i = 1; i < len; i++) { + instr_out(data, hw_offset, i, + " vertex data (%f float)\n", + int_as_float(data[i])); + } + } else { + unsigned int vertex = 0; + for (i = 1; i < len;) { + unsigned int tc; + +#define VERTEX_OUT(fmt, ...) do { \ + if (i < len) \ + instr_out(data, hw_offset, i, " V%d."fmt"\n", vertex, __VA_ARGS__); \ + else \ + fprintf(out, " missing data in V%d\n", vertex); \ + i++; \ +} while (0) + + VERTEX_OUT("X = %f", int_as_float(data[i])); + VERTEX_OUT("Y = %f", int_as_float(data[i])); + switch (saved_s4 >> 6 & 0x7) { + case 0x1: + VERTEX_OUT("Z = %f", int_as_float(data[i])); + break; + case 0x2: + VERTEX_OUT("Z = %f", int_as_float(data[i])); + VERTEX_OUT("W = %f", int_as_float(data[i])); + break; + case 0x3: + break; + case 0x4: + VERTEX_OUT("W = %f", int_as_float(data[i])); + break; + default: + fprintf(out, "bad S4 position mask\n"); + } + + if (saved_s4 & (1 << 10)) { + VERTEX_OUT("color = (A=0x%02x, R=0x%02x, G=0x%02x, " + "B=0x%02x)", + data[i] >> 24, + (data[i] >> 16) & 0xff, + (data[i] >> 8) & 0xff, + data[i] & 0xff); + } + if (saved_s4 & (1 << 11)) { + VERTEX_OUT("spec = (A=0x%02x, R=0x%02x, G=0x%02x, " + "B=0x%02x)", + data[i] >> 24, + (data[i] >> 16) & 0xff, + (data[i] >> 8) & 0xff, + data[i] & 0xff); + } + if (saved_s4 & (1 << 12)) + VERTEX_OUT("width = 0x%08x)", data[i]); + + for (tc = 0; tc <= 7; tc++) { + switch ((saved_s2 >> (tc * 4)) & 0xf) { + case 0x0: + VERTEX_OUT("T%d.X = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.Y = %f", tc, int_as_float(data[i])); + break; + case 0x1: + VERTEX_OUT("T%d.X = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.Y = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.Z = %f", tc, int_as_float(data[i])); + break; + case 0x2: + VERTEX_OUT("T%d.X = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.Y = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.Z = %f", tc, int_as_float(data[i])); + VERTEX_OUT("T%d.W = %f", tc, int_as_float(data[i])); + break; + case 0x3: + VERTEX_OUT("T%d.X = %f", tc, int_as_float(data[i])); + break; + case 0x4: + VERTEX_OUT("T%d.XY = 0x%08x half-float", tc, data[i]); + break; + case 0x5: + VERTEX_OUT("T%d.XY = 0x%08x half-float", tc, data[i]); + VERTEX_OUT("T%d.ZW = 0x%08x half-float", tc, data[i]); + break; + case 0xf: + break; + default: + fprintf(out, "bad S2.T%d format\n", tc); + } + } + vertex++; + } + } + } else { + /* indirect vertices */ + len = data[0] & 0x0000ffff; /* index count */ + if (data[0] & (1 << 17)) { + /* random vertex access */ + if (count < (len + 1) / 2 + 1) { + BUFFER_FAIL(count, (len + 1) / 2 + 1, + "3DPRIMITIVE random indirect"); + } + instr_out(data, hw_offset, 0, + "3DPRIMITIVE random indirect %s (%d)\n", primtype, len); + if (len == 0) { + /* vertex indices continue until 0xffff is found */ + for (i = 1; i < count; i++) { + if ((data[i] & 0xffff) == 0xffff) { + instr_out(data, hw_offset, i, + " indices: (terminator)\n"); + return i; + } else if ((data[i] >> 16) == 0xffff) { + instr_out(data, hw_offset, i, + " indices: 0x%04x, " + "(terminator)\n", + data[i] & 0xffff); + return i; + } else { + instr_out(data, hw_offset, i, + " indices: 0x%04x, 0x%04x\n", + data[i] & 0xffff, data[i] >> 16); + } + } + fprintf(out, + "3DPRIMITIVE: no terminator found in index buffer\n"); + (*failures)++; + return count; + } else { + /* fixed size vertex index buffer */ + for (i = 0; i < len; i += 2) { + if (i * 2 == len - 1) { + instr_out(data, hw_offset, i, + " indices: 0x%04x\n", + data[i] & 0xffff); + } else { + instr_out(data, hw_offset, i, + " indices: 0x%04x, 0x%04x\n", + data[i] & 0xffff, data[i] >> 16); + } + } + } + return (len + 1) / 2 + 1; + } else { + /* sequential vertex access */ + if (count < 2) + BUFFER_FAIL(count, 2, "3DPRIMITIVE seq indirect"); + instr_out(data, hw_offset, 0, + "3DPRIMITIVE sequential indirect %s, %d starting from " + "%d\n", primtype, len, data[1] & 0xffff); + instr_out(data, hw_offset, 1, " start\n"); + return 2; + } + } + + return len; +} + +static int +decode_3d(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + unsigned int opcode; + + struct { + uint32_t opcode; + int min_len; + int max_len; + char *name; + } opcodes_3d[] = { + { 0x06, 1, 1, "3DSTATE_ANTI_ALIASING" }, + { 0x08, 1, 1, "3DSTATE_BACKFACE_STENCIL_OPS" }, + { 0x09, 1, 1, "3DSTATE_BACKFACE_STENCIL_MASKS" }, + { 0x16, 1, 1, "3DSTATE_COORD_SET_BINDINGS" }, + { 0x15, 1, 1, "3DSTATE_FOG_COLOR" }, + { 0x0b, 1, 1, "3DSTATE_INDEPENDENT_ALPHA_BLEND" }, + { 0x0d, 1, 1, "3DSTATE_MODES_4" }, + { 0x0c, 1, 1, "3DSTATE_MODES_5" }, + { 0x07, 1, 1, "3DSTATE_RASTERIZATION_RULES" }, + }; + + switch ((data[0] & 0x1f000000) >> 24) { + case 0x1f: + return decode_3d_primitive(data, count, hw_offset, failures); + case 0x1d: + return decode_3d_1d(data, count, hw_offset, failures); + case 0x1c: + return decode_3d_1c(data, count, hw_offset, failures); + } + + for (opcode = 0; opcode < sizeof(opcodes_3d) / sizeof(opcodes_3d[0]); + opcode++) { + if ((data[0] & 0x1f000000) >> 24 == opcodes_3d[opcode].opcode) { + unsigned int len = 1, i; + + instr_out(data, hw_offset, 0, "%s\n", opcodes_3d[opcode].name); + if (opcodes_3d[opcode].max_len > 1) { + len = (data[0] & 0xff) + 2; + if (len < opcodes_3d[opcode].min_len || + len > opcodes_3d[opcode].max_len) + { + fprintf(out, "Bad count in %s\n", opcodes_3d[opcode].name); + } + } + + for (i = 1; i < len; i++) { + if (i >= count) + BUFFER_FAIL(count, len, opcodes_3d[opcode].name); + instr_out(data, hw_offset, i, "dword %d\n", i); + } + return len; + } + } + + instr_out(data, hw_offset, 0, "3D UNKNOWN\n"); + (*failures)++; + return 1; +} + +static const char * +get_965_surfacetype(unsigned int surfacetype) +{ + switch (surfacetype) { + case 0: return "1D"; + case 1: return "2D"; + case 2: return "3D"; + case 3: return "CUBE"; + case 4: return "BUFFER"; + case 7: return "NULL"; + default: return "unknown"; + } +} + +static const char * +get_965_depthformat(unsigned int depthformat) +{ + switch (depthformat) { + case 0: return "s8_z24float"; + case 1: return "z32float"; + case 2: return "z24s8"; + case 5: return "z16"; + default: return "unknown"; + } +} + +static int +decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures) +{ + unsigned int opcode, len; + + struct { + uint32_t opcode; + int min_len; + int max_len; + char *name; + } opcodes_3d[] = { + { 0x6000, 3, 3, "URB_FENCE" }, + { 0x6001, 2, 2, "CS_URB_STATE" }, + { 0x6002, 2, 2, "CONSTANT_BUFFER" }, + { 0x6101, 6, 6, "STATE_BASE_ADDRESS" }, + { 0x6102, 2, 2 , "STATE_SIP" }, + { 0x6104, 1, 1, "3DSTATE_PIPELINE_SELECT" }, + { 0x680b, 1, 1, "3DSTATE_VF_STATISTICS" }, + { 0x6904, 1, 1, "3DSTATE_PIPELINE_SELECT" }, + { 0x7800, 7, 7, "3DSTATE_PIPELINED_POINTERS" }, + { 0x7801, 6, 6, "3DSTATE_BINDING_TABLE_POINTERS" }, + { 0x780b, 1, 1, "3DSTATE_VF_STATISTICS" }, + { 0x7808, 5, 257, "3DSTATE_VERTEX_BUFFERS" }, + { 0x7809, 3, 256, "3DSTATE_VERTEX_ELEMENTS" }, + /* 0x7808: 3DSTATE_VERTEX_BUFFERS */ + /* 0x7809: 3DSTATE_VERTEX_ELEMENTS */ + { 0x7900, 4, 4, "3DSTATE_DRAWING_RECTANGLE" }, + { 0x7901, 5, 5, "3DSTATE_CONSTANT_COLOR" }, + { 0x7905, 5, 7, "3DSTATE_DEPTH_BUFFER" }, + { 0x7906, 2, 2, "3DSTATE_POLY_STIPPLE_OFFSET" }, + { 0x7907, 33, 33, "3DSTATE_POLY_STIPPLE_PATTERN" }, + { 0x7908, 3, 3, "3DSTATE_LINE_STIPPLE" }, + { 0x7909, 2, 2, "3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP" }, + { 0x790a, 3, 3, "3DSTATE_AA_LINE_PARAMETERS" }, + { 0x7b00, 6, 6, "3DPRIMITIVE" }, + }; + + len = (data[0] & 0x0000ffff) + 2; + + switch ((data[0] & 0xffff0000) >> 16) { + case 0x6101: + if (len != 6) + fprintf(out, "Bad count in STATE_BASE_ADDRESS\n"); + if (count < 6) + BUFFER_FAIL(count, len, "STATE_BASE_ADDRESS"); + + instr_out(data, hw_offset, 0, + "STATE_BASE_ADDRESS\n"); + + if (data[1] & 1) { + instr_out(data, hw_offset, 1, "General state at 0x%08x\n", + data[1] & ~1); + } else + instr_out(data, hw_offset, 1, "General state not updated\n"); + + if (data[2] & 1) { + instr_out(data, hw_offset, 2, "Surface state at 0x%08x\n", + data[2] & ~1); + } else + instr_out(data, hw_offset, 2, "Surface state not updated\n"); + + if (data[3] & 1) { + instr_out(data, hw_offset, 3, "Indirect state at 0x%08x\n", + data[3] & ~1); + } else + instr_out(data, hw_offset, 3, "Indirect state not updated\n"); + + if (data[4] & 1) { + instr_out(data, hw_offset, 4, "General state upper bound 0x%08x\n", + data[4] & ~1); + } else + instr_out(data, hw_offset, 4, "General state not updated\n"); + + if (data[5] & 1) { + instr_out(data, hw_offset, 5, "Indirect state upper bound 0x%08x\n", + data[5] & ~1); + } else + instr_out(data, hw_offset, 5, "Indirect state not updated\n"); + + return len; + case 0x7800: + if (len != 7) + fprintf(out, "Bad count in 3DSTATE_PIPELINED_POINTERS\n"); + if (count < 7) + BUFFER_FAIL(count, len, "3DSTATE_PIPELINED_POINTERS"); + + instr_out(data, hw_offset, 0, + "3DSTATE_PIPELINED_POINTERS\n"); + instr_out(data, hw_offset, 1, "VS state\n"); + instr_out(data, hw_offset, 2, "GS state\n"); + instr_out(data, hw_offset, 3, "Clip state\n"); + instr_out(data, hw_offset, 4, "SF state\n"); + instr_out(data, hw_offset, 5, "WM state\n"); + instr_out(data, hw_offset, 6, "CC state\n"); + return len; + case 0x7801: + if (len != 6) + fprintf(out, "Bad count in 3DSTATE_BINDING_TABLE_POINTERS\n"); + if (count < 6) + BUFFER_FAIL(count, len, "3DSTATE_BINDING_TABLE_POINTERS"); + + instr_out(data, hw_offset, 0, + "3DSTATE_BINDING_TABLE_POINTERS\n"); + instr_out(data, hw_offset, 1, "VS binding table\n"); + instr_out(data, hw_offset, 2, "GS binding table\n"); + instr_out(data, hw_offset, 3, "Clip binding table\n"); + instr_out(data, hw_offset, 4, "SF binding table\n"); + instr_out(data, hw_offset, 5, "WM binding table\n"); + + return len; + + case 0x7900: + if (len != 4) + fprintf(out, "Bad count in 3DSTATE_DRAWING_RECTANGLE\n"); + if (count < 4) + BUFFER_FAIL(count, len, "3DSTATE_DRAWING_RECTANGLE"); + + instr_out(data, hw_offset, 0, + "3DSTATE_DRAWING_RECTANGLE\n"); + instr_out(data, hw_offset, 1, "top left: %d,%d\n", + data[1] & 0xffff, + (data[1] >> 16) & 0xffff); + instr_out(data, hw_offset, 2, "bottom right: %d,%d\n", + data[2] & 0xffff, + (data[2] >> 16) & 0xffff); + instr_out(data, hw_offset, 3, "origin: %d,%d\n", + (int)data[3] & 0xffff, + ((int)data[3] >> 16) & 0xffff); + + return len; + + case 0x7905: + if (len != 5) + fprintf(out, "Bad count in 3DSTATE_DEPTH_BUFFER\n"); + if (count < 5) + BUFFER_FAIL(count, len, "3DSTATE_DEPTH_BUFFER"); + + instr_out(data, hw_offset, 0, + "3DSTATE_DEPTH_BUFFER\n"); + instr_out(data, hw_offset, 1, "%s, %s, pitch = %d bytes, %stiled\n", + get_965_surfacetype(data[1] >> 29), + get_965_depthformat((data[1] >> 18) & 0x7), + (data[1] & 0x0001ffff) + 1, + data[1] & (1 << 27) ? "" : "not "); + instr_out(data, hw_offset, 2, "depth offset\n"); + instr_out(data, hw_offset, 3, "%dx%d\n", + ((data[3] & 0x0007ffc0) >> 6) + 1, + ((data[3] & 0xfff80000) >> 19) + 1); + instr_out(data, hw_offset, 4, "volume depth\n"); + + return len; + } + + for (opcode = 0; opcode < sizeof(opcodes_3d) / sizeof(opcodes_3d[0]); + opcode++) { + if ((data[0] & 0xffff0000) >> 16 == opcodes_3d[opcode].opcode) { + unsigned int i; + len = 1; + + instr_out(data, hw_offset, 0, "%s\n", opcodes_3d[opcode].name); + if (opcodes_3d[opcode].max_len > 1) { + len = (data[0] & 0xff) + 2; + if (len < opcodes_3d[opcode].min_len || + len > opcodes_3d[opcode].max_len) + { + fprintf(out, "Bad count in %s\n", opcodes_3d[opcode].name); + } + } + + for (i = 1; i < len; i++) { + if (i >= count) + BUFFER_FAIL(count, len, opcodes_3d[opcode].name); + instr_out(data, hw_offset, i, "dword %d\n", i); + } + return len; + } + } + + instr_out(data, hw_offset, 0, "3D UNKNOWN\n"); + (*failures)++; + return 1; +} + +/** + * Decodes an i830-i915 batch buffer, writing the output to stdout. + * + * \param data batch buffer contents + * \param count number of DWORDs to decode in the batch buffer + * \param hw_offset hardware address for the buffer + */ +int +intel_decode(uint32_t *data, int count, uint32_t hw_offset, uint32_t devid) +{ + int index = 0; + int failures = 0; + + out = stderr; + + while (index < count) { + switch ((data[index] & 0xe0000000) >> 29) { + case 0x0: + index += decode_mi(data + index, count - index, + hw_offset + index * 4, &failures); + break; + case 0x2: + index += decode_2d(data + index, count - index, + hw_offset + index * 4, &failures); + break; + case 0x3: + if (IS_965(devid)) { + index += decode_3d_965(data + index, count - index, + hw_offset + index * 4, &failures); + } else { + index += decode_3d(data + index, count - index, + hw_offset + index * 4, &failures); + } + break; + default: + instr_out(data, hw_offset, index, "UNKNOWN\n"); + failures++; + index++; + break; + } + fflush(out); + } + + return failures; +} + +void intel_decode_context_reset(void) +{ + saved_s2_set = 0; + saved_s4_set = 1; +} + diff --git a/src/mesa/drivers/dri/intel/intel_decode.h b/src/mesa/drivers/dri/intel/intel_decode.h new file mode 100644 index 0000000000..c50644a46b --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_decode.h @@ -0,0 +1,29 @@ +/* + * Copyright © 2007 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +int intel_decode(uint32_t *data, int count, uint32_t hw_offset, uint32_t devid); +void intel_decode_context_reset(void); diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.c b/src/mesa/drivers/dri/intel/intel_depthstencil.c new file mode 100644 index 0000000000..70ba68e9e3 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_depthstencil.c @@ -0,0 +1,252 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "imports.h" +#include "context.h" +#include "depthstencil.h" +#include "fbobject.h" +#include "framebuffer.h" +#include "hash.h" +#include "mtypes.h" +#include "renderbuffer.h" + +#include "intel_context.h" +#include "intel_fbo.h" +#include "intel_depthstencil.h" +#include "intel_regions.h" +#include "intel_span.h" + +/** + * The GL_EXT_framebuffer_object allows the user to create their own + * framebuffer objects consisting of color renderbuffers (0 or more), + * depth renderbuffers (0 or 1) and stencil renderbuffers (0 or 1). + * + * The spec considers depth and stencil renderbuffers to be totally independent + * buffers. In reality, most graphics hardware today uses a combined + * depth+stencil buffer (one 32-bit pixel = 24 bits of Z + 8 bits of stencil). + * + * This causes difficulty because the user may create some number of depth + * renderbuffers and some number of stencil renderbuffers and bind them + * together in framebuffers in any combination. + * + * This code manages all that. + * + * 1. Depth renderbuffers are always allocated in hardware as 32bpp + * GL_DEPTH24_STENCIL8 buffers. + * + * 2. Stencil renderbuffers are initially allocated in software as 8bpp + * GL_STENCIL_INDEX8 buffers. + * + * 3. Depth and Stencil renderbuffers use the PairedStencil and PairedDepth + * fields (respectively) to indicate if the buffer's currently paired + * with another stencil or depth buffer (respectively). + * + * 4. When a depth and stencil buffer are initially both attached to the + * current framebuffer, we merge the stencil buffer values into the + * depth buffer (really a depth+stencil buffer). The then hardware uses + * the combined buffer. + * + * 5. Whenever a depth or stencil buffer is reallocated (with + * glRenderbufferStorage) we undo the pairing and copy the stencil values + * from the combined depth/stencil buffer back to the stencil-only buffer. + * + * 6. We also undo the pairing when we find a change in buffer bindings. + * + * 7. If a framebuffer is only using a depth renderbuffer (no stencil), we + * just use the combined depth/stencil buffer and ignore the stencil values. + * + * 8. If a framebuffer is only using a stencil renderbuffer (no depth) we have + * to promote the 8bpp software stencil buffer to a 32bpp hardware + * depth+stencil buffer. + * + */ + +/** + * Undo the pairing/interleaving between depth and stencil buffers. + * irb should be a depth/stencil or stencil renderbuffer. + */ +void +intel_unpair_depth_stencil(GLcontext *ctx, struct intel_renderbuffer *irb) +{ + struct intel_context *intel = intel_context(ctx); + struct gl_renderbuffer *rb = &irb->Base; + + if (irb->PairedStencil) { + /* irb is a depth/stencil buffer */ + struct gl_renderbuffer *stencilRb; + struct intel_renderbuffer *stencilIrb; + + ASSERT(rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT); + + stencilRb = _mesa_lookup_renderbuffer(ctx, irb->PairedStencil); + stencilIrb = intel_renderbuffer(stencilRb); + if (stencilIrb) { + /* need to extract stencil values from the depth buffer */ + ASSERT(stencilIrb->PairedDepth == rb->Name); + intel_renderbuffer_map(intel, rb); + intel_renderbuffer_map(intel, stencilRb); + _mesa_extract_stencil(ctx, rb, stencilRb); + intel_renderbuffer_unmap(intel, stencilRb); + intel_renderbuffer_unmap(intel, rb); + stencilIrb->PairedDepth = 0; + } + irb->PairedStencil = 0; + } + else if (irb->PairedDepth) { + /* irb is a stencil buffer */ + struct gl_renderbuffer *depthRb; + struct intel_renderbuffer *depthIrb; + + ASSERT(rb->_ActualFormat == GL_STENCIL_INDEX8_EXT || + rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT); + + depthRb = _mesa_lookup_renderbuffer(ctx, irb->PairedDepth); + depthIrb = intel_renderbuffer(depthRb); + if (depthIrb) { + /* need to extract stencil values from the depth buffer */ + ASSERT(depthIrb->PairedStencil == rb->Name); + intel_renderbuffer_map(intel, rb); + intel_renderbuffer_map(intel, depthRb); + _mesa_extract_stencil(ctx, depthRb, rb); + intel_renderbuffer_unmap(intel, depthRb); + intel_renderbuffer_unmap(intel, rb); + depthIrb->PairedStencil = 0; + } + irb->PairedDepth = 0; + } + else { + _mesa_problem(ctx, "Problem in undo_depth_stencil_pairing"); + } + + ASSERT(irb->PairedStencil == 0); + ASSERT(irb->PairedDepth == 0); +} + + +/** + * Examine the depth and stencil renderbuffers which are attached to the + * framebuffer. If both depth and stencil are attached, make sure that the + * renderbuffers are 'paired' (combined). If only depth or only stencil is + * attached, undo any previous pairing. + * + * Must be called if NewState & _NEW_BUFFER (when renderbuffer attachments + * change, for example). + */ +void +intel_validate_paired_depth_stencil(GLcontext * ctx, + struct gl_framebuffer *fb) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_renderbuffer *depthRb, *stencilRb; + + depthRb = intel_get_renderbuffer(fb, BUFFER_DEPTH); + stencilRb = intel_get_renderbuffer(fb, BUFFER_STENCIL); + + if (depthRb && stencilRb) { + if (depthRb == stencilRb) { + /* Using a user-created combined depth/stencil buffer. + * Nothing to do. + */ + ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_STENCIL_EXT); + ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + } + else { + /* Separate depth/stencil buffers, need to interleave now */ + ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_COMPONENT); + ASSERT(stencilRb->Base._BaseFormat == GL_STENCIL_INDEX); + /* may need to interleave depth/stencil now */ + if (depthRb->PairedStencil == stencilRb->Base.Name) { + /* OK, the depth and stencil buffers are already interleaved */ + ASSERT(stencilRb->PairedDepth == depthRb->Base.Name); + } + else { + /* need to setup new pairing/interleaving */ + if (depthRb->PairedStencil) { + intel_unpair_depth_stencil(ctx, depthRb); + } + if (stencilRb->PairedDepth) { + intel_unpair_depth_stencil(ctx, stencilRb); + } + + ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + ASSERT(stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT || + stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + + /* establish new pairing: interleave stencil into depth buffer */ + intel_renderbuffer_map(intel, &depthRb->Base); + intel_renderbuffer_map(intel, &stencilRb->Base); + _mesa_insert_stencil(ctx, &depthRb->Base, &stencilRb->Base); + intel_renderbuffer_unmap(intel, &stencilRb->Base); + intel_renderbuffer_unmap(intel, &depthRb->Base); + depthRb->PairedStencil = stencilRb->Base.Name; + stencilRb->PairedDepth = depthRb->Base.Name; + } + + } + } + else if (depthRb) { + /* Depth buffer but no stencil buffer. + * We'll use a GL_DEPTH24_STENCIL8 buffer and ignore the stencil bits. + */ + /* can't assert this until storage is allocated: + ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + */ + /* intel_undo any previous pairing */ + if (depthRb->PairedStencil) { + intel_unpair_depth_stencil(ctx, depthRb); + } + } + else if (stencilRb) { + /* Stencil buffer but no depth buffer. + * Since h/w doesn't typically support just 8bpp stencil w/out Z, + * we'll use a GL_DEPTH24_STENCIL8 buffer and ignore the depth bits. + */ + /* undo any previous pairing */ + if (stencilRb->PairedDepth) { + intel_unpair_depth_stencil(ctx, stencilRb); + } + if (stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT) { + /* promote buffer to GL_DEPTH24_STENCIL8 for hw rendering */ + _mesa_promote_stencil(ctx, &stencilRb->Base); + ASSERT(stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT); + } + } + + /* Finally, update the fb->_DepthBuffer and fb->_StencilBuffer fields */ + _mesa_update_depth_buffer(ctx, fb, BUFFER_DEPTH); + if (depthRb && depthRb->PairedStencil) + _mesa_update_stencil_buffer(ctx, fb, BUFFER_DEPTH); + else + _mesa_update_stencil_buffer(ctx, fb, BUFFER_STENCIL); + + + /* The hardware should use fb->Attachment[BUFFER_DEPTH].Renderbuffer + * first, if present, then fb->Attachment[BUFFER_STENCIL].Renderbuffer + * if present. + */ +} diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.h b/src/mesa/drivers/dri/intel/intel_depthstencil.h new file mode 100644 index 0000000000..740eb0d989 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_depthstencil.h @@ -0,0 +1,15 @@ + +#ifndef INTEL_DEPTH_STENCIL_H +#define INTEL_DEPTH_STENCIL_H + +#include "intel_fbo.h" + +extern void +intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb); + +extern void +intel_validate_paired_depth_stencil(GLcontext * ctx, + struct gl_framebuffer *fb); + + +#endif /* INTEL_DEPTH_STENCIL_H */ diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c new file mode 100644 index 0000000000..5bd2ebfdcf --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_fbo.c @@ -0,0 +1,709 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "imports.h" +#include "mtypes.h" +#include "fbobject.h" +#include "framebuffer.h" +#include "renderbuffer.h" +#include "context.h" +#include "texformat.h" +#include "texrender.h" + +#include "intel_context.h" +#include "intel_buffers.h" +#include "intel_depthstencil.h" +#include "intel_fbo.h" +#include "intel_mipmap_tree.h" +#include "intel_regions.h" +#include "intel_span.h" + + +#define FILE_DEBUG_FLAG DEBUG_FBO + +#define INTEL_RB_CLASS 0x12345678 + + +/* XXX FBO: move this to intel_context.h (inlined) */ +/** + * Return a gl_renderbuffer ptr casted to intel_renderbuffer. + * NULL will be returned if the rb isn't really an intel_renderbuffer. + * This is determiend by checking the ClassID. + */ +struct intel_renderbuffer * +intel_renderbuffer(struct gl_renderbuffer *rb) +{ + struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb; + if (irb && irb->Base.ClassID == INTEL_RB_CLASS) { + /*_mesa_warning(NULL, "Returning non-intel Rb\n");*/ + return irb; + } + else + return NULL; +} + + +struct intel_renderbuffer * +intel_get_renderbuffer(struct gl_framebuffer *fb, int attIndex) +{ + if (attIndex >= 0) + return intel_renderbuffer(fb->Attachment[attIndex].Renderbuffer); + else + return NULL; +} + + +void +intel_flip_renderbuffers(struct intel_framebuffer *intel_fb) +{ + int current_page = intel_fb->pf_current_page; + int next_page = (current_page + 1) % intel_fb->pf_num_pages; + struct gl_renderbuffer *tmp_rb; + + /* Exchange renderbuffers if necessary but make sure their reference counts + * are preserved. + */ + if (intel_fb->color_rb[current_page] && + intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer != + &intel_fb->color_rb[current_page]->Base) { + tmp_rb = NULL; + _mesa_reference_renderbuffer(&tmp_rb, + intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer); + tmp_rb = &intel_fb->color_rb[current_page]->Base; + _mesa_reference_renderbuffer( + &intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer, tmp_rb); + _mesa_reference_renderbuffer(&tmp_rb, NULL); + } + + if (intel_fb->color_rb[next_page] && + intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer != + &intel_fb->color_rb[next_page]->Base) { + tmp_rb = NULL; + _mesa_reference_renderbuffer(&tmp_rb, + intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer); + tmp_rb = &intel_fb->color_rb[next_page]->Base; + _mesa_reference_renderbuffer( + &intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer, tmp_rb); + _mesa_reference_renderbuffer(&tmp_rb, NULL); + } +} + + +struct intel_region * +intel_get_rb_region(struct gl_framebuffer *fb, GLuint attIndex) +{ + struct intel_renderbuffer *irb = intel_get_renderbuffer(fb, attIndex); + + if (irb) + return irb->region; + else + return NULL; +} + + + +/** + * Create a new framebuffer object. + */ +static struct gl_framebuffer * +intel_new_framebuffer(GLcontext * ctx, GLuint name) +{ + /* Only drawable state in intel_framebuffer at this time, just use Mesa's + * class + */ + return _mesa_new_framebuffer(ctx, name); +} + + +static void +intel_delete_renderbuffer(struct gl_renderbuffer *rb) +{ + GET_CURRENT_CONTEXT(ctx); + struct intel_context *intel = intel_context(ctx); + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + + ASSERT(irb); + + if (irb->PairedStencil || irb->PairedDepth) { + intel_unpair_depth_stencil(ctx, irb); + } + + if (irb->span_cache != NULL) + _mesa_free(irb->span_cache); + + if (intel && irb->region) { + intel_region_release(&irb->region); + } + + _mesa_free(irb); +} + + + +/** + * Return a pointer to a specific pixel in a renderbuffer. + */ +static void * +intel_get_pointer(GLcontext * ctx, struct gl_renderbuffer *rb, + GLint x, GLint y) +{ + /* By returning NULL we force all software rendering to go through + * the span routines. + */ + return NULL; +} + + + +/** + * Called via glRenderbufferStorageEXT() to set the format and allocate + * storage for a user-created renderbuffer. + */ +static GLboolean +intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, + GLuint width, GLuint height) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + GLboolean softwareBuffer = GL_FALSE; + int cpp; + + ASSERT(rb->Name != 0); + + switch (internalFormat) { + case GL_R3_G3_B2: + case GL_RGB4: + case GL_RGB5: + rb->_ActualFormat = GL_RGB5; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 5; + rb->GreenBits = 6; + rb->BlueBits = 5; + cpp = 2; + break; + case GL_RGB: + case GL_RGB8: + case GL_RGB10: + case GL_RGB12: + case GL_RGB16: + rb->_ActualFormat = GL_RGB8; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 8; + rb->GreenBits = 8; + rb->BlueBits = 8; + rb->AlphaBits = 0; + cpp = 4; + break; + case GL_RGBA: + case GL_RGBA2: + case GL_RGBA4: + case GL_RGB5_A1: + case GL_RGBA8: + case GL_RGB10_A2: + case GL_RGBA12: + case GL_RGBA16: + rb->_ActualFormat = GL_RGBA8; + rb->DataType = GL_UNSIGNED_BYTE; + rb->RedBits = 8; + rb->GreenBits = 8; + rb->BlueBits = 8; + rb->AlphaBits = 8; + cpp = 4; + break; + case GL_STENCIL_INDEX: + case GL_STENCIL_INDEX1_EXT: + case GL_STENCIL_INDEX4_EXT: + case GL_STENCIL_INDEX8_EXT: + case GL_STENCIL_INDEX16_EXT: + /* alloc a depth+stencil buffer */ + rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rb->DataType = GL_UNSIGNED_INT_24_8_EXT; + rb->StencilBits = 8; + cpp = 4; + break; + case GL_DEPTH_COMPONENT16: + rb->_ActualFormat = GL_DEPTH_COMPONENT16; + rb->DataType = GL_UNSIGNED_SHORT; + rb->DepthBits = 16; + cpp = 2; + break; + case GL_DEPTH_COMPONENT: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: + rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rb->DataType = GL_UNSIGNED_INT_24_8_EXT; + rb->DepthBits = 24; + cpp = 4; + break; + case GL_DEPTH_STENCIL_EXT: + case GL_DEPTH24_STENCIL8_EXT: + rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT; + rb->DataType = GL_UNSIGNED_INT_24_8_EXT; + rb->DepthBits = 24; + rb->StencilBits = 8; + cpp = 4; + break; + default: + _mesa_problem(ctx, + "Unexpected format in intel_alloc_renderbuffer_storage"); + return GL_FALSE; + } + + intelFlush(ctx); + + /* free old region */ + if (irb->region) { + intel_region_release(&irb->region); + } + + /* allocate new memory region/renderbuffer */ + if (softwareBuffer) { + return _mesa_soft_renderbuffer_storage(ctx, rb, internalFormat, + width, height); + } + else { + /* Choose a pitch to match hardware requirements: + */ + GLuint pitch = ((cpp * width + 63) & ~63) / cpp; + + /* alloc hardware renderbuffer */ + DBG("Allocating %d x %d Intel RBO (pitch %d)\n", width, + height, pitch); + + irb->region = intel_region_alloc(intel, cpp, pitch, height); + if (!irb->region) + return GL_FALSE; /* out of memory? */ + + ASSERT(irb->region->buffer); + + rb->Width = width; + rb->Height = height; + + return GL_TRUE; + } +} + + + +/** + * Called for each hardware renderbuffer when a _window_ is resized. + * Just update fields. + * Not used for user-created renderbuffers! + */ +static GLboolean +intel_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, GLuint width, GLuint height) +{ + ASSERT(rb->Name == 0); + rb->Width = width; + rb->Height = height; + rb->_ActualFormat = internalFormat; + + return GL_TRUE; +} + +static void +intel_resize_buffers(GLcontext *ctx, struct gl_framebuffer *fb, + GLuint width, GLuint height) +{ + struct intel_framebuffer *intel_fb = (struct intel_framebuffer*)fb; + int i; + + _mesa_resize_framebuffer(ctx, fb, width, height); + + fb->Initialized = GL_TRUE; /* XXX remove someday */ + + if (fb->Name != 0) { + return; + } + + /* Make sure all window system renderbuffers are up to date */ + for (i = 0; i < 3; i++) { + struct gl_renderbuffer *rb = &intel_fb->color_rb[i]->Base; + + /* only resize if size is changing */ + if (rb && (rb->Width != width || rb->Height != height)) { + rb->AllocStorage(ctx, rb, rb->InternalFormat, width, height); + } + } +} + +static GLboolean +intel_nop_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb, + GLenum internalFormat, GLuint width, GLuint height) +{ + _mesa_problem(ctx, "intel_op_alloc_storage should never be called."); + return GL_FALSE; +} + + +void +intel_renderbuffer_set_region(struct intel_renderbuffer *rb, + struct intel_region *region) +{ + struct intel_region *old; + + old = rb->region; + rb->region = NULL; + intel_region_reference(&rb->region, region); + intel_region_release(&old); + + rb->pfPitch = region->pitch; +} + +/** + * Create a new intel_renderbuffer which corresponds to an on-screen window, + * not a user-created renderbuffer. + */ +struct intel_renderbuffer * +intel_create_renderbuffer(GLenum intFormat) +{ + GET_CURRENT_CONTEXT(ctx); + + struct intel_renderbuffer *irb; + const GLuint name = 0; + + irb = CALLOC_STRUCT(intel_renderbuffer); + if (!irb) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer"); + return NULL; + } + + _mesa_init_renderbuffer(&irb->Base, name); + irb->Base.ClassID = INTEL_RB_CLASS; + + switch (intFormat) { + case GL_RGB5: + irb->Base._ActualFormat = GL_RGB5; + irb->Base._BaseFormat = GL_RGBA; + irb->Base.RedBits = 5; + irb->Base.GreenBits = 6; + irb->Base.BlueBits = 5; + irb->Base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_RGBA8: + irb->Base._ActualFormat = GL_RGBA8; + irb->Base._BaseFormat = GL_RGBA; + irb->Base.RedBits = 8; + irb->Base.GreenBits = 8; + irb->Base.BlueBits = 8; + irb->Base.AlphaBits = 8; + irb->Base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_STENCIL_INDEX8_EXT: + irb->Base._ActualFormat = GL_STENCIL_INDEX8_EXT; + irb->Base._BaseFormat = GL_STENCIL_INDEX; + irb->Base.StencilBits = 8; + irb->Base.DataType = GL_UNSIGNED_BYTE; + break; + case GL_DEPTH_COMPONENT16: + irb->Base._ActualFormat = GL_DEPTH_COMPONENT16; + irb->Base._BaseFormat = GL_DEPTH_COMPONENT; + irb->Base.DepthBits = 16; + irb->Base.DataType = GL_UNSIGNED_SHORT; + break; + case GL_DEPTH_COMPONENT24: + irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT; + irb->Base._BaseFormat = GL_DEPTH_COMPONENT; + irb->Base.DepthBits = 24; + irb->Base.DataType = GL_UNSIGNED_INT; + break; + case GL_DEPTH24_STENCIL8_EXT: + irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT; + irb->Base._BaseFormat = GL_DEPTH_STENCIL_EXT; + irb->Base.DepthBits = 24; + irb->Base.StencilBits = 8; + irb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT; + break; + default: + _mesa_problem(NULL, + "Unexpected intFormat in intel_create_renderbuffer"); + return NULL; + } + + irb->Base.InternalFormat = intFormat; + + /* intel-specific methods */ + irb->Base.Delete = intel_delete_renderbuffer; + irb->Base.AllocStorage = intel_alloc_window_storage; + irb->Base.GetPointer = intel_get_pointer; + + return irb; +} + + +/** + * Create a new renderbuffer object. + * Typically called via glBindRenderbufferEXT(). + */ +static struct gl_renderbuffer * +intel_new_renderbuffer(GLcontext * ctx, GLuint name) +{ + /*struct intel_context *intel = intel_context(ctx); */ + struct intel_renderbuffer *irb; + + irb = CALLOC_STRUCT(intel_renderbuffer); + if (!irb) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer"); + return NULL; + } + + _mesa_init_renderbuffer(&irb->Base, name); + irb->Base.ClassID = INTEL_RB_CLASS; + + /* intel-specific methods */ + irb->Base.Delete = intel_delete_renderbuffer; + irb->Base.AllocStorage = intel_alloc_renderbuffer_storage; + irb->Base.GetPointer = intel_get_pointer; + /* span routines set in alloc_storage function */ + + return &irb->Base; +} + + +/** + * Called via glBindFramebufferEXT(). + */ +static void +intel_bind_framebuffer(GLcontext * ctx, GLenum target, + struct gl_framebuffer *fb, struct gl_framebuffer *fbread) +{ + if (target == GL_FRAMEBUFFER_EXT || target == GL_DRAW_FRAMEBUFFER_EXT) { + intel_draw_buffer(ctx, fb); + /* Integer depth range depends on depth buffer bits */ + if (ctx->Driver.DepthRange != NULL) + ctx->Driver.DepthRange(ctx, ctx->Viewport.Near, ctx->Viewport.Far); + } + else { + /* don't need to do anything if target == GL_READ_FRAMEBUFFER_EXT */ + } +} + + +/** + * Called via glFramebufferRenderbufferEXT(). + */ +static void +intel_framebuffer_renderbuffer(GLcontext * ctx, + struct gl_framebuffer *fb, + GLenum attachment, struct gl_renderbuffer *rb) +{ + DBG("Intel FramebufferRenderbuffer %u %u\n", fb->Name, rb ? rb->Name : 0); + + intelFlush(ctx); + + _mesa_framebuffer_renderbuffer(ctx, fb, attachment, rb); + intel_draw_buffer(ctx, fb); +} + +static GLboolean +intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb, + struct gl_texture_image *texImage) +{ + if (texImage->TexFormat == &_mesa_texformat_argb8888) { + irb->Base._ActualFormat = GL_RGBA8; + irb->Base._BaseFormat = GL_RGBA; + DBG("Render to RGBA8 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_rgb565) { + irb->Base._ActualFormat = GL_RGB5; + irb->Base._BaseFormat = GL_RGB; + DBG("Render to RGB5 texture OK\n"); + } + else if (texImage->TexFormat == &_mesa_texformat_z16) { + irb->Base._ActualFormat = GL_DEPTH_COMPONENT16; + irb->Base._BaseFormat = GL_DEPTH_COMPONENT; + DBG("Render to DEPTH16 texture OK\n"); + } else if (texImage->TexFormat == &_mesa_texformat_z24_s8) { + irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT; + irb->Base._BaseFormat = GL_DEPTH_STENCIL_EXT; + DBG("Render to DEPTH_STENCIL texture OK\n"); + } + else { + DBG("Render to texture BAD FORMAT %d\n", + texImage->TexFormat->MesaFormat); + return GL_FALSE; + } + + irb->Base.InternalFormat = irb->Base._ActualFormat; + irb->Base.Width = texImage->Width; + irb->Base.Height = texImage->Height; + irb->Base.DataType = GL_UNSIGNED_BYTE; /* FBO XXX fix */ + irb->Base.RedBits = texImage->TexFormat->RedBits; + irb->Base.GreenBits = texImage->TexFormat->GreenBits; + irb->Base.BlueBits = texImage->TexFormat->BlueBits; + irb->Base.AlphaBits = texImage->TexFormat->AlphaBits; + irb->Base.DepthBits = texImage->TexFormat->DepthBits; + + irb->Base.Delete = intel_delete_renderbuffer; + irb->Base.AllocStorage = intel_nop_alloc_storage; + + irb->RenderToTexture = GL_TRUE; + + return GL_TRUE; +} + +/** + * When glFramebufferTexture[123]D is called this function sets up the + * gl_renderbuffer wrapper around the texture image. + * This will have the region info needed for hardware rendering. + */ +static struct intel_renderbuffer * +intel_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage) +{ + const GLuint name = ~0; /* not significant, but distinct for debugging */ + struct intel_renderbuffer *irb; + + /* make an intel_renderbuffer to wrap the texture image */ + irb = CALLOC_STRUCT(intel_renderbuffer); + if (!irb) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glFramebufferTexture"); + return NULL; + } + + _mesa_init_renderbuffer(&irb->Base, name); + irb->Base.ClassID = INTEL_RB_CLASS; + + if (!intel_update_wrapper(ctx, irb, texImage)) { + _mesa_free(irb); + return NULL; + } + + return irb; +} + + +/** + * Called by glFramebufferTexture[123]DEXT() (and other places) to + * prepare for rendering into texture memory. This might be called + * many times to choose different texture levels, cube faces, etc + * before intel_finish_render_texture() is ever called. + */ +static void +intel_render_texture(GLcontext * ctx, + struct gl_framebuffer *fb, + struct gl_renderbuffer_attachment *att) +{ + struct gl_texture_image *newImage + = att->Texture->Image[att->CubeMapFace][att->TextureLevel]; + struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer); + struct intel_texture_image *intel_image; + GLuint imageOffset; + + (void) fb; + + ASSERT(newImage); + + if (!irb) { + irb = intel_wrap_texture(ctx, newImage); + if (irb) { + /* bind the wrapper to the attachment point */ + _mesa_reference_renderbuffer(&att->Renderbuffer, &irb->Base); + } + else { + /* fallback to software rendering */ + _mesa_render_texture(ctx, fb, att); + return; + } + } if (!intel_update_wrapper(ctx, irb, newImage)) { + _mesa_reference_renderbuffer(&att->Renderbuffer, NULL); + _mesa_render_texture(ctx, fb, att); + return; + } + + DBG("Begin render texture tid %x tex=%u w=%d h=%d refcount=%d\n", + _glthread_GetID(), + att->Texture->Name, newImage->Width, newImage->Height, + irb->Base.RefCount); + + /* point the renderbufer's region to the texture image region */ + intel_image = intel_texture_image(newImage); + if (irb->region != intel_image->mt->region) { + if (irb->region) + intel_region_release(&irb->region); + intel_region_reference(&irb->region, intel_image->mt->region); + } + + /* compute offset of the particular 2D image within the texture region */ + imageOffset = intel_miptree_image_offset(intel_image->mt, + att->CubeMapFace, + att->TextureLevel); + + if (att->Texture->Target == GL_TEXTURE_3D) { + const GLuint *offsets = intel_miptree_depth_offsets(intel_image->mt, + att->TextureLevel); + imageOffset += offsets[att->Zoffset]; + } + + /* store that offset in the region */ + intel_image->mt->region->draw_offset = imageOffset; + + /* update drawing region, etc */ + intel_draw_buffer(ctx, fb); +} + + +/** + * Called by Mesa when rendering to a texture is done. + */ +static void +intel_finish_render_texture(GLcontext * ctx, + struct gl_renderbuffer_attachment *att) +{ + struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer); + + DBG("End render texture (tid %x) tex %u\n", _glthread_GetID(), att->Texture->Name); + + if (irb) { + /* just release the region */ + intel_region_release(&irb->region); + } + else if (att->Renderbuffer) { + /* software fallback */ + _mesa_finish_render_texture(ctx, att); + /* XXX FBO: Need to unmap the buffer (or in intelSpanRenderStart???) */ + } +} + + +/** + * Do one-time context initializations related to GL_EXT_framebuffer_object. + * Hook in device driver functions. + */ +void +intel_fbo_init(struct intel_context *intel) +{ + intel->ctx.Driver.NewFramebuffer = intel_new_framebuffer; + intel->ctx.Driver.NewRenderbuffer = intel_new_renderbuffer; + intel->ctx.Driver.BindFramebuffer = intel_bind_framebuffer; + intel->ctx.Driver.FramebufferRenderbuffer = intel_framebuffer_renderbuffer; + intel->ctx.Driver.RenderTexture = intel_render_texture; + intel->ctx.Driver.FinishRenderTexture = intel_finish_render_texture; + intel->ctx.Driver.ResizeBuffers = intel_resize_buffers; +} diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h new file mode 100644 index 0000000000..9d15582d78 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_fbo.h @@ -0,0 +1,115 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_FBO_H +#define INTEL_FBO_H + +#include "intel_screen.h" + +struct intel_context; + +/** + * Intel framebuffer, derived from gl_framebuffer. + */ +struct intel_framebuffer +{ + struct gl_framebuffer Base; + + struct intel_renderbuffer *color_rb[3]; + + /* Drawable page flipping state */ + GLboolean pf_active; + GLuint pf_seq; + GLint pf_planes; + GLint pf_current_page; + GLint pf_num_pages; + + /* VBI + */ + GLuint vbl_waited; + + int64_t swap_ust; + int64_t swap_missed_ust; + + GLuint swap_count; + GLuint swap_missed_count; +}; + + +/** + * Intel renderbuffer, derived from gl_renderbuffer. + * Note: The PairedDepth and PairedStencil fields use renderbuffer IDs, + * not pointers because in some circumstances a deleted renderbuffer could + * result in a dangling pointer here. + */ +struct intel_renderbuffer +{ + struct gl_renderbuffer Base; + struct intel_region *region; + GLuint pfPitch; /* possibly paged flipped pitch */ + GLboolean RenderToTexture; /* RTT? */ + + GLuint PairedDepth; /**< only used if this is a depth renderbuffer */ + GLuint PairedStencil; /**< only used if this is a stencil renderbuffer */ + + GLuint pf_pending; /**< sequence number of pending flip */ + + GLuint vbl_pending; /**< vblank sequence number of pending flip */ + + uint8_t *span_cache; + unsigned long span_cache_offset; +}; + +extern struct intel_renderbuffer *intel_renderbuffer(struct gl_renderbuffer + *rb); + +extern void +intel_renderbuffer_set_region(struct intel_renderbuffer *irb, + struct intel_region *region); + +extern struct intel_renderbuffer * +intel_create_renderbuffer(GLenum intFormat); + +extern void intel_fbo_init(struct intel_context *intel); + + +/* XXX make inline or macro */ +extern struct intel_renderbuffer *intel_get_renderbuffer(struct gl_framebuffer + *fb, + int attIndex); + +extern void intel_flip_renderbuffers(struct intel_framebuffer *intel_fb); + + +/* XXX make inline or macro */ +extern struct intel_region *intel_get_rb_region(struct gl_framebuffer *fb, + GLuint attIndex); + + + + +#endif /* INTEL_FBO_H */ diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c new file mode 100644 index 0000000000..9be7e02eff --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c @@ -0,0 +1,489 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "intel_context.h" +#include "intel_mipmap_tree.h" +#include "intel_regions.h" +#include "intel_chipset.h" +#include "enums.h" + +#define FILE_DEBUG_FLAG DEBUG_MIPTREE + +static GLenum +target_to_target(GLenum target) +{ + switch (target) { + case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB: + return GL_TEXTURE_CUBE_MAP_ARB; + default: + return target; + } +} + +static struct intel_mipmap_tree * +intel_miptree_create_internal(struct intel_context *intel, + GLenum target, + GLenum internal_format, + GLuint first_level, + GLuint last_level, + GLuint width0, + GLuint height0, + GLuint depth0, GLuint cpp, GLuint compress_byte) +{ + GLboolean ok; + struct intel_mipmap_tree *mt = calloc(sizeof(*mt), 1); + + DBG("%s target %s format %s level %d..%d\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(target), + _mesa_lookup_enum_by_nr(internal_format), first_level, last_level); + + mt->target = target_to_target(target); + mt->internal_format = internal_format; + mt->first_level = first_level; + mt->last_level = last_level; + mt->width0 = width0; + mt->height0 = height0; + mt->depth0 = depth0; + mt->cpp = compress_byte ? compress_byte : cpp; + mt->compressed = compress_byte ? 1 : 0; + mt->refcount = 1; + mt->pitch = 0; + +#ifdef I915 + if (IS_945(intel->intelScreen->deviceID)) + ok = i945_miptree_layout(intel, mt); + else + ok = i915_miptree_layout(intel, mt); +#else + ok = brw_miptree_layout(intel, mt); +#endif + + if (!ok) { + free(mt); + return NULL; + } + + return mt; +} + +struct intel_mipmap_tree * +intel_miptree_create(struct intel_context *intel, + GLenum target, + GLenum internal_format, + GLuint first_level, + GLuint last_level, + GLuint width0, + GLuint height0, + GLuint depth0, GLuint cpp, GLuint compress_byte) +{ + struct intel_mipmap_tree *mt; + + mt = intel_miptree_create_internal(intel, target, internal_format, + first_level, last_level, width0, + height0, depth0, cpp, compress_byte); + /* + * pitch == 0 indicates the null texture + */ + if (!mt || !mt->pitch) + return NULL; + + mt->region = intel_region_alloc(intel, + mt->cpp, mt->pitch, mt->total_height); + + if (!mt->region) { + free(mt); + return NULL; + } + + return mt; +} + +struct intel_mipmap_tree * +intel_miptree_create_for_region(struct intel_context *intel, + GLenum target, + GLenum internal_format, + GLuint first_level, + GLuint last_level, + struct intel_region *region, + GLuint depth0, + GLuint compress_byte) +{ + struct intel_mipmap_tree *mt; + + mt = intel_miptree_create_internal(intel, target, internal_format, + first_level, last_level, + region->pitch, region->height, depth0, + region->cpp, compress_byte); + if (!mt) + return mt; +#if 0 + if (mt->pitch != region->pitch) { + fprintf(stderr, + "region pitch (%d) doesn't match mipmap tree pitch (%d)\n", + region->pitch, mt->pitch); + free(mt); + return NULL; + } +#else + /* The mipmap tree pitch is aligned to 64 bytes to make sure render + * to texture works, but we don't need that for texturing from a + * pixmap. Just override it here. */ + mt->pitch = region->pitch; +#endif + + mt->region = region; + + return mt; + } + +/** + * intel_miptree_pitch_align: + * + * @intel: intel context pointer + * + * @mt: the miptree to compute pitch alignment for + * + * @pitch: the natural pitch value + * + * Given @pitch, compute a larger value which accounts for + * any necessary alignment required by the device + */ + +int intel_miptree_pitch_align (struct intel_context *intel, + struct intel_mipmap_tree *mt, + int pitch) +{ +#ifdef I915 + GLcontext *ctx = &intel->ctx; +#endif + + if (!mt->compressed) { + int pitch_align; + + if (intel->ttm) { + /* XXX: Align pitch to multiple of 64 bytes for now to allow + * render-to-texture to work in all cases. This should probably be + * replaced at some point by some scheme to only do this when really + * necessary. + */ + pitch_align = 64; + } else { + pitch_align = 4; + } + + pitch = ALIGN(pitch * mt->cpp, pitch_align); + +#ifdef I915 + /* XXX: At least the i915 seems very upset when the pitch is a multiple + * of 1024 and sometimes 512 bytes - performance can drop by several + * times. Go to the next multiple of the required alignment for now. + */ + if (!(pitch & 511) && + (pitch + pitch_align) < (1 << ctx->Const.MaxTextureLevels)) + pitch += pitch_align; +#endif + + pitch /= mt->cpp; + } + return pitch; +} + +void +intel_miptree_reference(struct intel_mipmap_tree **dst, + struct intel_mipmap_tree *src) +{ + src->refcount++; + *dst = src; + DBG("%s %p refcount now %d\n", __FUNCTION__, src, src->refcount); +} + +void +intel_miptree_release(struct intel_context *intel, + struct intel_mipmap_tree **mt) +{ + if (!*mt) + return; + + DBG("%s %p refcount will be %d\n", __FUNCTION__, *mt, (*mt)->refcount - 1); + if (--(*mt)->refcount <= 0) { + GLuint i; + + DBG("%s deleting %p\n", __FUNCTION__, *mt); + + intel_region_release(&((*mt)->region)); + + for (i = 0; i < MAX_TEXTURE_LEVELS; i++) + if ((*mt)->level[i].image_offset) + free((*mt)->level[i].image_offset); + + free(*mt); + } + *mt = NULL; +} + + + + +/* Can the image be pulled into a unified mipmap tree. This mirrors + * the completeness test in a lot of ways. + * + * Not sure whether I want to pass gl_texture_image here. + */ +GLboolean +intel_miptree_match_image(struct intel_mipmap_tree *mt, + struct gl_texture_image *image, + GLuint face, GLuint level) +{ + /* Images with borders are never pulled into mipmap trees. + */ + if (image->Border || + ((image->_BaseFormat == GL_DEPTH_COMPONENT) && + ((image->TexObject->WrapS == GL_CLAMP_TO_BORDER) || + (image->TexObject->WrapT == GL_CLAMP_TO_BORDER)))) + return GL_FALSE; + + if (image->InternalFormat != mt->internal_format || + image->IsCompressed != mt->compressed) + return GL_FALSE; + + if (!image->IsCompressed && + !mt->compressed && + image->TexFormat->TexelBytes != mt->cpp) + return GL_FALSE; + + /* Test image dimensions against the base level image adjusted for + * minification. This will also catch images not present in the + * tree, changed targets, etc. + */ + if (image->Width != mt->level[level].width || + image->Height != mt->level[level].height || + image->Depth != mt->level[level].depth) + return GL_FALSE; + + return GL_TRUE; +} + + +void +intel_miptree_set_level_info(struct intel_mipmap_tree *mt, + GLuint level, + GLuint nr_images, + GLuint x, GLuint y, + GLuint w, GLuint h, GLuint d) +{ + mt->level[level].width = w; + mt->level[level].height = h; + mt->level[level].depth = d; + mt->level[level].level_offset = (x + y * mt->pitch) * mt->cpp; + mt->level[level].nr_images = nr_images; + + DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__, + level, w, h, d, x, y, mt->level[level].level_offset); + + /* Not sure when this would happen, but anyway: + */ + if (mt->level[level].image_offset) { + free(mt->level[level].image_offset); + mt->level[level].image_offset = NULL; + } + + assert(nr_images); + + mt->level[level].image_offset = malloc(nr_images * sizeof(GLuint)); + mt->level[level].image_offset[0] = 0; +} + + + +void +intel_miptree_set_image_offset(struct intel_mipmap_tree *mt, + GLuint level, GLuint img, + GLuint x, GLuint y) +{ + if (img == 0 && level == 0) + assert(x == 0 && y == 0); + + assert(img < mt->level[level].nr_images); + + mt->level[level].image_offset[img] = (x + y * mt->pitch) * mt->cpp; + + DBG("%s level %d img %d pos %d,%d image_offset %x\n", + __FUNCTION__, level, img, x, y, mt->level[level].image_offset[img]); +} + + +/* Although we use the image_offset[] array to store relative offsets + * to cube faces, Mesa doesn't know anything about this and expects + * each cube face to be treated as a separate image. + * + * These functions present that view to mesa: + */ +const GLuint * +intel_miptree_depth_offsets(struct intel_mipmap_tree *mt, GLuint level) +{ + static const GLuint zero = 0; + + if (mt->target != GL_TEXTURE_3D || mt->level[level].nr_images == 1) + return &zero; + else + return mt->level[level].image_offset; +} + + +GLuint +intel_miptree_image_offset(struct intel_mipmap_tree *mt, + GLuint face, GLuint level) +{ + if (mt->target == GL_TEXTURE_CUBE_MAP_ARB) + return (mt->level[level].level_offset + + mt->level[level].image_offset[face]); + else + return mt->level[level].level_offset; +} + + + +/** + * Map a teximage in a mipmap tree. + * \param row_stride returns row stride in bytes + * \param image_stride returns image stride in bytes (for 3D textures). + * \param image_offsets pointer to array of pixel offsets from the returned + * pointer to each depth image + * \return address of mapping + */ +GLubyte * +intel_miptree_image_map(struct intel_context * intel, + struct intel_mipmap_tree * mt, + GLuint face, + GLuint level, + GLuint * row_stride, GLuint * image_offsets) +{ + DBG("%s \n", __FUNCTION__); + + if (row_stride) + *row_stride = mt->pitch * mt->cpp; + + if (mt->target == GL_TEXTURE_3D) { + int i; + + for (i = 0; i < mt->level[level].depth; i++) + image_offsets[i] = mt->level[level].image_offset[i] / mt->cpp; + } else { + assert(mt->level[level].depth == 1); + assert(mt->target == GL_TEXTURE_CUBE_MAP || + mt->level[level].image_offset[0] == 0); + image_offsets[0] = 0; + } + + return (intel_region_map(intel, mt->region) + + intel_miptree_image_offset(mt, face, level)); +} + +void +intel_miptree_image_unmap(struct intel_context *intel, + struct intel_mipmap_tree *mt) +{ + DBG("%s\n", __FUNCTION__); + intel_region_unmap(intel, mt->region); +} + + + +/* Upload data for a particular image. + */ +void +intel_miptree_image_data(struct intel_context *intel, + struct intel_mipmap_tree *dst, + GLuint face, + GLuint level, + void *src, + GLuint src_row_pitch, + GLuint src_image_pitch) +{ + GLuint depth = dst->level[level].depth; + GLuint dst_offset = intel_miptree_image_offset(dst, face, level); + const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level); + GLuint i; + GLuint height = 0; + + DBG("%s: %d/%d\n", __FUNCTION__, face, level); + for (i = 0; i < depth; i++) { + height = dst->level[level].height; + if(dst->compressed) + height = (height + 3) / 4; + intel_region_data(intel, + dst->region, + dst_offset + dst_depth_offset[i], /* dst_offset */ + 0, 0, /* dstx, dsty */ + src, + src_row_pitch, + 0, 0, /* source x, y */ + dst->level[level].width, height); /* width, height */ + + src += src_image_pitch * dst->cpp; + } +} + +extern GLuint intel_compressed_alignment(GLenum); +/* Copy mipmap image between trees + */ +void +intel_miptree_image_copy(struct intel_context *intel, + struct intel_mipmap_tree *dst, + GLuint face, GLuint level, + struct intel_mipmap_tree *src) +{ + GLuint width = src->level[level].width; + GLuint height = src->level[level].height; + GLuint depth = src->level[level].depth; + GLuint dst_offset = intel_miptree_image_offset(dst, face, level); + GLuint src_offset = intel_miptree_image_offset(src, face, level); + const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level); + const GLuint *src_depth_offset = intel_miptree_depth_offsets(src, level); + GLuint i; + + if (dst->compressed) { + GLuint alignment = intel_compressed_alignment(dst->internal_format); + height = (height + 3) / 4; + width = ((width + alignment - 1) & ~(alignment - 1)); + } + + for (i = 0; i < depth; i++) { + intel_region_copy(intel, + dst->region, dst_offset + dst_depth_offset[i], + 0, + 0, + src->region, src_offset + src_depth_offset[i], + 0, 0, width, height); + } + +} diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h new file mode 100644 index 0000000000..c9537dbb9a --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h @@ -0,0 +1,226 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_MIPMAP_TREE_H +#define INTEL_MIPMAP_TREE_H + +#include "intel_regions.h" + +/* A layer on top of the intel_regions code which adds: + * + * - Code to size and layout a region to hold a set of mipmaps. + * - Query to determine if a new image fits in an existing tree. + * - More refcounting + * - maybe able to remove refcounting from intel_region? + * - ? + * + * The fixed mipmap layout of intel hardware where one offset + * specifies the position of all images in a mipmap hierachy + * complicates the implementation of GL texture image commands, + * compared to hardware where each image is specified with an + * independent offset. + * + * In an ideal world, each texture object would be associated with a + * single bufmgr buffer or 2d intel_region, and all the images within + * the texture object would slot into the tree as they arrive. The + * reality can be a little messier, as images can arrive from the user + * with sizes that don't fit in the existing tree, or in an order + * where the tree layout cannot be guessed immediately. + * + * This structure encodes an idealized mipmap tree. The GL image + * commands build these where possible, otherwise store the images in + * temporary system buffers. + */ + + +/** + * Describes the location of each texture image within a texture region. + */ +struct intel_mipmap_level +{ + /** + * Byte offset to the base of this level. + * + * This is used for mipmap levels of 1D/2D/3D textures. However, CUBE + * layouts spread images around the whole tree, so the level offset is + * always zero in that case. + */ + GLuint level_offset; + GLuint width; + GLuint height; + /** Depth of the mipmap at this level: 1 for 1D/2D/CUBE, n for 3D. */ + GLuint depth; + /** Number of images at this level: 1 for 1D/2D, 6 for CUBE, depth for 3D */ + GLuint nr_images; + + /** + * Byte offset from level_offset to the image for each cube face or depth + * level. + * + * Pretty much have to accept that hardware formats + * are going to be so diverse that there is no unified way to + * compute the offsets of depth/cube images within a mipmap level, + * so have to store them as a lookup table. + */ + GLuint *image_offset; +}; + +struct intel_mipmap_tree +{ + /* Effectively the key: + */ + GLenum target; + GLenum internal_format; + + GLuint first_level; + GLuint last_level; + + GLuint width0, height0, depth0; /**< Level zero image dimensions */ + GLuint cpp; + GLboolean compressed; + + /* Derived from the above: + */ + GLuint pitch; + GLuint depth_pitch; /* per-image on i945? */ + GLuint total_height; + + /* Includes image offset tables: + */ + struct intel_mipmap_level level[MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct intel_region *region; + + /* These are also refcounted: + */ + GLuint refcount; +}; + + + +struct intel_mipmap_tree *intel_miptree_create(struct intel_context *intel, + GLenum target, + GLenum internal_format, + GLuint first_level, + GLuint last_level, + GLuint width0, + GLuint height0, + GLuint depth0, + GLuint cpp, + GLuint compress_byte); + +struct intel_mipmap_tree * +intel_miptree_create_for_region(struct intel_context *intel, + GLenum target, + GLenum internal_format, + GLuint first_level, + GLuint last_level, + struct intel_region *region, + GLuint depth0, + GLuint compress_byte); + +int intel_miptree_pitch_align (struct intel_context *intel, + struct intel_mipmap_tree *mt, + int pitch); + +void intel_miptree_reference(struct intel_mipmap_tree **dst, + struct intel_mipmap_tree *src); + +void intel_miptree_release(struct intel_context *intel, + struct intel_mipmap_tree **mt); + +/* Check if an image fits an existing mipmap tree layout + */ +GLboolean intel_miptree_match_image(struct intel_mipmap_tree *mt, + struct gl_texture_image *image, + GLuint face, GLuint level); + +/* Return a pointer to an image within a tree. Return image stride as + * well. + */ +GLubyte *intel_miptree_image_map(struct intel_context *intel, + struct intel_mipmap_tree *mt, + GLuint face, + GLuint level, + GLuint * row_stride, GLuint * image_stride); + +void intel_miptree_image_unmap(struct intel_context *intel, + struct intel_mipmap_tree *mt); + + +/* Return the linear offset of an image relative to the start of the + * tree: + */ +GLuint intel_miptree_image_offset(struct intel_mipmap_tree *mt, + GLuint face, GLuint level); + +/* Return pointers to each 2d slice within an image. Indexed by depth + * value. + */ +const GLuint *intel_miptree_depth_offsets(struct intel_mipmap_tree *mt, + GLuint level); + + +void intel_miptree_set_level_info(struct intel_mipmap_tree *mt, + GLuint level, + GLuint nr_images, + GLuint x, GLuint y, + GLuint w, GLuint h, GLuint d); + +void intel_miptree_set_image_offset(struct intel_mipmap_tree *mt, + GLuint level, + GLuint img, GLuint x, GLuint y); + + +/* Upload an image into a tree + */ +void intel_miptree_image_data(struct intel_context *intel, + struct intel_mipmap_tree *dst, + GLuint face, + GLuint level, + void *src, + GLuint src_row_pitch, GLuint src_image_pitch); + +/* Copy an image between two trees + */ +void intel_miptree_image_copy(struct intel_context *intel, + struct intel_mipmap_tree *dst, + GLuint face, GLuint level, + struct intel_mipmap_tree *src); + +/* i915_mipmap_tree.c: + */ +GLboolean i915_miptree_layout(struct intel_context *intel, + struct intel_mipmap_tree *mt); +GLboolean i945_miptree_layout(struct intel_context *intel, + struct intel_mipmap_tree *mt); +GLboolean brw_miptree_layout(struct intel_context *intel, + struct intel_mipmap_tree *mt); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_pixel.c b/src/mesa/drivers/dri/intel/intel_pixel.c new file mode 100644 index 0000000000..6417866b20 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_pixel.c @@ -0,0 +1,189 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portionsalloc + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "enums.h" +#include "state.h" +#include "swrast/swrast.h" + +#include "intel_context.h" +#include "intel_pixel.h" +#include "intel_regions.h" + +#define FILE_DEBUG_FLAG DEBUG_PIXEL + +static GLenum +effective_func(GLenum func, GLboolean src_alpha_is_one) +{ + if (src_alpha_is_one) { + if (func == GL_SRC_ALPHA) + return GL_ONE; + if (func == GL_ONE_MINUS_SRC_ALPHA) + return GL_ZERO; + } + + return func; +} + +/** + * Check if any fragment operations are in effect which might effect + * glDraw/CopyPixels. + */ +GLboolean +intel_check_blit_fragment_ops(GLcontext * ctx, GLboolean src_alpha_is_one) +{ + if (ctx->NewState) + _mesa_update_state(ctx); + + if (ctx->FragmentProgram._Enabled) { + DBG("fallback due to fragment program\n"); + return GL_FALSE; + } + + if (ctx->Color.BlendEnabled && + (effective_func(ctx->Color.BlendSrcRGB, src_alpha_is_one) != GL_ONE || + effective_func(ctx->Color.BlendDstRGB, src_alpha_is_one) != GL_ZERO || + ctx->Color.BlendEquationRGB != GL_FUNC_ADD || + effective_func(ctx->Color.BlendSrcA, src_alpha_is_one) != GL_ONE || + effective_func(ctx->Color.BlendDstA, src_alpha_is_one) != GL_ZERO || + ctx->Color.BlendEquationA != GL_FUNC_ADD)) { + DBG("fallback due to blend\n"); + return GL_FALSE; + } + + if (ctx->Texture._EnabledUnits) { + DBG("fallback due to texturing\n"); + return GL_FALSE; + } + + if (!(ctx->Color.ColorMask[0] && + ctx->Color.ColorMask[1] && + ctx->Color.ColorMask[2] && + ctx->Color.ColorMask[3])) { + DBG("fallback due to color masking\n"); + return GL_FALSE; + } + + if (ctx->Color.AlphaEnabled) { + DBG("fallback due to alpha\n"); + return GL_FALSE; + } + + if (ctx->Depth.Test) { + DBG("fallback due to depth test\n"); + return GL_FALSE; + } + + if (ctx->Fog.Enabled) { + DBG("fallback due to fog\n"); + return GL_FALSE; + } + + if (ctx->_ImageTransferState) { + DBG("fallback due to image transfer\n"); + return GL_FALSE; + } + + if (ctx->Stencil.Enabled) { + DBG("fallback due to image stencil\n"); + return GL_FALSE; + } + + if (ctx->Scissor.Enabled) { + /* XXX Note: Scissor could be done with the blitter */ + DBG("fallback due to image scissor\n"); + return GL_FALSE; + } + + if (ctx->RenderMode != GL_RENDER) { + DBG("fallback due to render mode\n"); + return GL_FALSE; + } + + return GL_TRUE; +} + + +GLboolean +intel_check_meta_tex_fragment_ops(GLcontext * ctx) +{ + if (ctx->NewState) + _mesa_update_state(ctx); + + /* Some of _ImageTransferState (scale, bias) could be done with + * fragment programs on i915. + */ + return !(ctx->_ImageTransferState || ctx->Fog.Enabled || /* not done yet */ + ctx->Texture._EnabledUnits || ctx->FragmentProgram._Enabled); +} + +/* The intel_region struct doesn't really do enough to capture the + * format of the pixels in the region. For now this code assumes that + * the region is a display surface and hence is either ARGB8888 or + * RGB565. + * XXX FBO: If we'd pass in the intel_renderbuffer instead of region, we'd + * know the buffer's pixel format. + * + * \param format as given to glDraw/ReadPixels + * \param type as given to glDraw/ReadPixels + */ +GLboolean +intel_check_blit_format(struct intel_region * region, + GLenum format, GLenum type) +{ + if (region->cpp == 4 && + (type == GL_UNSIGNED_INT_8_8_8_8_REV || + type == GL_UNSIGNED_BYTE) && format == GL_BGRA) { + return GL_TRUE; + } + + if (region->cpp == 2 && + type == GL_UNSIGNED_SHORT_5_6_5_REV && format == GL_BGR) { + return GL_TRUE; + } + + if (INTEL_DEBUG & DEBUG_PIXEL) + fprintf(stderr, "%s: bad format for blit (cpp %d, type %s format %s)\n", + __FUNCTION__, region->cpp, + _mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format)); + + return GL_FALSE; +} + + +void +intelInitPixelFuncs(struct dd_function_table *functions) +{ + functions->Accum = _swrast_Accum; + if (!getenv("INTEL_NO_BLIT")) { + functions->Bitmap = intelBitmap; + functions->CopyPixels = intelCopyPixels; +#ifdef I915 + functions->ReadPixels = intelReadPixels; + functions->DrawPixels = intelDrawPixels; +#endif + } +} diff --git a/src/mesa/drivers/dri/intel/intel_pixel.h b/src/mesa/drivers/dri/intel/intel_pixel.h new file mode 100644 index 0000000000..9c899b954c --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_pixel.h @@ -0,0 +1,70 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_PIXEL_H +#define INTEL_PIXEL_H + +#include "mtypes.h" + +void intelInitPixelFuncs(struct dd_function_table *functions); + +GLboolean intel_check_blit_fragment_ops(GLcontext * ctx, + GLboolean src_alpha_is_one); + +GLboolean intel_check_meta_tex_fragment_ops(GLcontext * ctx); + +GLboolean intel_check_blit_format(struct intel_region *region, + GLenum format, GLenum type); + + +void intelReadPixels(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *pack, + GLvoid * pixels); + +void intelDrawPixels(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + GLenum format, + GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid * pixels); + +void intelCopyPixels(GLcontext * ctx, + GLint srcx, GLint srcy, + GLsizei width, GLsizei height, + GLint destx, GLint desty, GLenum type); + +void intelBitmap(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + const struct gl_pixelstore_attrib *unpack, + const GLubyte * pixels); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c new file mode 100644 index 0000000000..daa6d4be17 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c @@ -0,0 +1,356 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portionsalloc + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "enums.h" +#include "image.h" +#include "colormac.h" +#include "mtypes.h" +#include "macros.h" +#include "bufferobj.h" +#include "swrast/swrast.h" + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_blit.h" +#include "intel_regions.h" +#include "intel_buffer_objects.h" +#include "intel_buffers.h" +#include "intel_pixel.h" +#include "intel_reg.h" + + +#define FILE_DEBUG_FLAG DEBUG_PIXEL + + +/* Unlike the other intel_pixel_* functions, the expectation here is + * that the incoming data is not in a PBO. With the XY_TEXT blit + * method, there's no benefit haveing it in a PBO, but we could + * implement a path based on XY_MONO_SRC_COPY_BLIT which might benefit + * PBO bitmaps. I think they are probably pretty rare though - I + * wonder if Xgl uses them? + */ +static const GLubyte *map_pbo( GLcontext *ctx, + GLsizei width, GLsizei height, + const struct gl_pixelstore_attrib *unpack, + const GLubyte *bitmap ) +{ + GLubyte *buf; + + if (!_mesa_validate_pbo_access(2, unpack, width, height, 1, + GL_COLOR_INDEX, GL_BITMAP, + (GLvoid *) bitmap)) { + _mesa_error(ctx, GL_INVALID_OPERATION,"glBitmap(invalid PBO access)"); + return NULL; + } + + buf = (GLubyte *) ctx->Driver.MapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT, + GL_READ_ONLY_ARB, + unpack->BufferObj); + if (!buf) { + _mesa_error(ctx, GL_INVALID_OPERATION, "glBitmap(PBO is mapped)"); + return NULL; + } + + return ADD_POINTERS(buf, bitmap); +} + +static GLboolean test_bit( const GLubyte *src, + GLuint bit ) +{ + return (src[bit/8] & (1<<(bit % 8))) ? 1 : 0; +} + +static void set_bit( GLubyte *dest, + GLuint bit ) +{ + dest[bit/8] |= 1 << (bit % 8); +} + +/* Extract a rectangle's worth of data from the bitmap. Called + * per-cliprect. + */ +static GLuint get_bitmap_rect(GLsizei width, GLsizei height, + const struct gl_pixelstore_attrib *unpack, + const GLubyte *bitmap, + GLuint x, GLuint y, + GLuint w, GLuint h, + GLubyte *dest, + GLuint row_align, + GLboolean invert) +{ + GLuint src_offset = (x + unpack->SkipPixels) & 0x7; + GLuint mask = unpack->LsbFirst ? 0 : 7; + GLuint bit = 0; + GLint row, col; + GLint first, last; + GLint incr; + GLuint count = 0; + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s %d,%d %dx%d bitmap %dx%d skip %d src_offset %d mask %d\n", + __FUNCTION__, x,y,w,h,width,height,unpack->SkipPixels, src_offset, mask); + + if (invert) { + first = h-1; + last = 0; + incr = -1; + } + else { + first = 0; + last = h-1; + incr = 1; + } + + /* Require that dest be pre-zero'd. + */ + for (row = first; row != (last+incr); row += incr) { + const GLubyte *rowsrc = _mesa_image_address2d(unpack, bitmap, + width, height, + GL_COLOR_INDEX, GL_BITMAP, + y + row, x); + + for (col = 0; col < w; col++, bit++) { + if (test_bit(rowsrc, (col + src_offset) ^ mask)) { + set_bit(dest, bit ^ 7); + count++; + } + } + + if (row_align) + bit = ALIGN(bit, row_align); + } + + return count; +} + + + + +/* + * Render a bitmap. + */ +static GLboolean +do_blit_bitmap( GLcontext *ctx, + GLint dstx, GLint dsty, + GLsizei width, GLsizei height, + const struct gl_pixelstore_attrib *unpack, + const GLubyte *bitmap ) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *dst = intel_drawbuf_region(intel); + GLfloat tmpColor[4]; + GLubyte ubcolor[4]; + GLuint color8888, color565; + + if (!dst) + return GL_FALSE; + + if (unpack->BufferObj->Name) { + bitmap = map_pbo(ctx, width, height, unpack, bitmap); + if (bitmap == NULL) + return GL_TRUE; /* even though this is an error, we're done */ + } + + COPY_4V(tmpColor, ctx->Current.RasterColor); + + if (NEED_SECONDARY_COLOR(ctx)) { + ADD_3V(tmpColor, tmpColor, ctx->Current.RasterSecondaryColor); + } + + UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[0], tmpColor[0]); + UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[1], tmpColor[1]); + UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[2], tmpColor[2]); + UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[3], tmpColor[3]); + + color8888 = INTEL_PACKCOLOR8888(ubcolor[0], ubcolor[1], ubcolor[2], ubcolor[3]); + color565 = INTEL_PACKCOLOR565(ubcolor[0], ubcolor[1], ubcolor[2]); + + /* Does zoom apply to bitmaps? + */ + if (!intel_check_blit_fragment_ops(ctx, tmpColor[3] == 1.0F) || + ctx->Pixel.ZoomX != 1.0F || + ctx->Pixel.ZoomY != 1.0F) + return GL_FALSE; + + LOCK_HARDWARE(intel); + + if (intel->driDrawable->numClipRects) { + __DRIdrawablePrivate *dPriv = intel->driDrawable; + drm_clip_rect_t *box = dPriv->pClipRects; + drm_clip_rect_t dest_rect; + GLint nbox = dPriv->numClipRects; + GLint srcx = 0, srcy = 0; + GLint orig_screen_x1, orig_screen_y2; + GLuint i; + + + orig_screen_x1 = dPriv->x + dstx; + orig_screen_y2 = dPriv->y + (dPriv->h - dsty); + + /* Do scissoring in GL coordinates: + */ + if (ctx->Scissor.Enabled) + { + GLint x = ctx->Scissor.X; + GLint y = ctx->Scissor.Y; + GLuint w = ctx->Scissor.Width; + GLuint h = ctx->Scissor.Height; + + if (!_mesa_clip_to_region(x, y, x+w-1, y+h-1, &dstx, &dsty, &width, &height)) + goto out; + } + + /* Convert from GL to hardware coordinates: + */ + dsty = dPriv->y + (dPriv->h - dsty - height); + dstx = dPriv->x + dstx; + + dest_rect.x1 = dstx < 0 ? 0 : dstx; + dest_rect.y1 = dsty < 0 ? 0 : dsty; + dest_rect.x2 = dstx + width < 0 ? 0 : dstx + width; + dest_rect.y2 = dsty + height < 0 ? 0 : dsty + height; + + for (i = 0; i < nbox; i++) { + drm_clip_rect_t rect; + int box_w, box_h; + GLint px, py; + GLuint stipple[32]; + + if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i])) + continue; + + /* Now go back to GL coordinates to figure out what subset of + * the bitmap we are uploading for this cliprect: + */ + box_w = rect.x2 - rect.x1; + box_h = rect.y2 - rect.y1; + srcx = rect.x1 - orig_screen_x1; + srcy = orig_screen_y2 - rect.y2; + + +#define DY 32 +#define DX 32 + + /* Then, finally, chop it all into chunks that can be + * digested by hardware: + */ + for (py = 0; py < box_h; py += DY) { + for (px = 0; px < box_w; px += DX) { + int h = MIN2(DY, box_h - py); + int w = MIN2(DX, box_w - px); + GLuint sz = ALIGN(ALIGN(w,8) * h, 64)/8; + GLenum logic_op = ctx->Color.ColorLogicOpEnabled ? + ctx->Color.LogicOp : GL_COPY; + + assert(sz <= sizeof(stipple)); + memset(stipple, 0, sz); + + /* May need to adjust this when padding has been introduced in + * sz above: + */ + if (get_bitmap_rect(width, height, unpack, + bitmap, + srcx + px, srcy + py, w, h, + (GLubyte *)stipple, + 8, + GL_TRUE) == 0) + continue; + + /* + */ + intelEmitImmediateColorExpandBlit( intel, + dst->cpp, + (GLubyte *)stipple, + sz, + (dst->cpp == 2) ? color565 : color8888, + dst->pitch, + dst->buffer, + 0, + dst->tiling, + rect.x1 + px, + rect.y2 - (py + h), + w, h, + logic_op); + } + } + } + } +out: + UNLOCK_HARDWARE(intel); + + + if (unpack->BufferObj->Name) { + /* done with PBO so unmap it now */ + ctx->Driver.UnmapBuffer(ctx, GL_PIXEL_UNPACK_BUFFER_EXT, + unpack->BufferObj); + } + + return GL_TRUE; +} + + + + + +/* There are a large number of possible ways to implement bitmap on + * this hardware, most of them have some sort of drawback. Here are a + * few that spring to mind: + * + * Blit: + * - XY_MONO_SRC_BLT_CMD + * - use XY_SETUP_CLIP_BLT for cliprect clipping. + * - XY_TEXT_BLT + * - XY_TEXT_IMMEDIATE_BLT + * - blit per cliprect, subject to maximum immediate data size. + * - XY_COLOR_BLT + * - per pixel or run of pixels + * - XY_PIXEL_BLT + * - good for sparse bitmaps + * + * 3D engine: + * - Point per pixel + * - Translate bitmap to an alpha texture and render as a quad + * - Chop bitmap up into 32x32 squares and render w/polygon stipple. + */ +void +intelBitmap(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + const struct gl_pixelstore_attrib *unpack, + const GLubyte * pixels) +{ + if (do_blit_bitmap(ctx, x, y, width, height, + unpack, pixels)) + return; + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s: fallback to swrast\n", __FUNCTION__); + + _swrast_Bitmap(ctx, x, y, width, height, unpack, pixels); +} diff --git a/src/mesa/drivers/dri/intel/intel_pixel_copy.c b/src/mesa/drivers/dri/intel/intel_pixel_copy.c new file mode 100644 index 0000000000..1c72d6179d --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_pixel_copy.c @@ -0,0 +1,376 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "enums.h" +#include "image.h" +#include "state.h" +#include "mtypes.h" +#include "macros.h" +#include "swrast/swrast.h" + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_buffers.h" +#include "intel_blit.h" +#include "intel_regions.h" +#include "intel_pixel.h" + +#define FILE_DEBUG_FLAG DEBUG_PIXEL + +static struct intel_region * +copypix_src_region(struct intel_context *intel, GLenum type) +{ + switch (type) { + case GL_COLOR: + return intel_readbuf_region(intel); + case GL_DEPTH: + /* Don't think this is really possible execpt at 16bpp, when we have no stencil. + */ + if (intel->depth_region && intel->depth_region->cpp == 2) + return intel->depth_region; + case GL_STENCIL: + /* Don't think this is really possible. + */ + break; + case GL_DEPTH_STENCIL_EXT: + /* Does it matter whether it is stencil/depth or depth/stencil? + */ + return intel->depth_region; + default: + break; + } + + return NULL; +} + + +/** + * Check if any fragment operations are in effect which might effect + * glCopyPixels. Differs from intel_check_blit_fragment_ops in that + * we allow Scissor. + */ +static GLboolean +intel_check_copypixel_blit_fragment_ops(GLcontext * ctx) +{ + if (ctx->NewState) + _mesa_update_state(ctx); + + /* Could do logicop with the blitter: + */ + return !(ctx->_ImageTransferState || + ctx->Color.AlphaEnabled || + ctx->Depth.Test || + ctx->Fog.Enabled || + ctx->Stencil.Enabled || + !ctx->Color.ColorMask[0] || + !ctx->Color.ColorMask[1] || + !ctx->Color.ColorMask[2] || + !ctx->Color.ColorMask[3] || + ctx->Texture._EnabledUnits || + ctx->FragmentProgram._Enabled || + ctx->Color.BlendEnabled); +} + +#ifdef I915 +/* Doesn't work for overlapping regions. Could do a double copy or + * just fallback. + */ +static GLboolean +do_texture_copypixels(GLcontext * ctx, + GLint srcx, GLint srcy, + GLsizei width, GLsizei height, + GLint dstx, GLint dsty, GLenum type) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *dst = intel_drawbuf_region(intel); + struct intel_region *src = copypix_src_region(intel, type); + GLenum src_format; + GLenum src_type; + + DBG("%s %d,%d %dx%d --> %d,%d\n", __FUNCTION__, + srcx, srcy, width, height, dstx, dsty); + + if (!src || !dst || type != GL_COLOR) + return GL_FALSE; + + /* Can't handle overlapping regions. Don't have sufficient control + * over rasterization to pull it off in-place. Punt on these for + * now. + * + * XXX: do a copy to a temporary. + */ + if (src->buffer == dst->buffer) { + drm_clip_rect_t srcbox; + drm_clip_rect_t dstbox; + drm_clip_rect_t tmp; + + srcbox.x1 = srcx; + srcbox.y1 = srcy; + srcbox.x2 = srcx + width; + srcbox.y2 = srcy + height; + + dstbox.x1 = dstx; + dstbox.y1 = dsty; + dstbox.x2 = dstx + width * ctx->Pixel.ZoomX; + dstbox.y2 = dsty + height * ctx->Pixel.ZoomY; + + DBG("src %d,%d %d,%d\n", srcbox.x1, srcbox.y1, srcbox.x2, srcbox.y2); + DBG("dst %d,%d %d,%d (%dx%d) (%f,%f)\n", dstbox.x1, dstbox.y1, dstbox.x2, dstbox.y2, + width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY); + + if (intel_intersect_cliprects(&tmp, &srcbox, &dstbox)) { + DBG("%s: regions overlap\n", __FUNCTION__); + return GL_FALSE; + } + } + + intelFlush(&intel->ctx); + + intel->vtbl.install_meta_state(intel); + + /* Is this true? Also will need to turn depth testing on according + * to state: + */ + intel->vtbl.meta_no_stencil_write(intel); + intel->vtbl.meta_no_depth_write(intel); + + /* Set the 3d engine to draw into the destination region: + */ + intel->vtbl.meta_draw_region(intel, dst, intel->depth_region); + + intel->vtbl.meta_import_pixel_state(intel); + + if (src->cpp == 2) { + src_format = GL_RGB; + src_type = GL_UNSIGNED_SHORT_5_6_5; + } + else { + src_format = GL_BGRA; + src_type = GL_UNSIGNED_BYTE; + } + + /* Set the frontbuffer up as a large rectangular texture. + */ + if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, 0, + src->pitch, + src->height, src_format, src_type)) { + intel->vtbl.leave_meta_state(intel); + return GL_FALSE; + } + + + intel->vtbl.meta_texture_blend_replace(intel); + + LOCK_HARDWARE(intel); + + if (intel->driDrawable->numClipRects) { + __DRIdrawablePrivate *dPriv = intel->driDrawable; + + + srcy = dPriv->h - srcy - height; /* convert from gl to hardware coords */ + + srcx += dPriv->x; + srcy += dPriv->y; + + /* Clip against the source region. This is the only source + * clipping we do. XXX: Just set the texcord wrap mode to clamp + * or similar. + * + */ + if (0) { + GLint orig_x = srcx; + GLint orig_y = srcy; + + if (!_mesa_clip_to_region(0, 0, src->pitch, src->height, + &srcx, &srcy, &width, &height)) + goto out; + + dstx += srcx - orig_x; + dsty += (srcy - orig_y) * ctx->Pixel.ZoomY; + } + + /* Just use the regular cliprect mechanism... Does this need to + * even hold the lock??? + */ + intel->vtbl.meta_draw_quad(intel, + dstx, + dstx + width * ctx->Pixel.ZoomX, + dPriv->h - (dsty + height * ctx->Pixel.ZoomY), + dPriv->h - (dsty), 0, /* XXX: what z value? */ + 0x00ff00ff, + srcx, srcx + width, srcy, srcy + height); + + out: + intel->vtbl.leave_meta_state(intel); + intel_batchbuffer_emit_mi_flush(intel->batch); + } + UNLOCK_HARDWARE(intel); + + DBG("%s: success\n", __FUNCTION__); + return GL_TRUE; +} +#endif /* I915 */ + + +/** + * CopyPixels with the blitter. Don't support zooming, pixel transfer, etc. + */ +static GLboolean +do_blit_copypixels(GLcontext * ctx, + GLint srcx, GLint srcy, + GLsizei width, GLsizei height, + GLint dstx, GLint dsty, GLenum type) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *dst = intel_drawbuf_region(intel); + struct intel_region *src = copypix_src_region(intel, type); + + /* Copypixels can be more than a straight copy. Ensure all the + * extra operations are disabled: + */ + if (!intel_check_copypixel_blit_fragment_ops(ctx) || + ctx->Pixel.ZoomX != 1.0F || ctx->Pixel.ZoomY != 1.0F) + return GL_FALSE; + + if (!src || !dst) + return GL_FALSE; + + + + intelFlush(&intel->ctx); + + LOCK_HARDWARE(intel); + + if (intel->driDrawable->numClipRects) { + __DRIdrawablePrivate *dPriv = intel->driDrawable; + __DRIdrawablePrivate *dReadPriv = intel->driReadDrawable; + drm_clip_rect_t *box = dPriv->pClipRects; + GLint nbox = dPriv->numClipRects; + GLint delta_x = 0; + GLint delta_y = 0; + GLuint i; + + /* Do scissoring in GL coordinates: + */ + if (ctx->Scissor.Enabled) + { + GLint x = ctx->Scissor.X; + GLint y = ctx->Scissor.Y; + GLuint w = ctx->Scissor.Width; + GLuint h = ctx->Scissor.Height; + GLint dx = dstx - srcx; + GLint dy = dsty - srcy; + + if (!_mesa_clip_to_region(x, y, x+w-1, y+h-1, &dstx, &dsty, &width, &height)) + goto out; + + srcx = dstx - dx; + srcy = dsty - dy; + } + + /* Convert from GL to hardware coordinates: + */ + dsty = dPriv->h - dsty - height; + srcy = dPriv->h - srcy - height; + dstx += dPriv->x; + dsty += dPriv->y; + srcx += dReadPriv->x; + srcy += dReadPriv->y; + + /* Clip against the source region. This is the only source + * clipping we do. Dst is clipped with cliprects below. + */ + { + delta_x = srcx - dstx; + delta_y = srcy - dsty; + + if (!_mesa_clip_to_region(0, 0, src->pitch, src->height, + &srcx, &srcy, &width, &height)) + goto out; + + dstx = srcx - delta_x; + dsty = srcy - delta_y; + } + + /* Could do slightly more clipping: Eg, take the intersection of + * the existing set of cliprects and those cliprects translated + * by delta_x, delta_y: + * + * This code will not overwrite other windows, but will + * introduce garbage when copying from obscured window regions. + */ + for (i = 0; i < nbox; i++) { + GLint clip_x = dstx; + GLint clip_y = dsty; + GLint clip_w = width; + GLint clip_h = height; + + if (!_mesa_clip_to_region(box[i].x1, box[i].y1, box[i].x2, box[i].y2, + &clip_x, &clip_y, &clip_w, &clip_h)) + continue; + + intelEmitCopyBlit(intel, dst->cpp, + src->pitch, src->buffer, 0, src->tiling, + dst->pitch, dst->buffer, 0, dst->tiling, + clip_x + delta_x, clip_y + delta_y, /* srcx, srcy */ + clip_x, clip_y, /* dstx, dsty */ + clip_w, clip_h, + ctx->Color.ColorLogicOpEnabled ? + ctx->Color.LogicOp : GL_COPY); + } + } +out: + UNLOCK_HARDWARE(intel); + + DBG("%s: success\n", __FUNCTION__); + return GL_TRUE; +} + + +void +intelCopyPixels(GLcontext * ctx, + GLint srcx, GLint srcy, + GLsizei width, GLsizei height, + GLint destx, GLint desty, GLenum type) +{ + if (INTEL_DEBUG & DEBUG_PIXEL) + fprintf(stderr, "%s\n", __FUNCTION__); + + if (do_blit_copypixels(ctx, srcx, srcy, width, height, destx, desty, type)) + return; + +#ifdef I915 + if (do_texture_copypixels(ctx, srcx, srcy, width, height, destx, desty, type)) + return; +#endif + + DBG("fallback to _swrast_CopyPixels\n"); + + _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type); +} diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c new file mode 100644 index 0000000000..47460da9aa --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c @@ -0,0 +1,379 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portionsalloc + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "enums.h" +#include "image.h" +#include "mtypes.h" +#include "macros.h" +#include "bufferobj.h" +#include "swrast/swrast.h" + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_blit.h" +#include "intel_buffers.h" +#include "intel_regions.h" +#include "intel_pixel.h" +#include "intel_buffer_objects.h" +#include "intel_tris.h" + + + +static GLboolean +do_texture_drawpixels(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid * pixels) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *dst = intel_drawbuf_region(intel); + struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj); + GLuint rowLength = unpack->RowLength ? unpack->RowLength : width; + GLuint src_offset; + + if (INTEL_DEBUG & DEBUG_PIXEL) + fprintf(stderr, "%s\n", __FUNCTION__); + + intelFlush(&intel->ctx); + + if (!dst) + return GL_FALSE; + + intel->vtbl.render_start(intel); + intel->vtbl.emit_state(intel); + + if (src) { + if (!_mesa_validate_pbo_access(2, unpack, width, height, 1, + format, type, pixels)) { + _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels"); + return GL_TRUE; + } + } + else { + /* PBO only for now: + */ + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - not PBO\n", __FUNCTION__); + return GL_FALSE; + } + + /* There are a couple of things we can't do yet, one of which is + * set the correct state for pixel operations when GL texturing is + * enabled. That's a pretty rare state and probably not worth the + * effort. A completely device-independent version of this may do + * more. + * + * Similarly, we make no attempt to merge metaops processing with + * an enabled fragment program, though it would certainly be + * possible. + */ + if (!intel_check_meta_tex_fragment_ops(ctx)) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad GL fragment state for metaops texture\n", + __FUNCTION__); + return GL_FALSE; + } + + intel->vtbl.install_meta_state(intel); + + + /* Is this true? Also will need to turn depth testing on according + * to state: + */ + intel->vtbl.meta_no_stencil_write(intel); + intel->vtbl.meta_no_depth_write(intel); + + /* Set the 3d engine to draw into the destination region: + */ + intel->vtbl.meta_draw_region(intel, dst, intel->depth_region); + + intel->vtbl.meta_import_pixel_state(intel); + + src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height, + format, type, 0, 0, 0); + + + /* Setup the pbo up as a rectangular texture, if possible. + * + * TODO: This is almost always possible if the i915 fragment + * program is adjusted to correctly swizzle the sampled colors. + * The major exception is any 24bit texture, like RGB888, for which + * there is no hardware support. + */ + if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, src_offset, + rowLength, height, format, type)) { + intel->vtbl.leave_meta_state(intel); + return GL_FALSE; + } + + intel->vtbl.meta_texture_blend_replace(intel); + + + LOCK_HARDWARE(intel); + + if (intel->driDrawable->numClipRects) { + __DRIdrawablePrivate *dPriv = intel->driDrawable; + GLint srcx, srcy; + GLint dstx, dsty; + + dstx = x; + dsty = dPriv->h - (y + height); + + srcx = 0; /* skiprows/pixels already done */ + srcy = 0; + + if (0) { + const GLint orig_x = dstx; + const GLint orig_y = dsty; + + if (!_mesa_clip_to_region(0, 0, dst->pitch, dst->height, + &dstx, &dsty, &width, &height)) + goto out; + + srcx += dstx - orig_x; + srcy += dsty - orig_y; + } + + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("draw %d,%d %dx%d\n", dstx, dsty, width, height); + + /* Must use the regular cliprect mechanism in order to get the + * drawing origin set correctly. Otherwise scissor state is in + * incorrect coordinate space. Does this even need to hold the + * lock??? + */ + intel->vtbl.meta_draw_quad(intel, + dstx, dstx + width * ctx->Pixel.ZoomX, + dPriv->h - (y + height * ctx->Pixel.ZoomY), + dPriv->h - (y), + -ctx->Current.RasterPos[2] * .5, + 0x00ff00ff, + srcx, srcx + width, srcy + height, srcy); + out: + intel->vtbl.leave_meta_state(intel); + intel_batchbuffer_emit_mi_flush(intel->batch); + } + UNLOCK_HARDWARE(intel); + return GL_TRUE; +} + + + + + +/* Pros: + * - no waiting for idle before updating framebuffer. + * + * Cons: + * - if upload is by memcpy, this may actually be slower than fallback path. + * - uploads the whole image even if destination is clipped + * + * Need to benchmark. + * + * Given the questions about performance, implement for pbo's only. + * This path is definitely a win if the pbo is already in agp. If it + * turns out otherwise, we can add the code necessary to upload client + * data to agp space before performing the blit. (Though it may turn + * out to be better/simpler just to use the texture engine). + */ +static GLboolean +do_blit_drawpixels(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid * pixels) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_region *dest = intel_drawbuf_region(intel); + struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj); + GLuint src_offset; + GLuint rowLength; + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s\n", __FUNCTION__); + + + if (!dest) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - no dest\n", __FUNCTION__); + return GL_FALSE; + } + + if (src) { + /* This validation should be done by core mesa: + */ + if (!_mesa_validate_pbo_access(2, unpack, width, height, 1, + format, type, pixels)) { + _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels"); + return GL_TRUE; + } + } + else { + /* PBO only for now: + */ + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - not PBO\n", __FUNCTION__); + return GL_FALSE; + } + + if (!intel_check_blit_format(dest, format, type)) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad format for blit\n", __FUNCTION__); + return GL_FALSE; + } + + if (!intel_check_blit_fragment_ops(ctx, GL_FALSE)) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad GL fragment state for blitter\n", + __FUNCTION__); + return GL_FALSE; + } + + if (ctx->Pixel.ZoomX != 1.0F) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__); + return GL_FALSE; + } + + + if (unpack->RowLength > 0) + rowLength = unpack->RowLength; + else + rowLength = width; + + if (ctx->Pixel.ZoomY == -1.0F) { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__); + return GL_FALSE; /* later */ + y -= height; + } + else if (ctx->Pixel.ZoomY == 1.0F) { + rowLength = -rowLength; + } + else { + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__); + return GL_FALSE; + } + + src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height, + format, type, 0, 0, 0); + + intelFlush(&intel->ctx); + LOCK_HARDWARE(intel); + + if (intel->driDrawable->numClipRects) { + __DRIdrawablePrivate *dPriv = intel->driDrawable; + int nbox = dPriv->numClipRects; + drm_clip_rect_t *box = dPriv->pClipRects; + drm_clip_rect_t rect; + drm_clip_rect_t dest_rect; + dri_bo *src_buffer = intel_bufferobj_buffer(intel, src, INTEL_READ); + int i; + + dest_rect.x1 = dPriv->x + x; + dest_rect.y1 = dPriv->y + dPriv->h - (y + height); + dest_rect.x2 = dest_rect.x1 + width; + dest_rect.y2 = dest_rect.y1 + height; + + for (i = 0; i < nbox; i++) { + if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i])) + continue; + + intelEmitCopyBlit(intel, + dest->cpp, + rowLength, src_buffer, src_offset, GL_FALSE, + dest->pitch, dest->buffer, 0, dest->tiling, + rect.x1 - dest_rect.x1, + rect.y2 - dest_rect.y2, + rect.x1, + rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1, + ctx->Color.ColorLogicOpEnabled ? + ctx->Color.LogicOp : GL_COPY); + } + } + UNLOCK_HARDWARE(intel); + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s - DONE\n", __FUNCTION__); + + return GL_TRUE; +} + + + +void +intelDrawPixels(GLcontext * ctx, + GLint x, GLint y, + GLsizei width, GLsizei height, + GLenum format, + GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid * pixels) +{ + if (do_blit_drawpixels(ctx, x, y, width, height, format, type, + unpack, pixels)) + return; + + if (do_texture_drawpixels(ctx, x, y, width, height, format, type, + unpack, pixels)) + return; + + + if (INTEL_DEBUG & DEBUG_PIXEL) + _mesa_printf("%s: fallback to swrast\n", __FUNCTION__); + + if (ctx->FragmentProgram._Current == ctx->FragmentProgram._TexEnvProgram) { + /* + * We don't want the i915 texenv program to be applied to DrawPixels. + * This is really just a performance optimization (mesa will other- + * wise happily run the fragment program on each pixel in the image). + */ + struct gl_fragment_program *fpSave = ctx->FragmentProgram._Current; + /* can't just set current frag prog to 0 here as on buffer resize + we'll get new state checks which will segfault. Remains a hack. */ + ctx->FragmentProgram._Current = NULL; + ctx->FragmentProgram._UseTexEnvProgram = GL_FALSE; + ctx->FragmentProgram._Active = GL_FALSE; + _swrast_DrawPixels( ctx, x, y, width, height, format, type, + unpack, pixels ); + ctx->FragmentProgram._Current = fpSave; + ctx->FragmentProgram._UseTexEnvProgram = GL_TRUE; + ctx->FragmentProgram._Active = GL_TRUE; + _swrast_InvalidateState(ctx, _NEW_PROGRAM); + } + else { + _swrast_DrawPixels( ctx, x, y, width, height, format, type, + unpack, pixels ); + } +} diff --git a/src/mesa/drivers/dri/intel/intel_reg.h b/src/mesa/drivers/dri/intel/intel_reg.h new file mode 100644 index 0000000000..96af7e1a03 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_reg.h @@ -0,0 +1,208 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#define CMD_MI (0x0 << 29) +#define CMD_2D (0x2 << 29) +#define CMD_3D (0x3 << 29) + +#define MI_BATCH_BUFFER_END (CMD_MI | 0xA << 23) + +#define MI_FLUSH (CMD_MI | (4 << 23)) +#define FLUSH_MAP_CACHE (1 << 0) +#define INHIBIT_FLUSH_RENDER_CACHE (1 << 2) + +/* Stalls command execution waiting for the given events to have occurred. */ +#define MI_WAIT_FOR_EVENT (CMD_MI | (0x3 << 23)) +#define MI_WAIT_FOR_PLANE_B_FLIP (1<<6) +#define MI_WAIT_FOR_PLANE_A_FLIP (1<<2) + +/* p189 */ +#define _3DSTATE_LOAD_STATE_IMMEDIATE_1 (CMD_3D | (0x1d<<24) | (0x04<<16)) +#define I1_LOAD_S(n) (1<<(4+n)) + +/** @{ + * 915 definitions + */ +#define S0_VB_OFFSET_MASK 0xffffffc +#define S0_AUTO_CACHE_INV_DISABLE (1<<0) +/** @} */ + +/** @{ + * 830 definitions + */ +#define S0_VB_OFFSET_MASK_830 0xffffff8 +#define S0_VB_PITCH_SHIFT_830 1 +#define S0_VB_ENABLE_830 0 +/** @} */ + +#define S1_VERTEX_WIDTH_SHIFT 24 +#define S1_VERTEX_WIDTH_MASK (0x3f<<24) +#define S1_VERTEX_PITCH_SHIFT 16 +#define S1_VERTEX_PITCH_MASK (0x3f<<16) + +#define TEXCOORDFMT_2D 0x0 +#define TEXCOORDFMT_3D 0x1 +#define TEXCOORDFMT_4D 0x2 +#define TEXCOORDFMT_1D 0x3 +#define TEXCOORDFMT_2D_16 0x4 +#define TEXCOORDFMT_4D_16 0x5 +#define TEXCOORDFMT_NOT_PRESENT 0xf +#define S2_TEXCOORD_FMT0_MASK 0xf +#define S2_TEXCOORD_FMT1_SHIFT 4 +#define S2_TEXCOORD_FMT(unit, type) ((type)<<(unit*4)) +#define S2_TEXCOORD_NONE (~0) +#define S2_TEX_COUNT_SHIFT_830 12 +#define S2_VERTEX_0_WIDTH_SHIFT_830 0 +#define S2_VERTEX_1_WIDTH_SHIFT_830 6 +/* S3 not interesting */ + +#define S4_POINT_WIDTH_SHIFT 23 +#define S4_POINT_WIDTH_MASK (0x1ff<<23) +#define S4_LINE_WIDTH_SHIFT 19 +#define S4_LINE_WIDTH_ONE (0x2<<19) +#define S4_LINE_WIDTH_MASK (0xf<<19) +#define S4_FLATSHADE_ALPHA (1<<18) +#define S4_FLATSHADE_FOG (1<<17) +#define S4_FLATSHADE_SPECULAR (1<<16) +#define S4_FLATSHADE_COLOR (1<<15) +#define S4_CULLMODE_BOTH (0<<13) +#define S4_CULLMODE_NONE (1<<13) +#define S4_CULLMODE_CW (2<<13) +#define S4_CULLMODE_CCW (3<<13) +#define S4_CULLMODE_MASK (3<<13) +#define S4_VFMT_POINT_WIDTH (1<<12) +#define S4_VFMT_SPEC_FOG (1<<11) +#define S4_VFMT_COLOR (1<<10) +#define S4_VFMT_DEPTH_OFFSET (1<<9) +#define S4_VFMT_XYZ (1<<6) +#define S4_VFMT_XYZW (2<<6) +#define S4_VFMT_XY (3<<6) +#define S4_VFMT_XYW (4<<6) +#define S4_VFMT_XYZW_MASK (7<<6) +#define S4_FORCE_DEFAULT_DIFFUSE (1<<5) +#define S4_FORCE_DEFAULT_SPECULAR (1<<4) +#define S4_LOCAL_DEPTH_OFFSET_ENABLE (1<<3) +#define S4_VFMT_FOG_PARAM (1<<2) +#define S4_SPRITE_POINT_ENABLE (1<<1) +#define S4_LINE_ANTIALIAS_ENABLE (1<<0) + +#define S4_VFMT_MASK (S4_VFMT_POINT_WIDTH | \ + S4_VFMT_SPEC_FOG | \ + S4_VFMT_COLOR | \ + S4_VFMT_DEPTH_OFFSET | \ + S4_VFMT_XYZW_MASK | \ + S4_VFMT_FOG_PARAM) + + +#define S5_WRITEDISABLE_ALPHA (1<<31) +#define S5_WRITEDISABLE_RED (1<<30) +#define S5_WRITEDISABLE_GREEN (1<<29) +#define S5_WRITEDISABLE_BLUE (1<<28) +#define S5_WRITEDISABLE_MASK (0xf<<28) +#define S5_FORCE_DEFAULT_POINT_SIZE (1<<27) +#define S5_LAST_PIXEL_ENABLE (1<<26) +#define S5_GLOBAL_DEPTH_OFFSET_ENABLE (1<<25) +#define S5_FOG_ENABLE (1<<24) +#define S5_STENCIL_REF_SHIFT 16 +#define S5_STENCIL_REF_MASK (0xff<<16) +#define S5_STENCIL_TEST_FUNC_SHIFT 13 +#define S5_STENCIL_TEST_FUNC_MASK (0x7<<13) +#define S5_STENCIL_FAIL_SHIFT 10 +#define S5_STENCIL_FAIL_MASK (0x7<<10) +#define S5_STENCIL_PASS_Z_FAIL_SHIFT 7 +#define S5_STENCIL_PASS_Z_FAIL_MASK (0x7<<7) +#define S5_STENCIL_PASS_Z_PASS_SHIFT 4 +#define S5_STENCIL_PASS_Z_PASS_MASK (0x7<<4) +#define S5_STENCIL_WRITE_ENABLE (1<<3) +#define S5_STENCIL_TEST_ENABLE (1<<2) +#define S5_COLOR_DITHER_ENABLE (1<<1) +#define S5_LOGICOP_ENABLE (1<<0) + + +#define S6_ALPHA_TEST_ENABLE (1<<31) +#define S6_ALPHA_TEST_FUNC_SHIFT 28 +#define S6_ALPHA_TEST_FUNC_MASK (0x7<<28) +#define S6_ALPHA_REF_SHIFT 20 +#define S6_ALPHA_REF_MASK (0xff<<20) +#define S6_DEPTH_TEST_ENABLE (1<<19) +#define S6_DEPTH_TEST_FUNC_SHIFT 16 +#define S6_DEPTH_TEST_FUNC_MASK (0x7<<16) +#define S6_CBUF_BLEND_ENABLE (1<<15) +#define S6_CBUF_BLEND_FUNC_SHIFT 12 +#define S6_CBUF_BLEND_FUNC_MASK (0x7<<12) +#define S6_CBUF_SRC_BLEND_FACT_SHIFT 8 +#define S6_CBUF_SRC_BLEND_FACT_MASK (0xf<<8) +#define S6_CBUF_DST_BLEND_FACT_SHIFT 4 +#define S6_CBUF_DST_BLEND_FACT_MASK (0xf<<4) +#define S6_DEPTH_WRITE_ENABLE (1<<3) +#define S6_COLOR_WRITE_ENABLE (1<<2) +#define S6_TRISTRIP_PV_SHIFT 0 +#define S6_TRISTRIP_PV_MASK (0x3<<0) + +#define S7_DEPTH_OFFSET_CONST_MASK ~0 + +/* Primitive dispatch on 830-945 */ +#define _3DPRIMITIVE (CMD_3D | (0x1f << 24)) +#define PRIM_INDIRECT (1<<23) +#define PRIM_INLINE (0<<23) +#define PRIM_INDIRECT_SEQUENTIAL (0<<17) +#define PRIM_INDIRECT_ELTS (1<<17) + +#define PRIM3D_TRILIST (0x0<<18) +#define PRIM3D_TRISTRIP (0x1<<18) +#define PRIM3D_TRISTRIP_RVRSE (0x2<<18) +#define PRIM3D_TRIFAN (0x3<<18) +#define PRIM3D_POLY (0x4<<18) +#define PRIM3D_LINELIST (0x5<<18) +#define PRIM3D_LINESTRIP (0x6<<18) +#define PRIM3D_RECTLIST (0x7<<18) +#define PRIM3D_POINTLIST (0x8<<18) +#define PRIM3D_DIB (0x9<<18) +#define PRIM3D_MASK (0x1f<<18) + +#define XY_SETUP_BLT_CMD (CMD_2D | (0x01 << 22) | 6) + +#define XY_COLOR_BLT_CMD (CMD_2D | (0x50 << 22) | 4) + +#define XY_SRC_COPY_BLT_CMD (CMD_2D | (0x53 << 22) | 6) + +#define XY_TEXT_IMMEDIATE_BLIT_CMD (CMD_2D | (0x31 << 22)) +# define XY_TEXT_BYTE_PACKED (1 << 16) + +/* BR00 */ +#define XY_BLT_WRITE_ALPHA (1 << 21) +#define XY_BLT_WRITE_RGB (1 << 20) +#define XY_SRC_TILED (1 << 15) +#define XY_DST_TILED (1 << 11) + +/* BR13 */ +#define BR13_565 (0x1 << 24) +#define BR13_8888 (0x3 << 24) + +#define FENCE_LINEAR 0 +#define FENCE_XMAJOR 1 +#define FENCE_YMAJOR 2 diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c new file mode 100644 index 0000000000..1cfc8ddd64 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_regions.c @@ -0,0 +1,564 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Provide additional functionality on top of bufmgr buffers: + * - 2d semantics and blit operations + * - refcounting of buffers for multiple images in a buffer. + * - refcounting of buffer mappings. + * - some logic for moving the buffers to the best memory pools for + * given operations. + * + * Most of this is to make it easier to implement the fixed-layout + * mipmap tree required by intel hardware in the face of GL's + * programming interface where each image can be specifed in random + * order and it isn't clear what layout the tree should have until the + * last moment. + */ + +#include <sys/ioctl.h> +#include <errno.h> + +#include "intel_context.h" +#include "intel_regions.h" +#include "intel_blit.h" +#include "intel_buffer_objects.h" +#include "intel_bufmgr.h" +#include "intel_batchbuffer.h" +#include "intel_chipset.h" + +#define FILE_DEBUG_FLAG DEBUG_REGION + +/* XXX: Thread safety? + */ +GLubyte * +intel_region_map(struct intel_context *intel, struct intel_region *region) +{ + DBG("%s\n", __FUNCTION__); + if (!region->map_refcount++) { + if (region->pbo) + intel_region_cow(intel, region); + + dri_bo_map(region->buffer, GL_TRUE); + region->map = region->buffer->virtual; + } + + return region->map; +} + +void +intel_region_unmap(struct intel_context *intel, struct intel_region *region) +{ + DBG("%s\n", __FUNCTION__); + if (!--region->map_refcount) { + dri_bo_unmap(region->buffer); + region->map = NULL; + } +} + +static int +intel_set_region_tiling_gem(struct intel_context *intel, + struct intel_region *region, + uint32_t bo_handle) +{ + struct drm_i915_gem_get_tiling get_tiling; + int ret; + + memset(&get_tiling, 0, sizeof(get_tiling)); + + get_tiling.handle = bo_handle; + ret = ioctl(intel->driFd, DRM_IOCTL_I915_GEM_GET_TILING, &get_tiling); + if (ret != 0) { + fprintf(stderr, "Failed to get tiling state for region: %s\n", + strerror(errno)); + return ret; + } + + region->tiling = get_tiling.tiling_mode; + region->bit_6_swizzle = get_tiling.swizzle_mode; + + return 0; +} + +static struct intel_region * +intel_region_alloc_internal(struct intel_context *intel, + GLuint cpp, GLuint pitch, GLuint height, + dri_bo *buffer) +{ + struct intel_region *region; + + DBG("%s\n", __FUNCTION__); + + if (buffer == NULL) + return NULL; + + region = calloc(sizeof(*region), 1); + region->cpp = cpp; + region->pitch = pitch; + region->height = height; /* needed? */ + region->refcount = 1; + region->buffer = buffer; + + /* Default to no tiling */ + region->tiling = I915_TILING_NONE; + region->bit_6_swizzle = I915_BIT_6_SWIZZLE_NONE; + + return region; +} + +struct intel_region * +intel_region_alloc(struct intel_context *intel, + GLuint cpp, GLuint pitch, GLuint height) +{ + dri_bo *buffer; + + buffer = dri_bo_alloc(intel->bufmgr, "region", + pitch * cpp * height, 64); + + return intel_region_alloc_internal(intel, cpp, pitch, height, buffer); +} + +struct intel_region * +intel_region_alloc_for_handle(struct intel_context *intel, + GLuint cpp, GLuint pitch, GLuint height, + GLuint handle, const char *name) +{ + struct intel_region *region; + dri_bo *buffer; + + buffer = intel_bo_gem_create_from_name(intel->bufmgr, name, handle); + + region = intel_region_alloc_internal(intel, cpp, pitch, height, buffer); + if (region == NULL) + return region; + + intel_set_region_tiling_gem(intel, region, handle); + + return region; +} + +void +intel_region_reference(struct intel_region **dst, struct intel_region *src) +{ + if (src) + DBG("%s %d\n", __FUNCTION__, src->refcount); + + assert(*dst == NULL); + if (src) { + src->refcount++; + *dst = src; + } +} + +void +intel_region_release(struct intel_region **region_handle) +{ + struct intel_region *region = *region_handle; + + if (region == NULL) + return; + + DBG("%s %d\n", __FUNCTION__, region->refcount - 1); + + ASSERT(region->refcount > 0); + region->refcount--; + + if (region->refcount == 0) { + assert(region->map_refcount == 0); + + if (region->pbo) + region->pbo->region = NULL; + region->pbo = NULL; + dri_bo_unreference(region->buffer); + + if (region->classic_map != NULL) { + drmUnmap(region->classic_map, + region->pitch * region->cpp * region->height); + } + + free(region); + } + *region_handle = NULL; +} + +/* + * XXX Move this into core Mesa? + */ +void +_mesa_copy_rect(GLubyte * dst, + GLuint cpp, + GLuint dst_pitch, + GLuint dst_x, + GLuint dst_y, + GLuint width, + GLuint height, + const GLubyte * src, + GLuint src_pitch, GLuint src_x, GLuint src_y) +{ + GLuint i; + + dst_pitch *= cpp; + src_pitch *= cpp; + dst += dst_x * cpp; + src += src_x * cpp; + dst += dst_y * dst_pitch; + src += src_y * dst_pitch; + width *= cpp; + + if (width == dst_pitch && width == src_pitch) + memcpy(dst, src, height * width); + else { + for (i = 0; i < height; i++) { + memcpy(dst, src, width); + dst += dst_pitch; + src += src_pitch; + } + } +} + + +/* Upload data to a rectangular sub-region. Lots of choices how to do this: + * + * - memcpy by span to current destination + * - upload data as new buffer and blit + * + * Currently always memcpy. + */ +void +intel_region_data(struct intel_context *intel, + struct intel_region *dst, + GLuint dst_offset, + GLuint dstx, GLuint dsty, + const void *src, GLuint src_pitch, + GLuint srcx, GLuint srcy, GLuint width, GLuint height) +{ + GLboolean locked = GL_FALSE; + + DBG("%s\n", __FUNCTION__); + + if (intel == NULL) + return; + + if (dst->pbo) { + if (dstx == 0 && + dsty == 0 && width == dst->pitch && height == dst->height) + intel_region_release_pbo(intel, dst); + else + intel_region_cow(intel, dst); + } + + if (!intel->locked) { + LOCK_HARDWARE(intel); + locked = GL_TRUE; + } + + _mesa_copy_rect(intel_region_map(intel, dst) + dst_offset, + dst->cpp, + dst->pitch, + dstx, dsty, width, height, src, src_pitch, srcx, srcy); + + intel_region_unmap(intel, dst); + + if (locked) + UNLOCK_HARDWARE(intel); + +} + +/* Copy rectangular sub-regions. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +void +intel_region_copy(struct intel_context *intel, + struct intel_region *dst, + GLuint dst_offset, + GLuint dstx, GLuint dsty, + struct intel_region *src, + GLuint src_offset, + GLuint srcx, GLuint srcy, GLuint width, GLuint height) +{ + DBG("%s\n", __FUNCTION__); + + if (intel == NULL) + return; + + if (dst->pbo) { + if (dstx == 0 && + dsty == 0 && width == dst->pitch && height == dst->height) + intel_region_release_pbo(intel, dst); + else + intel_region_cow(intel, dst); + } + + assert(src->cpp == dst->cpp); + + intelEmitCopyBlit(intel, + dst->cpp, + src->pitch, src->buffer, src_offset, src->tiling, + dst->pitch, dst->buffer, dst_offset, dst->tiling, + srcx, srcy, dstx, dsty, width, height, + GL_COPY); +} + +/* Fill a rectangular sub-region. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +void +intel_region_fill(struct intel_context *intel, + struct intel_region *dst, + GLuint dst_offset, + GLuint dstx, GLuint dsty, + GLuint width, GLuint height, GLuint color) +{ + DBG("%s\n", __FUNCTION__); + + if (intel == NULL) + return; + + if (dst->pbo) { + if (dstx == 0 && + dsty == 0 && width == dst->pitch && height == dst->height) + intel_region_release_pbo(intel, dst); + else + intel_region_cow(intel, dst); + } + + intelEmitFillBlit(intel, + dst->cpp, + dst->pitch, dst->buffer, dst_offset, dst->tiling, + dstx, dsty, width, height, color); +} + +/* Attach to a pbo, discarding our data. Effectively zero-copy upload + * the pbo's data. + */ +void +intel_region_attach_pbo(struct intel_context *intel, + struct intel_region *region, + struct intel_buffer_object *pbo) +{ + if (region->pbo == pbo) + return; + + /* If there is already a pbo attached, break the cow tie now. + * Don't call intel_region_release_pbo() as that would + * unnecessarily allocate a new buffer we would have to immediately + * discard. + */ + if (region->pbo) { + region->pbo->region = NULL; + region->pbo = NULL; + } + + if (region->buffer) { + dri_bo_unreference(region->buffer); + region->buffer = NULL; + } + + region->pbo = pbo; + region->pbo->region = region; + dri_bo_reference(pbo->buffer); + region->buffer = pbo->buffer; +} + + +/* Break the COW tie to the pbo and allocate a new buffer. + * The pbo gets to keep the data. + */ +void +intel_region_release_pbo(struct intel_context *intel, + struct intel_region *region) +{ + assert(region->buffer == region->pbo->buffer); + region->pbo->region = NULL; + region->pbo = NULL; + dri_bo_unreference(region->buffer); + region->buffer = NULL; + + region->buffer = dri_bo_alloc(intel->bufmgr, "region", + region->pitch * region->cpp * region->height, + 64); +} + +/* Break the COW tie to the pbo. Both the pbo and the region end up + * with a copy of the data. + */ +void +intel_region_cow(struct intel_context *intel, struct intel_region *region) +{ + struct intel_buffer_object *pbo = region->pbo; + GLboolean was_locked = intel->locked; + + if (intel == NULL) + return; + + intel_region_release_pbo(intel, region); + + assert(region->cpp * region->pitch * region->height == pbo->Base.Size); + + DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size); + + /* Now blit from the texture buffer to the new buffer: + */ + + was_locked = intel->locked; + if (!was_locked) + LOCK_HARDWARE(intel); + + intelEmitCopyBlit(intel, + region->cpp, + region->pitch, region->buffer, 0, region->tiling, + region->pitch, pbo->buffer, 0, region->tiling, + 0, 0, 0, 0, + region->pitch, region->height, + GL_COPY); + + if (!was_locked) + UNLOCK_HARDWARE(intel); +} + +dri_bo * +intel_region_buffer(struct intel_context *intel, + struct intel_region *region, GLuint flag) +{ + if (region->pbo) { + if (flag == INTEL_WRITE_PART) + intel_region_cow(intel, region); + else if (flag == INTEL_WRITE_FULL) + intel_region_release_pbo(intel, region); + } + + return region->buffer; +} + +static struct intel_region * +intel_recreate_static(struct intel_context *intel, + const char *name, + struct intel_region *region, + intelRegion *region_desc) +{ + intelScreenPrivate *intelScreen = intel->intelScreen; + int ret; + + if (region == NULL) { + region = calloc(sizeof(*region), 1); + region->refcount = 1; + } + + if (intel->ctx.Visual.rgbBits == 24) + region->cpp = 4; + else + region->cpp = intel->ctx.Visual.rgbBits / 8; + region->pitch = intelScreen->pitch; + region->height = intelScreen->height; /* needed? */ + + if (intel->ttm) { + assert(region_desc->bo_handle != -1); + region->buffer = intel_bo_gem_create_from_name(intel->bufmgr, + name, + region_desc->bo_handle); + + intel_set_region_tiling_gem(intel, region, region_desc->bo_handle); + } else { + ret = drmMap(intel->driFd, region_desc->handle, + region->pitch * region->cpp * region->height, + ®ion->classic_map); + if (ret != 0) { + fprintf(stderr, "Failed to drmMap %s buffer\n", name); + free(region); + return NULL; + } + + region->buffer = intel_bo_fake_alloc_static(intel->bufmgr, + name, + region_desc->offset, + region->pitch * region->cpp * + region->height, + region->classic_map); + + /* The sarea just gives us a boolean for whether it's tiled or not, + * instead of which tiling mode it is. Guess. + */ + if (region_desc->tiled) { + if (IS_965(intel->intelScreen->deviceID) && + region_desc == &intelScreen->depth) + region->tiling = I915_TILING_Y; + else + region->tiling = I915_TILING_X; + } else { + region->tiling = I915_TILING_NONE; + } + + region->bit_6_swizzle = I915_BIT_6_SWIZZLE_NONE; + } + + assert(region->buffer != NULL); + + return region; +} + +/** + * Create intel_region structs to describe the static front, back, and depth + * buffers created by the xserver. + * + * Although FBO's mean we now no longer use these as render targets in + * all circumstances, they won't go away until the back and depth + * buffers become private, and the front buffer will remain even then. + * + * Note that these don't allocate video memory, just describe + * allocations alread made by the X server. + */ +void +intel_recreate_static_regions(struct intel_context *intel) +{ + intelScreenPrivate *intelScreen = intel->intelScreen; + + intel->front_region = + intel_recreate_static(intel, "front", + intel->front_region, + &intelScreen->front); + + intel->back_region = + intel_recreate_static(intel, "back", + intel->back_region, + &intelScreen->back); + +#ifdef I915 + if (intelScreen->third.handle) { + intel->third_region = + intel_recreate_static(intel, "third", + intel->third_region, + &intelScreen->third); + } +#endif /* I915 */ + + /* Still assumes front.cpp == depth.cpp. We can kill this when we move to + * private buffers. + */ + intel->depth_region = + intel_recreate_static(intel, "depth", + intel->depth_region, + &intelScreen->depth); +} diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h new file mode 100644 index 0000000000..a561de485e --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_regions.h @@ -0,0 +1,149 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTEL_REGIONS_H +#define INTEL_REGIONS_H + +/** @file intel_regions.h + * + * Structure definitions and prototypes for intel_region handling, which is + * the basic structure for rectangular collections of pixels stored in a dri_bo. + */ + +#include <xf86drm.h> + +#include "mtypes.h" +#include "intel_bufmgr.h" + +struct intel_context; +struct intel_buffer_object; + +/** + * A layer on top of the bufmgr buffers that adds a few useful things: + * + * - Refcounting for local buffer references. + * - Refcounting for buffer maps + * - Buffer dimensions - pitch and height. + * - Blitter commands for copying 2D regions between buffers. (really???) + */ +struct intel_region +{ + dri_bo *buffer; /**< buffer manager's buffer */ + GLuint refcount; /**< Reference count for region */ + GLuint cpp; /**< bytes per pixel */ + GLuint pitch; /**< in pixels */ + GLuint height; /**< in pixels */ + GLubyte *map; /**< only non-NULL when region is actually mapped */ + GLuint map_refcount; /**< Reference count for mapping */ + + GLuint draw_offset; /**< Offset of drawing address within the region */ + uint32_t tiling; /**< Which tiling mode the region is in */ + uint32_t bit_6_swizzle; /**< GEM flag for address swizzling requirement */ + drmAddress classic_map; /**< drmMap of the region when not in GEM mode */ + struct intel_buffer_object *pbo; /* zero-copy uploads */ +}; + + +/* Allocate a refcounted region. Pointers to regions should only be + * copied by calling intel_reference_region(). + */ +struct intel_region *intel_region_alloc(struct intel_context *intel, + GLuint cpp, + GLuint pitch, GLuint height); + +struct intel_region * +intel_region_alloc_for_handle(struct intel_context *intel, + GLuint cpp, GLuint pitch, GLuint height, + unsigned int handle, const char *name); + +void intel_region_reference(struct intel_region **dst, + struct intel_region *src); + +void intel_region_release(struct intel_region **ib); + +void intel_recreate_static_regions(struct intel_context *intel); + +/* Map/unmap regions. This is refcounted also: + */ +GLubyte *intel_region_map(struct intel_context *intel, + struct intel_region *ib); + +void intel_region_unmap(struct intel_context *intel, struct intel_region *ib); + + +/* Upload data to a rectangular sub-region + */ +void intel_region_data(struct intel_context *intel, + struct intel_region *dest, + GLuint dest_offset, + GLuint destx, GLuint desty, + const void *src, GLuint src_stride, + GLuint srcx, GLuint srcy, GLuint width, GLuint height); + +/* Copy rectangular sub-regions + */ +void intel_region_copy(struct intel_context *intel, + struct intel_region *dest, + GLuint dest_offset, + GLuint destx, GLuint desty, + struct intel_region *src, + GLuint src_offset, + GLuint srcx, GLuint srcy, GLuint width, GLuint height); + +/* Fill a rectangular sub-region + */ +void intel_region_fill(struct intel_context *intel, + struct intel_region *dest, + GLuint dest_offset, + GLuint destx, GLuint desty, + GLuint width, GLuint height, GLuint color); + +/* Helpers for zerocopy uploads, particularly texture image uploads: + */ +void intel_region_attach_pbo(struct intel_context *intel, + struct intel_region *region, + struct intel_buffer_object *pbo); +void intel_region_release_pbo(struct intel_context *intel, + struct intel_region *region); +void intel_region_cow(struct intel_context *intel, + struct intel_region *region); + +dri_bo *intel_region_buffer(struct intel_context *intel, + struct intel_region *region, + GLuint flag); + +void _mesa_copy_rect(GLubyte * dst, + GLuint cpp, + GLuint dst_pitch, + GLuint dst_x, + GLuint dst_y, + GLuint width, + GLuint height, + const GLubyte * src, + GLuint src_pitch, GLuint src_x, GLuint src_y); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c new file mode 100644 index 0000000000..1de1ac49ea --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_screen.c @@ -0,0 +1,753 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "context.h" +#include "framebuffer.h" +#include "matrix.h" +#include "renderbuffer.h" +#include "simple_list.h" +#include "utils.h" +#include "vblank.h" +#include "xmlpool.h" + + +#include "intel_screen.h" + +#include "intel_buffers.h" +#include "intel_tex.h" +#include "intel_span.h" +#include "intel_fbo.h" +#include "intel_chipset.h" + +#include "i915_drm.h" +#include "i830_dri.h" +#include "intel_regions.h" +#include "intel_batchbuffer.h" +#include "intel_bufmgr.h" + +PUBLIC const char __driConfigOptions[] = + DRI_CONF_BEGIN + DRI_CONF_SECTION_PERFORMANCE + DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS) + DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_ALWAYS_SYNC) + /* Options correspond to DRI_CONF_BO_REUSE_DISABLED, + * DRI_CONF_BO_REUSE_ALL + */ + DRI_CONF_OPT_BEGIN_V(bo_reuse, enum, 1, "0:1") + DRI_CONF_DESC_BEGIN(en, "Buffer object reuse") + DRI_CONF_ENUM(0, "Disable buffer object reuse") + DRI_CONF_ENUM(1, "Enable reuse of all sizes of buffer objects") + DRI_CONF_DESC_END + DRI_CONF_OPT_END + DRI_CONF_SECTION_END + DRI_CONF_SECTION_QUALITY + DRI_CONF_FORCE_S3TC_ENABLE(false) + DRI_CONF_ALLOW_LARGE_TEXTURES(2) + DRI_CONF_SECTION_END + DRI_CONF_SECTION_DEBUG + DRI_CONF_NO_RAST(false) + DRI_CONF_SECTION_END +DRI_CONF_END; + +const GLuint __driNConfigOptions = 6; + +#ifdef USE_NEW_INTERFACE +static PFNGLXCREATECONTEXTMODES create_context_modes = NULL; +#endif /*USE_NEW_INTERFACE */ + +/** + * Map all the memory regions described by the screen. + * \return GL_TRUE if success, GL_FALSE if error. + */ +GLboolean +intelMapScreenRegions(__DRIscreenPrivate * sPriv) +{ + intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private; + + if (0) + _mesa_printf("TEX 0x%08x ", intelScreen->tex.handle); + if (intelScreen->tex.size != 0) { + if (drmMap(sPriv->fd, + intelScreen->tex.handle, + intelScreen->tex.size, + (drmAddress *) & intelScreen->tex.map) != 0) { + intelUnmapScreenRegions(intelScreen); + return GL_FALSE; + } + } + + return GL_TRUE; +} + +void +intelUnmapScreenRegions(intelScreenPrivate * intelScreen) +{ + if (intelScreen->tex.map) { + drmUnmap(intelScreen->tex.map, intelScreen->tex.size); + intelScreen->tex.map = NULL; + } +} + + +static void +intelPrintDRIInfo(intelScreenPrivate * intelScreen, + __DRIscreenPrivate * sPriv, I830DRIPtr gDRIPriv) +{ + fprintf(stderr, "*** Front size: 0x%x offset: 0x%x pitch: %d\n", + intelScreen->front.size, intelScreen->front.offset, + intelScreen->pitch); + fprintf(stderr, "*** Back size: 0x%x offset: 0x%x pitch: %d\n", + intelScreen->back.size, intelScreen->back.offset, + intelScreen->pitch); + fprintf(stderr, "*** Depth size: 0x%x offset: 0x%x pitch: %d\n", + intelScreen->depth.size, intelScreen->depth.offset, + intelScreen->pitch); + fprintf(stderr, "*** Texture size: 0x%x offset: 0x%x\n", + intelScreen->tex.size, intelScreen->tex.offset); + fprintf(stderr, "*** Memory : 0x%x\n", gDRIPriv->mem); +} + + +static void +intelPrintSAREA(const struct drm_i915_sarea * sarea) +{ + fprintf(stderr, "SAREA: sarea width %d height %d\n", sarea->width, + sarea->height); + fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch); + fprintf(stderr, + "SAREA: front offset: 0x%08x size: 0x%x handle: 0x%x tiled: %d\n", + sarea->front_offset, sarea->front_size, + (unsigned) sarea->front_handle, sarea->front_tiled); + fprintf(stderr, + "SAREA: back offset: 0x%08x size: 0x%x handle: 0x%x tiled: %d\n", + sarea->back_offset, sarea->back_size, + (unsigned) sarea->back_handle, sarea->back_tiled); + fprintf(stderr, "SAREA: depth offset: 0x%08x size: 0x%x handle: 0x%x tiled: %d\n", + sarea->depth_offset, sarea->depth_size, + (unsigned) sarea->depth_handle, sarea->depth_tiled); + fprintf(stderr, "SAREA: tex offset: 0x%08x size: 0x%x handle: 0x%x\n", + sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle); +} + + +/** + * A number of the screen parameters are obtained/computed from + * information in the SAREA. This function updates those parameters. + */ +void +intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen, + struct drm_i915_sarea * sarea) +{ + intelScreen->width = sarea->width; + intelScreen->height = sarea->height; + intelScreen->pitch = sarea->pitch; + + intelScreen->front.offset = sarea->front_offset; + intelScreen->front.handle = sarea->front_handle; + intelScreen->front.size = sarea->front_size; + intelScreen->front.tiled = sarea->front_tiled; + + intelScreen->back.offset = sarea->back_offset; + intelScreen->back.handle = sarea->back_handle; + intelScreen->back.size = sarea->back_size; + intelScreen->back.tiled = sarea->back_tiled; + + if (intelScreen->driScrnPriv->ddx_version.minor >= 8) { + intelScreen->third.offset = sarea->third_offset; + intelScreen->third.handle = sarea->third_handle; + intelScreen->third.size = sarea->third_size; + intelScreen->third.tiled = sarea->third_tiled; + } + + intelScreen->depth.offset = sarea->depth_offset; + intelScreen->depth.handle = sarea->depth_handle; + intelScreen->depth.size = sarea->depth_size; + intelScreen->depth.tiled = sarea->depth_tiled; + + if (intelScreen->driScrnPriv->ddx_version.minor >= 9) { + intelScreen->front.bo_handle = sarea->front_bo_handle; + intelScreen->back.bo_handle = sarea->back_bo_handle; + intelScreen->third.bo_handle = sarea->third_bo_handle; + intelScreen->depth.bo_handle = sarea->depth_bo_handle; + } else { + intelScreen->front.bo_handle = -1; + intelScreen->back.bo_handle = -1; + intelScreen->third.bo_handle = -1; + intelScreen->depth.bo_handle = -1; + } + + intelScreen->tex.offset = sarea->tex_offset; + intelScreen->logTextureGranularity = sarea->log_tex_granularity; + intelScreen->tex.handle = sarea->tex_handle; + intelScreen->tex.size = sarea->tex_size; + + if (0) + intelPrintSAREA(sarea); +} + +static const __DRItexOffsetExtension intelTexOffsetExtension = { + { __DRI_TEX_OFFSET }, + intelSetTexOffset, +}; + +static const __DRItexBufferExtension intelTexBufferExtension = { + { __DRI_TEX_BUFFER, __DRI_TEX_BUFFER_VERSION }, + intelSetTexBuffer, +}; + +static const __DRIextension *intelScreenExtensions[] = { + &driReadDrawableExtension, + &driCopySubBufferExtension.base, + &driSwapControlExtension.base, + &driFrameTrackingExtension.base, + &driMediaStreamCounterExtension.base, + &intelTexOffsetExtension.base, + &intelTexBufferExtension.base, + NULL +}; + +static GLboolean +intel_get_param(__DRIscreenPrivate *psp, int param, int *value) +{ + int ret; + struct drm_i915_getparam gp; + + gp.param = param; + gp.value = value; + + ret = drmCommandWriteRead(psp->fd, DRM_I915_GETPARAM, &gp, sizeof(gp)); + if (ret) { + fprintf(stderr, "drm_i915_getparam: %d\n", ret); + return GL_FALSE; + } + + return GL_TRUE; +} + +static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv) +{ + intelScreenPrivate *intelScreen; + I830DRIPtr gDRIPriv = (I830DRIPtr) sPriv->pDevPriv; + struct drm_i915_sarea *sarea; + + if (sPriv->devPrivSize != sizeof(I830DRIRec)) { + fprintf(stderr, + "\nERROR! sizeof(I830DRIRec) does not match passed size from device driver\n"); + return GL_FALSE; + } + + /* Allocate the private area */ + intelScreen = (intelScreenPrivate *) CALLOC(sizeof(intelScreenPrivate)); + if (!intelScreen) { + fprintf(stderr, "\nERROR! Allocating private area failed\n"); + return GL_FALSE; + } + /* parse information in __driConfigOptions */ + driParseOptionInfo(&intelScreen->optionCache, + __driConfigOptions, __driNConfigOptions); + + intelScreen->driScrnPriv = sPriv; + sPriv->private = (void *) intelScreen; + sarea = (struct drm_i915_sarea *) + (((GLubyte *) sPriv->pSAREA) + gDRIPriv->sarea_priv_offset); + intelScreen->sarea = sarea; + + intelScreen->deviceID = gDRIPriv->deviceID; + + intelUpdateScreenFromSAREA(intelScreen, sarea); + + if (!intelMapScreenRegions(sPriv)) { + fprintf(stderr, "\nERROR! mapping regions\n"); + _mesa_free(intelScreen); + sPriv->private = NULL; + return GL_FALSE; + } + + if (0) + intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv); + + intelScreen->drmMinor = sPriv->drm_version.minor; + + /* Determine if IRQs are active? */ + if (!intel_get_param(sPriv, I915_PARAM_IRQ_ACTIVE, + &intelScreen->irq_active)) + return GL_FALSE; + + /* Determine if batchbuffers are allowed */ + if (!intel_get_param(sPriv, I915_PARAM_ALLOW_BATCHBUFFER, + &intelScreen->allow_batchbuffer)) + return GL_FALSE; + + sPriv->extensions = intelScreenExtensions; + + return GL_TRUE; +} + + +static void +intelDestroyScreen(__DRIscreenPrivate * sPriv) +{ + intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private; + + intelUnmapScreenRegions(intelScreen); + + FREE(intelScreen); + sPriv->private = NULL; +} + + +/** + * This is called when we need to set up GL rendering to a new X window. + */ +static GLboolean +intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, + __DRIdrawablePrivate * driDrawPriv, + const __GLcontextModes * mesaVis, GLboolean isPixmap) +{ + intelScreenPrivate *screen = (intelScreenPrivate *) driScrnPriv->private; + + if (isPixmap) { + return GL_FALSE; /* not implemented */ + } + else { + GLboolean swStencil = (mesaVis->stencilBits > 0 && + mesaVis->depthBits != 24); + GLenum rgbFormat = (mesaVis->redBits == 5 ? GL_RGB5 : GL_RGBA8); + + struct intel_framebuffer *intel_fb = CALLOC_STRUCT(intel_framebuffer); + + if (!intel_fb) + return GL_FALSE; + + _mesa_initialize_framebuffer(&intel_fb->Base, mesaVis); + + /* setup the hardware-based renderbuffers */ + intel_fb->color_rb[0] = intel_create_renderbuffer(rgbFormat); + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT, + &intel_fb->color_rb[0]->Base); + + if (mesaVis->doubleBufferMode) { + intel_fb->color_rb[1] = intel_create_renderbuffer(rgbFormat); + + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT, + &intel_fb->color_rb[1]->Base); + + if (screen->third.handle) { + struct gl_renderbuffer *tmp_rb = NULL; + + intel_fb->color_rb[2] = intel_create_renderbuffer(rgbFormat); + _mesa_reference_renderbuffer(&tmp_rb, &intel_fb->color_rb[2]->Base); + } + } + + if (mesaVis->depthBits == 24) { + if (mesaVis->stencilBits == 8) { + /* combined depth/stencil buffer */ + struct intel_renderbuffer *depthStencilRb + = intel_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT); + /* note: bind RB to two attachment points */ + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, + &depthStencilRb->Base); + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_STENCIL, + &depthStencilRb->Base); + } else { + struct intel_renderbuffer *depthRb + = intel_create_renderbuffer(GL_DEPTH_COMPONENT24); + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, + &depthRb->Base); + } + } + else if (mesaVis->depthBits == 16) { + /* just 16-bit depth buffer, no hw stencil */ + struct intel_renderbuffer *depthRb + = intel_create_renderbuffer(GL_DEPTH_COMPONENT16); + _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base); + } + + /* now add any/all software-based renderbuffers we may need */ + _mesa_add_soft_renderbuffers(&intel_fb->Base, + GL_FALSE, /* never sw color */ + GL_FALSE, /* never sw depth */ + swStencil, mesaVis->accumRedBits > 0, + GL_FALSE, /* never sw alpha */ + GL_FALSE /* never sw aux */ ); + driDrawPriv->driverPrivate = (void *) intel_fb; + + return GL_TRUE; + } +} + +static void +intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv) +{ + _mesa_unreference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate))); +} + + +/** + * Get information about previous buffer swaps. + */ +static int +intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo) +{ + struct intel_framebuffer *intel_fb; + + if ((dPriv == NULL) || (dPriv->driverPrivate == NULL) + || (sInfo == NULL)) { + return -1; + } + + intel_fb = dPriv->driverPrivate; + sInfo->swap_count = intel_fb->swap_count; + sInfo->swap_ust = intel_fb->swap_ust; + sInfo->swap_missed_count = intel_fb->swap_missed_count; + + sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0) + ? driCalculateSwapUsage(dPriv, 0, intel_fb->swap_missed_ust) + : 0.0; + + return 0; +} + + +/* There are probably better ways to do this, such as an + * init-designated function to register chipids and createcontext + * functions. + */ +extern GLboolean i830CreateContext(const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate); + +extern GLboolean i915CreateContext(const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate); +extern GLboolean brwCreateContext(const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate); + +static GLboolean +intelCreateContext(const __GLcontextModes * mesaVis, + __DRIcontextPrivate * driContextPriv, + void *sharedContextPrivate) +{ + __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; + intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private; + +#ifdef I915 + if (IS_9XX(intelScreen->deviceID)) { + if (!IS_965(intelScreen->deviceID)) { + return i915CreateContext(mesaVis, driContextPriv, + sharedContextPrivate); + } + } else { + return i830CreateContext(mesaVis, driContextPriv, sharedContextPrivate); + } +#else + if (IS_965(intelScreen->deviceID)) + return brwCreateContext(mesaVis, driContextPriv, sharedContextPrivate); +#endif + fprintf(stderr, "Unrecognized deviceID %x\n", intelScreen->deviceID); + return GL_FALSE; +} + + +static __DRIconfig ** +intelFillInModes(__DRIscreenPrivate *psp, + unsigned pixel_bits, unsigned depth_bits, + unsigned stencil_bits, GLboolean have_back_buffer) +{ + __DRIconfig **configs; + __GLcontextModes *m; + unsigned depth_buffer_factor; + unsigned back_buffer_factor; + GLenum fb_format; + GLenum fb_type; + int i; + + /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't + * support pageflipping at all. + */ + static const GLenum back_buffer_modes[] = { + GLX_NONE, GLX_SWAP_UNDEFINED_OML, GLX_SWAP_COPY_OML + }; + + u_int8_t depth_bits_array[3]; + u_int8_t stencil_bits_array[3]; + + depth_bits_array[0] = 0; + depth_bits_array[1] = depth_bits; + depth_bits_array[2] = depth_bits; + + /* Just like with the accumulation buffer, always provide some modes + * with a stencil buffer. It will be a sw fallback, but some apps won't + * care about that. + */ + stencil_bits_array[0] = 0; + stencil_bits_array[1] = 0; + if (depth_bits == 24) + stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits; + + stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits; + + depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1; + back_buffer_factor = (have_back_buffer) ? 3 : 1; + + if (pixel_bits == 16) { + fb_format = GL_RGB; + fb_type = GL_UNSIGNED_SHORT_5_6_5; + } + else { + fb_format = GL_BGRA; + fb_type = GL_UNSIGNED_INT_8_8_8_8_REV; + } + + configs = driCreateConfigs(fb_format, fb_type, + depth_bits_array, stencil_bits_array, + depth_buffer_factor, back_buffer_modes, + back_buffer_factor); + if (configs == NULL) { + fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__, + __LINE__); + return NULL; + } + + /* Mark the visual as slow if there are "fake" stencil bits. + */ + for (i = 0; configs[i]; i++) { + m = &configs[i]->modes; + if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) { + m->visualRating = GLX_SLOW_CONFIG; + } + } + + return configs; +} + +static GLboolean +intel_init_bufmgr(intelScreenPrivate *intelScreen) +{ + GLboolean gem_disable = getenv("INTEL_NO_GEM") != NULL; + int gem_kernel = 0; + GLboolean gem_supported; + struct drm_i915_getparam gp; + __DRIscreenPrivate *spriv = intelScreen->driScrnPriv; + + intelScreen->no_hw = getenv("INTEL_NO_HW") != NULL; + + gp.param = I915_PARAM_HAS_GEM; + gp.value = &gem_kernel; + + (void) drmCommandWriteRead(spriv->fd, DRM_I915_GETPARAM, &gp, sizeof(gp)); + + /* If we've got a new enough DDX that's initializing GEM and giving us + * object handles for the shared buffers, use that. + */ + intelScreen->ttm = GL_FALSE; + if (intelScreen->driScrnPriv->dri2.enabled) + gem_supported = GL_TRUE; + else if (intelScreen->driScrnPriv->ddx_version.minor >= 9 && + gem_kernel && + intelScreen->front.bo_handle != -1) + gem_supported = GL_TRUE; + else + gem_supported = GL_FALSE; + + if (!gem_disable && gem_supported) { + intelScreen->bufmgr = intel_bufmgr_gem_init(spriv->fd, BATCH_SZ); + if (intelScreen->bufmgr != NULL) + intelScreen->ttm = GL_TRUE; + } + /* Otherwise, use the classic buffer manager. */ + if (intelScreen->bufmgr == NULL) { + if (gem_disable) { + fprintf(stderr, "GEM disabled. Using classic.\n"); + } else { + fprintf(stderr, "Failed to initialize GEM. " + "Falling back to classic.\n"); + } + + if (intelScreen->tex.size == 0) { + fprintf(stderr, "[%s:%u] Error initializing buffer manager.\n", + __func__, __LINE__); + return GL_FALSE; + } + + intelScreen->bufmgr = + intel_bufmgr_fake_init(spriv->fd, + intelScreen->tex.offset, + intelScreen->tex.map, + intelScreen->tex.size, + (unsigned int * volatile) + &intelScreen->sarea->last_dispatch); + } + + /* XXX bufmgr should be per-screen, not per-context */ + intelScreen->ttm = intelScreen->ttm; + + return GL_TRUE; +} + +/** + * This is the driver specific part of the createNewScreen entry point. + * Called when using legacy DRI. + * + * \todo maybe fold this into intelInitDriver + * + * \return the __GLcontextModes supported by this driver + */ +static const __DRIconfig **intelInitScreen(__DRIscreenPrivate *psp) +{ + intelScreenPrivate *intelScreen; +#ifdef I915 + static const __DRIversion ddx_expected = { 1, 5, 0 }; +#else + static const __DRIversion ddx_expected = { 1, 6, 0 }; +#endif + static const __DRIversion dri_expected = { 4, 0, 0 }; + static const __DRIversion drm_expected = { 1, 5, 0 }; + I830DRIPtr dri_priv = (I830DRIPtr) psp->pDevPriv; + + if (!driCheckDriDdxDrmVersions2("i915", + &psp->dri_version, &dri_expected, + &psp->ddx_version, &ddx_expected, + &psp->drm_version, &drm_expected)) { + return NULL; + } + + /* Calling driInitExtensions here, with a NULL context pointer, + * does not actually enable the extensions. It just makes sure + * that all the dispatch offsets for all the extensions that + * *might* be enables are known. This is needed because the + * dispatch offsets need to be known when _mesa_context_create is + * called, but we can't enable the extensions until we have a + * context pointer. + * + * Hello chicken. Hello egg. How are you two today? + */ + intelInitExtensions(NULL, GL_TRUE); + + if (!intelInitDriver(psp)) + return NULL; + + psp->extensions = intelScreenExtensions; + + intelScreen = psp->private; + if (!intel_init_bufmgr(intelScreen)) + return GL_FALSE; + + return (const __DRIconfig **) + intelFillInModes(psp, dri_priv->cpp * 8, + (dri_priv->cpp == 2) ? 16 : 24, + (dri_priv->cpp == 2) ? 0 : 8, 1); +} + +struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen) +{ + /* + * This should probably change to have the screen allocate a dummy + * context at screen creation. For now just use the current context. + */ + + GET_CURRENT_CONTEXT(ctx); + if (ctx == NULL) { + _mesa_problem(NULL, "No current context in intelScreenContext\n"); + return NULL; + } + return intel_context(ctx); +} + +/** + * This is the driver specific part of the createNewScreen entry point. + * Called when using DRI2. + * + * \return the __GLcontextModes supported by this driver + */ +static const +__DRIconfig **intelInitScreen2(__DRIscreenPrivate *psp) +{ + intelScreenPrivate *intelScreen; + + /* Calling driInitExtensions here, with a NULL context pointer, + * does not actually enable the extensions. It just makes sure + * that all the dispatch offsets for all the extensions that + * *might* be enables are known. This is needed because the + * dispatch offsets need to be known when _mesa_context_create is + * called, but we can't enable the extensions until we have a + * context pointer. + * + * Hello chicken. Hello egg. How are you two today? + */ + intelInitExtensions(NULL, GL_TRUE); + + /* Allocate the private area */ + intelScreen = (intelScreenPrivate *) CALLOC(sizeof(intelScreenPrivate)); + if (!intelScreen) { + fprintf(stderr, "\nERROR! Allocating private area failed\n"); + return GL_FALSE; + } + /* parse information in __driConfigOptions */ + driParseOptionInfo(&intelScreen->optionCache, + __driConfigOptions, __driNConfigOptions); + + intelScreen->driScrnPriv = psp; + psp->private = (void *) intelScreen; + + intelScreen->drmMinor = psp->drm_version.minor; + + /* Determine chipset ID */ + if (!intel_get_param(psp, I915_PARAM_CHIPSET_ID, + &intelScreen->deviceID)) + return GL_FALSE; + + if (!intel_init_bufmgr(intelScreen)) + return GL_FALSE; + + intelScreen->irq_active = 1; + psp->extensions = intelScreenExtensions; + + return driConcatConfigs(intelFillInModes(psp, 16, 16, 0, 1), + intelFillInModes(psp, 32, 24, 8, 1)); +} + +const struct __DriverAPIRec driDriverAPI = { + .InitScreen = intelInitScreen, + .DestroyScreen = intelDestroyScreen, + .CreateContext = intelCreateContext, + .DestroyContext = intelDestroyContext, + .CreateBuffer = intelCreateBuffer, + .DestroyBuffer = intelDestroyBuffer, + .SwapBuffers = intelSwapBuffers, + .MakeCurrent = intelMakeCurrent, + .UnbindContext = intelUnbindContext, + .GetSwapInfo = intelGetSwapInfo, + .GetDrawableMSC = driDrawableGetMSC32, + .WaitForMSC = driWaitForMSC32, + .CopySubBuffer = intelCopySubBuffer, + + .InitScreen2 = intelInitScreen2, +}; diff --git a/src/mesa/drivers/dri/intel/intel_screen.h b/src/mesa/drivers/dri/intel/intel_screen.h new file mode 100644 index 0000000000..daf2ab4c58 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_screen.h @@ -0,0 +1,116 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _INTEL_INIT_H_ +#define _INTEL_INIT_H_ + +#include <sys/time.h> +#include "dri_util.h" +#include "intel_bufmgr.h" +#include "i915_drm.h" +#include "xmlconfig.h" + +/* XXX: change name or eliminate to avoid conflict with "struct + * intel_region"!!! + */ +typedef struct +{ + drm_handle_t handle; + drmSize size; /* region size in bytes */ + char *map; /* memory map */ + int offset; /* from start of video mem, in bytes */ + unsigned int bo_handle; /* buffer object id if available, or -1 */ + /** + * Flags if the region is tiled. + * + * Not included is Y versus X tiling. + */ + GLboolean tiled; +} intelRegion; + +typedef struct +{ + intelRegion front; + intelRegion back; + intelRegion third; + intelRegion depth; + intelRegion tex; + + int deviceID; + int width; + int height; + int pitch; /* common row stride, in pixels */ + + int logTextureGranularity; + + __DRIscreenPrivate *driScrnPriv; + + volatile struct drm_i915_sarea *sarea; + + int drmMinor; + + int irq_active; + int allow_batchbuffer; + + GLboolean no_hw; + + int ttm; + dri_bufmgr *bufmgr; + + /** + * Configuration cache with default values for all contexts + */ + driOptionCache optionCache; +} intelScreenPrivate; + + + +extern GLboolean intelMapScreenRegions(__DRIscreenPrivate * sPriv); + +extern void intelUnmapScreenRegions(intelScreenPrivate * intelScreen); + +extern void +intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen, + struct drm_i915_sarea * sarea); + +extern void intelDestroyContext(__DRIcontextPrivate * driContextPriv); + +extern GLboolean intelUnbindContext(__DRIcontextPrivate * driContextPriv); + +extern GLboolean +intelMakeCurrent(__DRIcontextPrivate * driContextPriv, + __DRIdrawablePrivate * driDrawPriv, + __DRIdrawablePrivate * driReadPriv); + +extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv); + +extern void +intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h); + +extern struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c new file mode 100644 index 0000000000..0ea8576a02 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_span.c @@ -0,0 +1,814 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "glheader.h" +#include "macros.h" +#include "mtypes.h" +#include "colormac.h" + +#include "intel_fbo.h" +#include "intel_screen.h" +#include "intel_span.h" +#include "intel_regions.h" +#include "intel_tex.h" + +#include "swrast/swrast.h" + +static void +intel_set_span_functions(struct intel_context *intel, + struct gl_renderbuffer *rb); + +#define SPAN_CACHE_SIZE 4096 + +static void +get_span_cache(struct intel_renderbuffer *irb, uint32_t offset) +{ + if (irb->span_cache == NULL) { + irb->span_cache = _mesa_malloc(SPAN_CACHE_SIZE); + irb->span_cache_offset = -1; + } + + if ((offset & ~(SPAN_CACHE_SIZE - 1)) != irb->span_cache_offset) { + irb->span_cache_offset = offset & ~(SPAN_CACHE_SIZE - 1); + dri_bo_get_subdata(irb->region->buffer, irb->span_cache_offset, + SPAN_CACHE_SIZE, irb->span_cache); + } +} + +static void +clear_span_cache(struct intel_renderbuffer *irb) +{ + irb->span_cache_offset = -1; +} + +static uint32_t +pread_32(struct intel_renderbuffer *irb, uint32_t offset) +{ + get_span_cache(irb, offset); + + return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))); +} + +static uint32_t +pread_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset) +{ + get_span_cache(irb, offset); + + return *(uint32_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))) | + 0xff000000; +} + +static uint16_t +pread_16(struct intel_renderbuffer *irb, uint32_t offset) +{ + get_span_cache(irb, offset); + + return *(uint16_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))); +} + +static uint8_t +pread_8(struct intel_renderbuffer *irb, uint32_t offset) +{ + get_span_cache(irb, offset); + + return *(uint8_t *)(irb->span_cache + (offset & (SPAN_CACHE_SIZE - 1))); +} + +static void +pwrite_32(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val) +{ + clear_span_cache(irb); + + dri_bo_subdata(irb->region->buffer, offset, 4, &val); +} + +static void +pwrite_xrgb8888(struct intel_renderbuffer *irb, uint32_t offset, uint32_t val) +{ + clear_span_cache(irb); + + dri_bo_subdata(irb->region->buffer, offset, 3, &val); +} + +static void +pwrite_16(struct intel_renderbuffer *irb, uint32_t offset, uint16_t val) +{ + clear_span_cache(irb); + + dri_bo_subdata(irb->region->buffer, offset, 2, &val); +} + +static void +pwrite_8(struct intel_renderbuffer *irb, uint32_t offset, uint8_t val) +{ + clear_span_cache(irb); + + dri_bo_subdata(irb->region->buffer, offset, 1, &val); +} + +static uint32_t no_tile_swizzle(struct intel_renderbuffer *irb, + struct intel_context *intel, + int x, int y) +{ + x += intel->drawX; + y += intel->drawY; + + return (y * irb->region->pitch + x) * irb->region->cpp; +} + +/* + * Deal with tiled surfaces + */ + +static uint32_t x_tile_swizzle(struct intel_renderbuffer *irb, + struct intel_context *intel, + int x, int y) +{ + int tile_stride; + int xbyte; + int x_tile_off, y_tile_off; + int x_tile_number, y_tile_number; + int tile_off, tile_base; + + tile_stride = (irb->pfPitch * irb->region->cpp) << 3; + + x += intel->drawX; + y += intel->drawY; + + xbyte = x * irb->region->cpp; + + x_tile_off = xbyte & 0x1ff; + y_tile_off = y & 7; + + x_tile_number = xbyte >> 9; + y_tile_number = y >> 3; + + tile_off = (y_tile_off << 9) + x_tile_off; + + switch (irb->region->bit_6_swizzle) { + case I915_BIT_6_SWIZZLE_NONE: + break; + case I915_BIT_6_SWIZZLE_9: + tile_off ^= ((tile_off >> 3) & 64); + break; + case I915_BIT_6_SWIZZLE_9_10: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64); + break; + case I915_BIT_6_SWIZZLE_9_11: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64); + break; + case I915_BIT_6_SWIZZLE_9_10_11: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^ + ((tile_off >> 5) & 64); + break; + default: + fprintf(stderr, "Unknown tile swizzling mode %d\n", + irb->region->bit_6_swizzle); + exit(1); + } + + tile_base = (x_tile_number << 12) + y_tile_number * tile_stride; + +#if 0 + printf("(%d,%d) -> %d + %d = %d (pitch = %d, tstride = %d)\n", + x, y, tile_off, tile_base, + tile_off + tile_base, + irb->pfPitch, tile_stride); +#endif + + return tile_base + tile_off; +} + +static uint32_t y_tile_swizzle(struct intel_renderbuffer *irb, + struct intel_context *intel, + int x, int y) +{ + int tile_stride; + int xbyte; + int x_tile_off, y_tile_off; + int x_tile_number, y_tile_number; + int tile_off, tile_base; + + tile_stride = (irb->pfPitch * irb->region->cpp) << 5; + + x += intel->drawX; + y += intel->drawY; + + xbyte = x * irb->region->cpp; + + x_tile_off = xbyte & 0x7f; + y_tile_off = y & 0x1f; + + x_tile_number = xbyte >> 7; + y_tile_number = y >> 5; + + tile_off = ((x_tile_off & ~0xf) << 5) + (y_tile_off << 4) + + (x_tile_off & 0xf); + + switch (irb->region->bit_6_swizzle) { + case I915_BIT_6_SWIZZLE_NONE: + break; + case I915_BIT_6_SWIZZLE_9: + tile_off ^= ((tile_off >> 3) & 64); + break; + case I915_BIT_6_SWIZZLE_9_10: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64); + break; + case I915_BIT_6_SWIZZLE_9_11: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 5) & 64); + break; + case I915_BIT_6_SWIZZLE_9_10_11: + tile_off ^= ((tile_off >> 3) & 64) ^ ((tile_off >> 4) & 64) ^ + ((tile_off >> 5) & 64); + break; + default: + fprintf(stderr, "Unknown tile swizzling mode %d\n", + irb->region->bit_6_swizzle); + exit(1); + } + + tile_base = (x_tile_number << 12) + y_tile_number * tile_stride; + + return tile_base + tile_off; +} + +/* + break intelWriteRGBASpan_ARGB8888 +*/ + +#undef DBG +#define DBG 0 + +#define LOCAL_VARS \ + struct intel_context *intel = intel_context(ctx); \ + struct intel_renderbuffer *irb = intel_renderbuffer(rb); \ + const GLint yScale = irb->RenderToTexture ? 1 : -1; \ + const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1; \ + GLuint p; \ + (void) p; + +/* XXX FBO: this is identical to the macro in spantmp2.h except we get + * the cliprect info from the context, not the driDrawable. + * Move this into spantmp2.h someday. + */ +#define HW_CLIPLOOP() \ + do { \ + int _nc = intel->numClipRects; \ + while ( _nc-- ) { \ + int minx = intel->pClipRects[_nc].x1 - intel->drawX; \ + int miny = intel->pClipRects[_nc].y1 - intel->drawY; \ + int maxx = intel->pClipRects[_nc].x2 - intel->drawX; \ + int maxy = intel->pClipRects[_nc].y2 - intel->drawY; + +#if 0 + }} +#endif + +#define Y_FLIP(_y) ((_y) * yScale + yBias) + +/* XXX with GEM, these need to tell the kernel */ +#define HW_LOCK() + +#define HW_UNLOCK() + +/* 16 bit, RGB565 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_RGB +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 + +#define TAG(x) intel##x##_RGB565 +#define TAG2(x,y) intel##x##_RGB565##y +#define GET_VALUE(X, Y) pread_16(irb, no_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_16(irb, no_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +/* 32 bit, ARGB8888 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel##x##_ARGB8888 +#define TAG2(x,y) intel##x##_ARGB8888##y +#define GET_VALUE(X, Y) pread_32(irb, no_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_32(irb, no_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +/* 32 bit, xRGB8888 color spanline and pixel functions + */ +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel##x##_xRGB8888 +#define TAG2(x,y) intel##x##_xRGB8888##y +#define GET_VALUE(X, Y) pread_xrgb8888(irb, no_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, no_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +/* 16 bit RGB565 color tile spanline and pixel functions + */ + +#define SPANTMP_PIXEL_FMT GL_RGB +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 + +#define TAG(x) intel_XTile_##x##_RGB565 +#define TAG2(x,y) intel_XTile_##x##_RGB565##y +#define GET_VALUE(X, Y) pread_16(irb, x_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_16(irb, x_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +#define SPANTMP_PIXEL_FMT GL_RGB +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 + +#define TAG(x) intel_YTile_##x##_RGB565 +#define TAG2(x,y) intel_YTile_##x##_RGB565##y +#define GET_VALUE(X, Y) pread_16(irb, y_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_16(irb, y_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +/* 32 bit ARGB888 color tile spanline and pixel functions + */ + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_XTile_##x##_ARGB8888 +#define TAG2(x,y) intel_XTile_##x##_ARGB8888##y +#define GET_VALUE(X, Y) pread_32(irb, x_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_32(irb, x_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_YTile_##x##_ARGB8888 +#define TAG2(x,y) intel_YTile_##x##_ARGB8888##y +#define GET_VALUE(X, Y) pread_32(irb, y_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_32(irb, y_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +/* 32 bit xRGB888 color tile spanline and pixel functions + */ + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_XTile_##x##_xRGB8888 +#define TAG2(x,y) intel_XTile_##x##_xRGB8888##y +#define GET_VALUE(X, Y) pread_xrgb8888(irb, x_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, x_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_YTile_##x##_xRGB8888 +#define TAG2(x,y) intel_YTile_##x##_xRGB8888##y +#define GET_VALUE(X, Y) pread_xrgb8888(irb, y_tile_swizzle(irb, intel, X, Y)) +#define PUT_VALUE(X, Y, V) pwrite_xrgb8888(irb, y_tile_swizzle(irb, intel, X, Y), V) +#include "spantmp2.h" + +#define LOCAL_DEPTH_VARS \ + struct intel_context *intel = intel_context(ctx); \ + struct intel_renderbuffer *irb = intel_renderbuffer(rb); \ + const GLint yScale = irb->RenderToTexture ? 1 : -1; \ + const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1; + + +#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS + +/** + ** 16-bit depthbuffer functions. + **/ +#define VALUE_TYPE GLushort +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_16(irb, no_tile_swizzle(irb, intel, _x, _y), d) +#define READ_DEPTH(d, _x, _y) \ + d = pread_16(irb, no_tile_swizzle(irb, intel, _x, _y)) +#define TAG(x) intel##x##_z16 +#include "depthtmp.h" + + +/** + ** 16-bit x tile depthbuffer functions. + **/ +#define VALUE_TYPE GLushort +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_16(irb, x_tile_swizzle(irb, intel, _x, _y), d) +#define READ_DEPTH(d, _x, _y) \ + d = pread_16(irb, x_tile_swizzle(irb, intel, _x, _y)) +#define TAG(x) intel_XTile_##x##_z16 +#include "depthtmp.h" + +/** + ** 16-bit y tile depthbuffer functions. + **/ +#define VALUE_TYPE GLushort +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_16(irb, y_tile_swizzle(irb, intel, _x, _y), d) +#define READ_DEPTH(d, _x, _y) \ + d = pread_16(irb, y_tile_swizzle(irb, intel, _x, _y)) +#define TAG(x) intel_YTile_##x##_z16 +#include "depthtmp.h" + + +/** + ** 24/8-bit interleaved depth/stencil functions + ** Note: we're actually reading back combined depth+stencil values. + ** The wrappers in main/depthstencil.c are used to extract the depth + ** and stencil values. + **/ +#define VALUE_TYPE GLuint + +/* Change ZZZS -> SZZZ */ +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_32(irb, no_tile_swizzle(irb, intel, _x, _y), \ + ((d) >> 8) | ((d) << 24)) + +/* Change SZZZ -> ZZZS */ +#define READ_DEPTH( d, _x, _y ) { \ + GLuint tmp = pread_32(irb, no_tile_swizzle(irb, intel, _x, _y)); \ + d = (tmp << 8) | (tmp >> 24); \ +} + +#define TAG(x) intel##x##_z24_s8 +#include "depthtmp.h" + + +/** + ** 24/8-bit x-tile interleaved depth/stencil functions + ** Note: we're actually reading back combined depth+stencil values. + ** The wrappers in main/depthstencil.c are used to extract the depth + ** and stencil values. + **/ +#define VALUE_TYPE GLuint + +/* Change ZZZS -> SZZZ */ +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_32(irb, x_tile_swizzle(irb, intel, _x, _y), \ + ((d) >> 8) | ((d) << 24)) \ + +/* Change SZZZ -> ZZZS */ +#define READ_DEPTH( d, _x, _y ) { \ + GLuint tmp = pread_32(irb, x_tile_swizzle(irb, intel, _x, _y)); \ + d = (tmp << 8) | (tmp >> 24); \ +} + +#define TAG(x) intel_XTile_##x##_z24_s8 +#include "depthtmp.h" + +/** + ** 24/8-bit y-tile interleaved depth/stencil functions + ** Note: we're actually reading back combined depth+stencil values. + ** The wrappers in main/depthstencil.c are used to extract the depth + ** and stencil values. + **/ +#define VALUE_TYPE GLuint + +/* Change ZZZS -> SZZZ */ +#define WRITE_DEPTH(_x, _y, d) \ + pwrite_32(irb, y_tile_swizzle(irb, intel, _x, _y), \ + ((d) >> 8) | ((d) << 24)) + +/* Change SZZZ -> ZZZS */ +#define READ_DEPTH( d, _x, _y ) { \ + GLuint tmp = pread_32(irb, y_tile_swizzle(irb, intel, _x, _y)); \ + d = (tmp << 8) | (tmp >> 24); \ +} + +#define TAG(x) intel_YTile_##x##_z24_s8 +#include "depthtmp.h" + + +/** + ** 8-bit stencil function (XXX FBO: This is obsolete) + **/ +#define WRITE_STENCIL(_x, _y, d) \ + pwrite_8(irb, no_tile_swizzle(irb, intel, _x, _y) + 3, d) + +#define READ_STENCIL(d, _x, _y) \ + d = pread_8(irb, no_tile_swizzle(irb, intel, _x, _y) + 3); + +#define TAG(x) intel##x##_z24_s8 +#include "stenciltmp.h" + +/** + ** 8-bit x-tile stencil function (XXX FBO: This is obsolete) + **/ +#define WRITE_STENCIL(_x, _y, d) \ + pwrite_8(irb, x_tile_swizzle(irb, intel, _x, _y) + 3, d) + +#define READ_STENCIL(d, _x, _y) \ + d = pread_8(irb, x_tile_swizzle(irb, intel, _x, _y) + 3); + +#define TAG(x) intel_XTile_##x##_z24_s8 +#include "stenciltmp.h" + +/** + ** 8-bit y-tile stencil function (XXX FBO: This is obsolete) + **/ +#define WRITE_STENCIL(_x, _y, d) \ + pwrite_8(irb, y_tile_swizzle(irb, intel, _x, _y) + 3, d) + +#define READ_STENCIL(d, _x, _y) \ + d = pread_8(irb, y_tile_swizzle(irb, intel, _x, _y) + 3) + +#define TAG(x) intel_YTile_##x##_z24_s8 +#include "stenciltmp.h" + +void +intel_renderbuffer_map(struct intel_context *intel, struct gl_renderbuffer *rb) +{ + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + + if (irb == NULL || irb->region == NULL) + return; + + irb->pfPitch = irb->region->pitch; + + intel_set_span_functions(intel, rb); +} + +void +intel_renderbuffer_unmap(struct intel_context *intel, + struct gl_renderbuffer *rb) +{ + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + + if (irb == NULL || irb->region == NULL) + return; + + clear_span_cache(irb); + irb->pfPitch = 0; + + rb->GetRow = NULL; + rb->PutRow = NULL; +} + +/** + * Map or unmap all the renderbuffers which we may need during + * software rendering. + * XXX in the future, we could probably convey extra information to + * reduce the number of mappings needed. I.e. if doing a glReadPixels + * from the depth buffer, we really only need one mapping. + * + * XXX Rewrite this function someday. + * We can probably just loop over all the renderbuffer attachments, + * map/unmap all of them, and not worry about the _ColorDrawBuffers + * _ColorReadBuffer, _DepthBuffer or _StencilBuffer fields. + */ +static void +intel_map_unmap_buffers(struct intel_context *intel, GLboolean map) +{ + GLcontext *ctx = &intel->ctx; + GLuint i, j; + + /* color draw buffers */ + for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers; j++) { + if (map) + intel_renderbuffer_map(intel, ctx->DrawBuffer->_ColorDrawBuffers[j]); + else + intel_renderbuffer_unmap(intel, ctx->DrawBuffer->_ColorDrawBuffers[j]); + } + + /* check for render to textures */ + for (i = 0; i < BUFFER_COUNT; i++) { + struct gl_renderbuffer_attachment *att = + ctx->DrawBuffer->Attachment + i; + struct gl_texture_object *tex = att->Texture; + if (tex) { + /* render to texture */ + ASSERT(att->Renderbuffer); + if (map) { + struct gl_texture_image *texImg; + texImg = tex->Image[att->CubeMapFace][att->TextureLevel]; + intel_tex_map_images(intel, intel_texture_object(tex)); + } + else { + intel_tex_unmap_images(intel, intel_texture_object(tex)); + } + } + } + + /* color read buffers */ + if (map) + intel_renderbuffer_map(intel, ctx->ReadBuffer->_ColorReadBuffer); + else + intel_renderbuffer_unmap(intel, ctx->ReadBuffer->_ColorReadBuffer); + + /* depth buffer (Note wrapper!) */ + if (ctx->DrawBuffer->_DepthBuffer) { + if (map) + intel_renderbuffer_map(intel, ctx->DrawBuffer->_DepthBuffer->Wrapped); + else + intel_renderbuffer_unmap(intel, + ctx->DrawBuffer->_DepthBuffer->Wrapped); + } + + /* stencil buffer (Note wrapper!) */ + if (ctx->DrawBuffer->_StencilBuffer) { + if (map) + intel_renderbuffer_map(intel, + ctx->DrawBuffer->_StencilBuffer->Wrapped); + else + intel_renderbuffer_unmap(intel, + ctx->DrawBuffer->_StencilBuffer->Wrapped); + } +} + + + +/** + * Prepare for softare rendering. Map current read/draw framebuffers' + * renderbuffes and all currently bound texture objects. + * + * Old note: Moved locking out to get reasonable span performance. + */ +void +intelSpanRenderStart(GLcontext * ctx) +{ + struct intel_context *intel = intel_context(ctx); + GLuint i; + + intelFlush(&intel->ctx); + LOCK_HARDWARE(intel); + + for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) { + struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current; + intel_tex_map_images(intel, intel_texture_object(texObj)); + } + } + + intel_map_unmap_buffers(intel, GL_TRUE); +} + +/** + * Called when done softare rendering. Unmap the buffers we mapped in + * the above function. + */ +void +intelSpanRenderFinish(GLcontext * ctx) +{ + struct intel_context *intel = intel_context(ctx); + GLuint i; + + _swrast_flush(ctx); + + for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) { + if (ctx->Texture.Unit[i]._ReallyEnabled) { + struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current; + intel_tex_unmap_images(intel, intel_texture_object(texObj)); + } + } + + intel_map_unmap_buffers(intel, GL_FALSE); + + UNLOCK_HARDWARE(intel); +} + + +void +intelInitSpanFuncs(GLcontext * ctx) +{ + struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx); + swdd->SpanRenderStart = intelSpanRenderStart; + swdd->SpanRenderFinish = intelSpanRenderFinish; +} + + +/** + * Plug in appropriate span read/write functions for the given renderbuffer. + * These are used for the software fallbacks. + */ +static void +intel_set_span_functions(struct intel_context *intel, + struct gl_renderbuffer *rb) +{ + struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb; + uint32_t tiling; + + /* If in GEM mode, we need to do the tile address swizzling ourselves, + * instead of the fence registers handling it. + */ + if (intel->ttm) + tiling = irb->region->tiling; + else + tiling = I915_TILING_NONE; + + if (rb->_ActualFormat == GL_RGB5) { + /* 565 RGB */ + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitPointers_RGB565(rb); + break; + case I915_TILING_X: + intel_XTile_InitPointers_RGB565(rb); + break; + case I915_TILING_Y: + intel_YTile_InitPointers_RGB565(rb); + break; + } + } + else if (rb->_ActualFormat == GL_RGB8) { + /* 8888 RGBx */ + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitPointers_xRGB8888(rb); + break; + case I915_TILING_X: + intel_XTile_InitPointers_xRGB8888(rb); + break; + case I915_TILING_Y: + intel_YTile_InitPointers_xRGB8888(rb); + break; + } + } + else if (rb->_ActualFormat == GL_RGBA8) { + /* 8888 RGBA */ + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitPointers_ARGB8888(rb); + break; + case I915_TILING_X: + intel_XTile_InitPointers_ARGB8888(rb); + break; + case I915_TILING_Y: + intel_YTile_InitPointers_ARGB8888(rb); + break; + } + } + else if (rb->_ActualFormat == GL_DEPTH_COMPONENT16) { + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitDepthPointers_z16(rb); + break; + case I915_TILING_X: + intel_XTile_InitDepthPointers_z16(rb); + break; + case I915_TILING_Y: + intel_YTile_InitDepthPointers_z16(rb); + break; + } + } + else if (rb->_ActualFormat == GL_DEPTH_COMPONENT24 || /* XXX FBO remove */ + rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) { + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitDepthPointers_z24_s8(rb); + break; + case I915_TILING_X: + intel_XTile_InitDepthPointers_z24_s8(rb); + break; + case I915_TILING_Y: + intel_YTile_InitDepthPointers_z24_s8(rb); + break; + } + } + else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) { + switch (tiling) { + case I915_TILING_NONE: + default: + intelInitStencilPointers_z24_s8(rb); + break; + case I915_TILING_X: + intel_XTile_InitStencilPointers_z24_s8(rb); + break; + case I915_TILING_Y: + intel_YTile_InitStencilPointers_z24_s8(rb); + break; + } + } + else { + _mesa_problem(NULL, + "Unexpected _ActualFormat in intelSetSpanFunctions"); + } +} diff --git a/src/mesa/drivers/dri/intel/intel_span.h b/src/mesa/drivers/dri/intel/intel_span.h new file mode 100644 index 0000000000..acbeb4abe1 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_span.h @@ -0,0 +1,40 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _INTEL_SPAN_H +#define _INTEL_SPAN_H + +extern void intelInitSpanFuncs(GLcontext * ctx); + +extern void intelSpanRenderFinish(GLcontext * ctx); +extern void intelSpanRenderStart(GLcontext * ctx); +void intel_renderbuffer_map(struct intel_context *intel, + struct gl_renderbuffer *rb); +void intel_renderbuffer_unmap(struct intel_context *intel, + struct gl_renderbuffer *rb); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_tex.c b/src/mesa/drivers/dri/intel/intel_tex.c new file mode 100644 index 0000000000..4fa18e2d54 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex.c @@ -0,0 +1,256 @@ +#include "swrast/swrast.h" +#include "texobj.h" +#include "teximage.h" +#include "mipmap.h" +#include "intel_context.h" +#include "intel_mipmap_tree.h" +#include "intel_tex.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +static GLboolean +intelIsTextureResident(GLcontext * ctx, struct gl_texture_object *texObj) +{ +#if 0 + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + + return + intelObj->mt && + intelObj->mt->region && + intel_is_region_resident(intel, intelObj->mt->region); +#endif + return 1; +} + + + +static struct gl_texture_image * +intelNewTextureImage(GLcontext * ctx) +{ + DBG("%s\n", __FUNCTION__); + (void) ctx; + return (struct gl_texture_image *) CALLOC_STRUCT(intel_texture_image); +} + + +static struct gl_texture_object * +intelNewTextureObject(GLcontext * ctx, GLuint name, GLenum target) +{ + struct intel_texture_object *obj = CALLOC_STRUCT(intel_texture_object); + + DBG("%s\n", __FUNCTION__); + _mesa_initialize_texture_object(&obj->base, name, target); + + return &obj->base; +} + +static void +intelDeleteTextureObject(GLcontext *ctx, + struct gl_texture_object *texObj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + + if (intelObj->mt) + intel_miptree_release(intel, &intelObj->mt); + + _mesa_delete_texture_object(ctx, texObj); +} + + +static void +intelFreeTextureImageData(GLcontext * ctx, struct gl_texture_image *texImage) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_image *intelImage = intel_texture_image(texImage); + + DBG("%s\n", __FUNCTION__); + + if (intelImage->mt) { + intel_miptree_release(intel, &intelImage->mt); + } + + if (texImage->Data) { + _mesa_free_texmemory(texImage->Data); + texImage->Data = NULL; + } +} + + +/* The system memcpy (at least on ubuntu 5.10) has problems copying + * to agp (writecombined) memory from a source which isn't 64-byte + * aligned - there is a 4x performance falloff. + * + * The x86 __memcpy is immune to this but is slightly slower + * (10%-ish) than the system memcpy. + * + * The sse_memcpy seems to have a slight cliff at 64/32 bytes, but + * isn't much faster than x86_memcpy for agp copies. + * + * TODO: switch dynamically. + */ +static void * +do_memcpy(void *dest, const void *src, size_t n) +{ + if ((((unsigned) src) & 63) || (((unsigned) dest) & 63)) { + return __memcpy(dest, src, n); + } + else + return memcpy(dest, src, n); +} + + +#if DO_DEBUG && !defined(__ia64__) + +#ifndef __x86_64__ +static unsigned +fastrdtsc(void) +{ + unsigned eax; + __asm__ volatile ("\t" + "pushl %%ebx\n\t" + "cpuid\n\t" ".byte 0x0f, 0x31\n\t" + "popl %%ebx\n":"=a" (eax) + :"0"(0) + :"ecx", "edx", "cc"); + + return eax; +} +#else +static unsigned +fastrdtsc(void) +{ + unsigned eax; + __asm__ volatile ("\t" "cpuid\n\t" ".byte 0x0f, 0x31\n\t":"=a" (eax) + :"0"(0) + :"ecx", "edx", "ebx", "cc"); + + return eax; +} +#endif + +static unsigned +time_diff(unsigned t, unsigned t2) +{ + return ((t < t2) ? t2 - t : 0xFFFFFFFFU - (t - t2 - 1)); +} + + +static void * +timed_memcpy(void *dest, const void *src, size_t n) +{ + void *ret; + unsigned t1, t2; + double rate; + + if ((((unsigned) src) & 63) || (((unsigned) dest) & 63)) + _mesa_printf("Warning - non-aligned texture copy!\n"); + + t1 = fastrdtsc(); + ret = do_memcpy(dest, src, n); + t2 = fastrdtsc(); + + rate = time_diff(t1, t2); + rate /= (double) n; + _mesa_printf("timed_memcpy: %u %u --> %f clocks/byte\n", t1, t2, rate); + return ret; +} +#endif /* DO_DEBUG */ + +/** + * Generate new mipmap data from BASE+1 to BASE+p (the minimally-sized mipmap + * level). + * + * The texture object's miptree must be mapped. + * + * It would be really nice if this was just called by Mesa whenever mipmaps + * needed to be regenerated, rather than us having to remember to do so in + * each texture image modification path. + * + * This function should also include an accelerated path. + */ +void +intel_generate_mipmap(GLcontext *ctx, GLenum target, + struct gl_texture_object *texObj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + int face, i; + + _mesa_generate_mipmap(ctx, target, texObj); + + /* Update the level information in our private data in the new images, since + * it didn't get set as part of a normal TexImage path. + */ + for (face = 0; face < nr_faces; face++) { + for (i = texObj->BaseLevel + 1; i < texObj->MaxLevel; i++) { + struct intel_texture_image *intelImage; + + intelImage = intel_texture_image(texObj->Image[face][i]); + if (intelImage == NULL) + break; + + intelImage->level = i; + intelImage->face = face; + /* Unreference the miptree to signal that the new Data is a bare + * pointer from mesa. + */ + intel_miptree_release(intel, &intelImage->mt); + } + } +} + +static void intelGenerateMipmap(GLcontext *ctx, GLenum target, struct gl_texture_object *texObj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + + intel_tex_map_level_images(intel, intelObj, texObj->BaseLevel); + intel_generate_mipmap(ctx, target, texObj); + intel_tex_unmap_level_images(intel, intelObj, texObj->BaseLevel); +} + +void +intelInitTextureFuncs(struct dd_function_table *functions) +{ + functions->ChooseTextureFormat = intelChooseTextureFormat; + functions->TexImage1D = intelTexImage1D; + functions->TexImage2D = intelTexImage2D; + functions->TexImage3D = intelTexImage3D; + functions->TexSubImage1D = intelTexSubImage1D; + functions->TexSubImage2D = intelTexSubImage2D; + functions->TexSubImage3D = intelTexSubImage3D; +#ifdef I915 + functions->CopyTexImage1D = intelCopyTexImage1D; + functions->CopyTexImage2D = intelCopyTexImage2D; + functions->CopyTexSubImage1D = intelCopyTexSubImage1D; + functions->CopyTexSubImage2D = intelCopyTexSubImage2D; +#else + functions->CopyTexImage1D = _swrast_copy_teximage1d; + functions->CopyTexImage2D = _swrast_copy_teximage2d; + functions->CopyTexSubImage1D = _swrast_copy_texsubimage1d; + functions->CopyTexSubImage2D = _swrast_copy_texsubimage2d; +#endif + functions->GetTexImage = intelGetTexImage; + functions->GenerateMipmap = intelGenerateMipmap; + + /* compressed texture functions */ + functions->CompressedTexImage2D = intelCompressedTexImage2D; + functions->GetCompressedTexImage = intelGetCompressedTexImage; + + functions->NewTextureObject = intelNewTextureObject; + functions->NewTextureImage = intelNewTextureImage; + functions->DeleteTexture = intelDeleteTextureObject; + functions->FreeTexImageData = intelFreeTextureImageData; + functions->UpdateTexturePalette = 0; + functions->IsTextureResident = intelIsTextureResident; + +#if DO_DEBUG && !defined(__ia64__) + if (INTEL_DEBUG & DEBUG_BUFMGR) + functions->TextureMemCpy = timed_memcpy; + else +#endif + functions->TextureMemCpy = do_memcpy; +} diff --git a/src/mesa/drivers/dri/intel/intel_tex.h b/src/mesa/drivers/dri/intel/intel_tex.h new file mode 100644 index 0000000000..fe7a8badf8 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex.h @@ -0,0 +1,164 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef INTELTEX_INC +#define INTELTEX_INC + +#include "mtypes.h" +#include "intel_context.h" +#include "texmem.h" + + +void intelInitTextureFuncs(struct dd_function_table *functions); + +const struct gl_texture_format *intelChooseTextureFormat(GLcontext * ctx, + GLint internalFormat, + GLenum format, + GLenum type); + + +void intelTexImage3D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelTexSubImage3D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelTexImage2D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelTexSubImage2D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelTexImage1D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelTexSubImage1D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, + GLsizei width, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLint border); + +void intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLsizei height, + GLint border); + +void intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint x, GLint y, GLsizei width); + +void intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLint x, GLint y, GLsizei width, GLsizei height); + +void intelGetTexImage(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage ); + +void intelGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, + GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage); + +void intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname, + unsigned long long offset, GLint depth, GLuint pitch); +void intelSetTexBuffer(__DRIcontext *pDRICtx, + GLint target, __DRIdrawable *pDraw); + +GLuint intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit); + +void intel_tex_map_level_images(struct intel_context *intel, + struct intel_texture_object *intelObj, + int level); + +void intel_tex_unmap_level_images(struct intel_context *intel, + struct intel_texture_object *intelObj, + int level); + +void intel_tex_map_images(struct intel_context *intel, + struct intel_texture_object *intelObj); + +void intel_tex_unmap_images(struct intel_context *intel, + struct intel_texture_object *intelObj); + +int intel_compressed_num_bytes(GLuint mesaFormat); + +void intel_generate_mipmap(GLcontext *ctx, GLenum target, + struct gl_texture_object *texObj); + +#endif diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c new file mode 100644 index 0000000000..d0ab464a1c --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c @@ -0,0 +1,298 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "mtypes.h" +#include "enums.h" +#include "image.h" +#include "teximage.h" +#include "mipmap.h" +#include "swrast/swrast.h" + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_buffers.h" +#include "intel_mipmap_tree.h" +#include "intel_regions.h" +#include "intel_fbo.h" +#include "intel_tex.h" +#include "intel_blit.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +/** + * Get the intel_region which is the source for any glCopyTex[Sub]Image call. + * + * Do the best we can using the blitter. A future project is to use + * the texture engine and fragment programs for these copies. + */ +static const struct intel_region * +get_teximage_source(struct intel_context *intel, GLenum internalFormat) +{ + struct intel_renderbuffer *irb; + + DBG("%s %s\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(internalFormat)); + + switch (internalFormat) { + case GL_DEPTH_COMPONENT: + case GL_DEPTH_COMPONENT16_ARB: + irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH); + if (irb && irb->region && irb->region->cpp == 2) + return irb->region; + return NULL; + case GL_DEPTH24_STENCIL8_EXT: + case GL_DEPTH_STENCIL_EXT: + irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH); + if (irb && irb->region && irb->region->cpp == 4) + return irb->region; + return NULL; + case GL_RGBA: + case GL_RGBA8: + return intel_readbuf_region(intel); + case GL_RGB: + if (intel->ctx.Visual.rgbBits == 16) + return intel_readbuf_region(intel); + return NULL; + default: + return NULL; + } +} + + +static GLboolean +do_copy_texsubimage(struct intel_context *intel, + GLenum target, + struct intel_texture_image *intelImage, + GLenum internalFormat, + GLint dstx, GLint dsty, + GLint x, GLint y, GLsizei width, GLsizei height) +{ + GLcontext *ctx = &intel->ctx; + struct gl_texture_object *texObj = intelImage->base.TexObject; + const struct intel_region *src = + get_teximage_source(intel, internalFormat); + + if (!intelImage->mt || !src) { + DBG("%s fail %p %p\n", __FUNCTION__, intelImage->mt, src); + return GL_FALSE; + } + + intelFlush(ctx); + LOCK_HARDWARE(intel); + { + GLuint image_offset = intel_miptree_image_offset(intelImage->mt, + intelImage->face, + intelImage->level); + const GLint orig_x = x; + const GLint orig_y = y; + const struct gl_framebuffer *fb = ctx->DrawBuffer; + + if (_mesa_clip_to_region(fb->_Xmin, fb->_Ymin, fb->_Xmax, fb->_Ymax, + &x, &y, &width, &height)) { + /* Update dst for clipped src. Need to also clip the source rect. + */ + dstx += x - orig_x; + dsty += y - orig_y; + + if (ctx->ReadBuffer->Name == 0) { + /* reading from a window, adjust x, y */ + __DRIdrawablePrivate *dPriv = intel->driDrawable; + GLuint window_y; + /* window_y = position of window on screen if y=0=bottom */ + window_y = intel->intelScreen->height - (dPriv->y + dPriv->h); + y = window_y + y; + x += dPriv->x; + } + else { + /* reading from a FBO */ + /* invert Y */ + y = ctx->ReadBuffer->Height - y - 1; + } + + + /* A bit of fiddling to get the blitter to work with -ve + * pitches. But we get a nice inverted blit this way, so it's + * worth it: + */ + intelEmitCopyBlit(intel, + intelImage->mt->cpp, + -src->pitch, + src->buffer, + src->height * src->pitch * src->cpp, + src->tiling, + intelImage->mt->pitch, + intelImage->mt->region->buffer, + image_offset, + intelImage->mt->region->tiling, + x, y + height, dstx, dsty, width, height, + GL_COPY); /* ? */ + } + } + + + UNLOCK_HARDWARE(intel); + + /* GL_SGIS_generate_mipmap */ + if (intelImage->level == texObj->BaseLevel && texObj->GenerateMipmap) { + intel_generate_mipmap(ctx, target, texObj); + } + + return GL_TRUE; +} + + + + + +void +intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLint border) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + + if (border) + goto fail; + + /* Setup or redefine the texture object, mipmap tree and texture + * image. Don't populate yet. + */ + ctx->Driver.TexImage1D(ctx, target, level, internalFormat, + width, border, + GL_RGBA, CHAN_TYPE, NULL, + &ctx->DefaultPacking, texObj, texImage); + + if (!do_copy_texsubimage(intel_context(ctx), target, + intel_texture_image(texImage), + internalFormat, 0, 0, x, y, width, 1)) + goto fail; + + return; + + fail: + _swrast_copy_teximage1d(ctx, target, level, internalFormat, x, y, + width, border); +} + +void +intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level, + GLenum internalFormat, + GLint x, GLint y, GLsizei width, GLsizei height, + GLint border) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + + if (border) + goto fail; + + /* Setup or redefine the texture object, mipmap tree and texture + * image. Don't populate yet. + */ + ctx->Driver.TexImage2D(ctx, target, level, internalFormat, + width, height, border, + GL_RGBA, CHAN_TYPE, NULL, + &ctx->DefaultPacking, texObj, texImage); + + + if (!do_copy_texsubimage(intel_context(ctx), target, + intel_texture_image(texImage), + internalFormat, 0, 0, x, y, width, height)) + goto fail; + + return; + + fail: + _swrast_copy_teximage2d(ctx, target, level, internalFormat, x, y, + width, height, border); +} + + +void +intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint x, GLint y, GLsizei width) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + GLenum internalFormat = texImage->InternalFormat; + + /* XXX need to check <border> as in above function? */ + + /* Need to check texture is compatible with source format. + */ + + if (!do_copy_texsubimage(intel_context(ctx), target, + intel_texture_image(texImage), + internalFormat, xoffset, 0, x, y, width, 1)) { + _swrast_copy_texsubimage1d(ctx, target, level, xoffset, x, y, width); + } +} + + + +void +intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, + GLint xoffset, GLint yoffset, + GLint x, GLint y, GLsizei width, GLsizei height) +{ + struct gl_texture_unit *texUnit = + &ctx->Texture.Unit[ctx->Texture.CurrentUnit]; + struct gl_texture_object *texObj = + _mesa_select_tex_object(ctx, texUnit, target); + struct gl_texture_image *texImage = + _mesa_select_tex_image(ctx, texObj, target, level); + GLenum internalFormat = texImage->InternalFormat; + + + /* Need to check texture is compatible with source format. + */ + + if (!do_copy_texsubimage(intel_context(ctx), target, + intel_texture_image(texImage), + internalFormat, + xoffset, yoffset, x, y, width, height)) { + + DBG("%s - fallback to swrast\n", __FUNCTION__); + + _swrast_copy_texsubimage2d(ctx, target, level, + xoffset, yoffset, x, y, width, height); + } +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_format.c b/src/mesa/drivers/dri/intel/intel_tex_format.c new file mode 100644 index 0000000000..8ae80e1601 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_format.c @@ -0,0 +1,193 @@ +#include "intel_context.h" +#include "intel_tex.h" +#include "texformat.h" +#include "enums.h" + +/* It works out that this function is fine for all the supported + * hardware. However, there is still a need to map the formats onto + * hardware descriptors. + */ +/* Note that the i915 can actually support many more formats than + * these if we take the step of simply swizzling the colors + * immediately after sampling... + */ +const struct gl_texture_format * +intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat, + GLenum format, GLenum type) +{ + struct intel_context *intel = intel_context(ctx); + const GLboolean do32bpt = (intel->ctx.Visual.rgbBits == 32); + + switch (internalFormat) { + case 4: + case GL_RGBA: + case GL_COMPRESSED_RGBA: + if (format == GL_BGRA) { + if (type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) { + return &_mesa_texformat_argb8888; + } + else if (type == GL_UNSIGNED_SHORT_4_4_4_4_REV) { + return &_mesa_texformat_argb4444; + } + else if (type == GL_UNSIGNED_SHORT_1_5_5_5_REV) { + return &_mesa_texformat_argb1555; + } + } + return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444; + + case 3: + case GL_RGB: + case GL_COMPRESSED_RGB: + if (format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5) { + return &_mesa_texformat_rgb565; + } + return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565; + + case GL_RGBA8: + case GL_RGB10_A2: + case GL_RGBA12: + case GL_RGBA16: + return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444; + + case GL_RGBA4: + case GL_RGBA2: + return &_mesa_texformat_argb4444; + + case GL_RGB5_A1: + return &_mesa_texformat_argb1555; + + case GL_RGB8: + case GL_RGB10: + case GL_RGB12: + case GL_RGB16: + return &_mesa_texformat_argb8888; + + case GL_RGB5: + case GL_RGB4: + case GL_R3_G3_B2: + return &_mesa_texformat_rgb565; + + case GL_ALPHA: + case GL_ALPHA4: + case GL_ALPHA8: + case GL_ALPHA12: + case GL_ALPHA16: + case GL_COMPRESSED_ALPHA: + return &_mesa_texformat_a8; + + case 1: + case GL_LUMINANCE: + case GL_LUMINANCE4: + case GL_LUMINANCE8: + case GL_LUMINANCE12: + case GL_LUMINANCE16: + case GL_COMPRESSED_LUMINANCE: + return &_mesa_texformat_l8; + + case 2: + case GL_LUMINANCE_ALPHA: + case GL_LUMINANCE4_ALPHA4: + case GL_LUMINANCE6_ALPHA2: + case GL_LUMINANCE8_ALPHA8: + case GL_LUMINANCE12_ALPHA4: + case GL_LUMINANCE12_ALPHA12: + case GL_LUMINANCE16_ALPHA16: + case GL_COMPRESSED_LUMINANCE_ALPHA: + return &_mesa_texformat_al88; + + case GL_INTENSITY: + case GL_INTENSITY4: + case GL_INTENSITY8: + case GL_INTENSITY12: + case GL_INTENSITY16: + case GL_COMPRESSED_INTENSITY: + return &_mesa_texformat_i8; + + case GL_YCBCR_MESA: + if (type == GL_UNSIGNED_SHORT_8_8_MESA || type == GL_UNSIGNED_BYTE) + return &_mesa_texformat_ycbcr; + else + return &_mesa_texformat_ycbcr_rev; + + case GL_COMPRESSED_RGB_FXT1_3DFX: + return &_mesa_texformat_rgb_fxt1; + case GL_COMPRESSED_RGBA_FXT1_3DFX: + return &_mesa_texformat_rgba_fxt1; + + case GL_RGB_S3TC: + case GL_RGB4_S3TC: + case GL_COMPRESSED_RGB_S3TC_DXT1_EXT: + return &_mesa_texformat_rgb_dxt1; + + case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT: + return &_mesa_texformat_rgba_dxt1; + + case GL_RGBA_S3TC: + case GL_RGBA4_S3TC: + case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT: + return &_mesa_texformat_rgba_dxt3; + + case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT: + return &_mesa_texformat_rgba_dxt5; + + case GL_DEPTH_COMPONENT: + case GL_DEPTH_COMPONENT16: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: + return &_mesa_texformat_z16; + + case GL_DEPTH_STENCIL_EXT: + case GL_DEPTH24_STENCIL8_EXT: + return &_mesa_texformat_s8_z24; + +#ifndef I915 + case GL_SRGB_EXT: + case GL_SRGB8_EXT: + case GL_SRGB_ALPHA_EXT: + case GL_SRGB8_ALPHA8_EXT: + case GL_SLUMINANCE_EXT: + case GL_SLUMINANCE8_EXT: + case GL_SLUMINANCE_ALPHA_EXT: + case GL_SLUMINANCE8_ALPHA8_EXT: + case GL_COMPRESSED_SRGB_EXT: + case GL_COMPRESSED_SRGB_ALPHA_EXT: + case GL_COMPRESSED_SLUMINANCE_EXT: + case GL_COMPRESSED_SLUMINANCE_ALPHA_EXT: + return &_mesa_texformat_srgba8; + case GL_COMPRESSED_SRGB_S3TC_DXT1_EXT: + case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT: + case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT: + case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT: + return &_mesa_texformat_srgb_dxt1; +#endif + + default: + fprintf(stderr, "unexpected texture format %s in %s\n", + _mesa_lookup_enum_by_nr(internalFormat), __FUNCTION__); + return NULL; + } + + return NULL; /* never get here */ +} + +int intel_compressed_num_bytes(GLuint mesaFormat) +{ + int bytes = 0; + switch(mesaFormat) { + + case MESA_FORMAT_RGB_FXT1: + case MESA_FORMAT_RGBA_FXT1: + case MESA_FORMAT_RGB_DXT1: + case MESA_FORMAT_RGBA_DXT1: + bytes = 2; + break; + + case MESA_FORMAT_RGBA_DXT3: + case MESA_FORMAT_RGBA_DXT5: + bytes = 4; + default: + break; + } + + return bytes; +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c new file mode 100644 index 0000000000..d1f540b0f1 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_image.c @@ -0,0 +1,772 @@ + +#include <stdlib.h> +#include <stdio.h> + +#include "glheader.h" +#include "macros.h" +#include "mtypes.h" +#include "enums.h" +#include "colortab.h" +#include "convolve.h" +#include "context.h" +#include "simple_list.h" +#include "texcompress.h" +#include "texformat.h" +#include "texobj.h" +#include "texstore.h" +#include "teximage.h" + +#include "intel_context.h" +#include "intel_mipmap_tree.h" +#include "intel_buffer_objects.h" +#include "intel_batchbuffer.h" +#include "intel_tex.h" +#include "intel_blit.h" +#include "intel_fbo.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +/* Functions to store texture images. Where possible, mipmap_tree's + * will be created or further instantiated with image data, otherwise + * images will be stored in malloc'd memory. A validation step is + * required to pull those images into a mipmap tree, or otherwise + * decide a fallback is required. + */ + + +static int +logbase2(int n) +{ + GLint i = 1; + GLint log2 = 0; + + while (n > i) { + i *= 2; + log2++; + } + + return log2; +} + + +/* Otherwise, store it in memory if (Border != 0) or (any dimension == + * 1). + * + * Otherwise, if max_level >= level >= min_level, create tree with + * space for textures from min_level down to max_level. + * + * Otherwise, create tree with space for textures from (level + * 0)..(1x1). Consider pruning this tree at a validation if the + * saving is worth it. + */ +static void +guess_and_alloc_mipmap_tree(struct intel_context *intel, + struct intel_texture_object *intelObj, + struct intel_texture_image *intelImage) +{ + GLuint firstLevel; + GLuint lastLevel; + GLuint width = intelImage->base.Width; + GLuint height = intelImage->base.Height; + GLuint depth = intelImage->base.Depth; + GLuint l2width, l2height, l2depth; + GLuint i, comp_byte = 0; + + DBG("%s\n", __FUNCTION__); + + if (intelImage->base.Border || + ((intelImage->base._BaseFormat == GL_DEPTH_COMPONENT) && + ((intelObj->base.WrapS == GL_CLAMP_TO_BORDER) || + (intelObj->base.WrapT == GL_CLAMP_TO_BORDER)))) + return; + + if (intelImage->level > intelObj->base.BaseLevel && + (intelImage->base.Width == 1 || + (intelObj->base.Target != GL_TEXTURE_1D && + intelImage->base.Height == 1) || + (intelObj->base.Target == GL_TEXTURE_3D && + intelImage->base.Depth == 1))) + return; + + /* If this image disrespects BaseLevel, allocate from level zero. + * Usually BaseLevel == 0, so it's unlikely to happen. + */ + if (intelImage->level < intelObj->base.BaseLevel) + firstLevel = 0; + else + firstLevel = intelObj->base.BaseLevel; + + + /* Figure out image dimensions at start level. + */ + for (i = intelImage->level; i > firstLevel; i--) { + width <<= 1; + if (height != 1) + height <<= 1; + if (depth != 1) + depth <<= 1; + } + + /* Guess a reasonable value for lastLevel. This is probably going + * to be wrong fairly often and might mean that we have to look at + * resizable buffers, or require that buffers implement lazy + * pagetable arrangements. + */ + if ((intelObj->base.MinFilter == GL_NEAREST || + intelObj->base.MinFilter == GL_LINEAR) && + intelImage->level == firstLevel) { + lastLevel = firstLevel; + } + else { + l2width = logbase2(width); + l2height = logbase2(height); + l2depth = logbase2(depth); + lastLevel = firstLevel + MAX2(MAX2(l2width, l2height), l2depth); + } + + assert(!intelObj->mt); + if (intelImage->base.IsCompressed) + comp_byte = intel_compressed_num_bytes(intelImage->base.TexFormat->MesaFormat); + intelObj->mt = intel_miptree_create(intel, + intelObj->base.Target, + intelImage->base.InternalFormat, + firstLevel, + lastLevel, + width, + height, + depth, + intelImage->base.TexFormat->TexelBytes, + comp_byte); + + DBG("%s - success\n", __FUNCTION__); +} + + + + +static GLuint +target_to_face(GLenum target) +{ + switch (target) { + case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB: + case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB: + case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB: + return ((GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X); + default: + return 0; + } +} + +/* There are actually quite a few combinations this will work for, + * more than what I've listed here. + */ +static GLboolean +check_pbo_format(GLint internalFormat, + GLenum format, GLenum type, + const struct gl_texture_format *mesa_format) +{ + switch (internalFormat) { + case 4: + case GL_RGBA: + return (format == GL_BGRA && + (type == GL_UNSIGNED_BYTE || + type == GL_UNSIGNED_INT_8_8_8_8_REV) && + mesa_format == &_mesa_texformat_argb8888); + case 3: + case GL_RGB: + return (format == GL_RGB && + type == GL_UNSIGNED_SHORT_5_6_5 && + mesa_format == &_mesa_texformat_rgb565); + case GL_YCBCR_MESA: + return (type == GL_UNSIGNED_SHORT_8_8_MESA || type == GL_UNSIGNED_BYTE); + default: + return GL_FALSE; + } +} + + +/* XXX: Do this for TexSubImage also: + */ +static GLboolean +try_pbo_upload(struct intel_context *intel, + struct intel_texture_image *intelImage, + const struct gl_pixelstore_attrib *unpack, + GLint internalFormat, + GLint width, GLint height, + GLenum format, GLenum type, const void *pixels) +{ + struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj); + GLuint src_offset, src_stride; + GLuint dst_offset, dst_stride; + + if (!pbo || + intel->ctx._ImageTransferState || + unpack->SkipPixels || unpack->SkipRows) { + _mesa_printf("%s: failure 1\n", __FUNCTION__); + return GL_FALSE; + } + + src_offset = (GLuint) pixels; + + if (unpack->RowLength > 0) + src_stride = unpack->RowLength; + else + src_stride = width; + + dst_offset = intel_miptree_image_offset(intelImage->mt, + intelImage->face, + intelImage->level); + + dst_stride = intelImage->mt->pitch; + + intelFlush(&intel->ctx); + LOCK_HARDWARE(intel); + { + dri_bo *src_buffer = intel_bufferobj_buffer(intel, pbo, INTEL_READ); + dri_bo *dst_buffer = intel_region_buffer(intel, + intelImage->mt->region, + INTEL_WRITE_FULL); + + + intelEmitCopyBlit(intel, + intelImage->mt->cpp, + src_stride, src_buffer, src_offset, GL_FALSE, + dst_stride, dst_buffer, dst_offset, GL_FALSE, + 0, 0, 0, 0, width, height, + GL_COPY); + } + UNLOCK_HARDWARE(intel); + + return GL_TRUE; +} + + + +static GLboolean +try_pbo_zcopy(struct intel_context *intel, + struct intel_texture_image *intelImage, + const struct gl_pixelstore_attrib *unpack, + GLint internalFormat, + GLint width, GLint height, + GLenum format, GLenum type, const void *pixels) +{ + struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj); + GLuint src_offset, src_stride; + GLuint dst_offset, dst_stride; + + if (!pbo || + intel->ctx._ImageTransferState || + unpack->SkipPixels || unpack->SkipRows) { + _mesa_printf("%s: failure 1\n", __FUNCTION__); + return GL_FALSE; + } + + src_offset = (GLuint) pixels; + + if (unpack->RowLength > 0) + src_stride = unpack->RowLength; + else + src_stride = width; + + dst_offset = intel_miptree_image_offset(intelImage->mt, + intelImage->face, + intelImage->level); + + dst_stride = intelImage->mt->pitch; + + if (src_stride != dst_stride || dst_offset != 0 || src_offset != 0) { + _mesa_printf("%s: failure 2\n", __FUNCTION__); + return GL_FALSE; + } + + intel_region_attach_pbo(intel, intelImage->mt->region, pbo); + + return GL_TRUE; +} + + + + + + +static void +intelTexImage(GLcontext * ctx, + GLint dims, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, GLsizei imageSize, int compressed) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + struct intel_texture_image *intelImage = intel_texture_image(texImage); + GLint postConvWidth = width; + GLint postConvHeight = height; + GLint texelBytes, sizeInBytes; + GLuint dstRowStride, srcRowStride = texImage->RowStride; + + + DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(target), level, width, height, depth, border); + + intelFlush(ctx); + + intelImage->face = target_to_face(target); + intelImage->level = level; + + if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) { + _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth, + &postConvHeight); + } + + /* choose the texture format */ + texImage->TexFormat = intelChooseTextureFormat(ctx, internalFormat, + format, type); + + _mesa_set_fetch_functions(texImage, dims); + + if (texImage->TexFormat->TexelBytes == 0) { + /* must be a compressed format */ + texelBytes = 0; + texImage->IsCompressed = GL_TRUE; + texImage->CompressedSize = + ctx->Driver.CompressedTextureSize(ctx, texImage->Width, + texImage->Height, texImage->Depth, + texImage->TexFormat->MesaFormat); + } else { + texelBytes = texImage->TexFormat->TexelBytes; + + /* Minimum pitch of 32 bytes */ + if (postConvWidth * texelBytes < 32) { + postConvWidth = 32 / texelBytes; + texImage->RowStride = postConvWidth; + } + + if (!intelImage->mt) { + assert(texImage->RowStride == postConvWidth); + } + } + + /* Release the reference to a potentially orphaned buffer. + * Release any old malloced memory. + */ + if (intelImage->mt) { + intel_miptree_release(intel, &intelImage->mt); + assert(!texImage->Data); + } + else if (texImage->Data) { + _mesa_free_texmemory(texImage->Data); + texImage->Data = NULL; + } + + /* If this is the only texture image in the tree, could call + * bmBufferData with NULL data to free the old block and avoid + * waiting on any outstanding fences. + */ + if (intelObj->mt && + intelObj->mt->first_level == level && + intelObj->mt->last_level == level && + intelObj->mt->target != GL_TEXTURE_CUBE_MAP_ARB && + !intel_miptree_match_image(intelObj->mt, &intelImage->base, + intelImage->face, intelImage->level)) { + + DBG("release it\n"); + intel_miptree_release(intel, &intelObj->mt); + assert(!intelObj->mt); + } + + if (!intelObj->mt) { + guess_and_alloc_mipmap_tree(intel, intelObj, intelImage); + if (!intelObj->mt) { + DBG("guess_and_alloc_mipmap_tree: failed\n"); + } + } + + assert(!intelImage->mt); + + if (intelObj->mt && + intel_miptree_match_image(intelObj->mt, &intelImage->base, + intelImage->face, intelImage->level)) { + + intel_miptree_reference(&intelImage->mt, intelObj->mt); + assert(intelImage->mt); + } else if (intelImage->base.Border == 0) { + int comp_byte = 0; + + if (intelImage->base.IsCompressed) { + comp_byte = + intel_compressed_num_bytes(intelImage->base.TexFormat->MesaFormat); + } + + /* Didn't fit in the object miptree, but it's suitable for inclusion in + * a miptree, so create one just for our level and store it in the image. + * It'll get moved into the object miptree at validate time. + */ + intelImage->mt = intel_miptree_create(intel, target, internalFormat, + level, level, + width, height, depth, + intelImage->base.TexFormat->TexelBytes, + comp_byte); + + } + + /* PBO fastpaths: + */ + if (dims <= 2 && + intelImage->mt && + intel_buffer_object(unpack->BufferObj) && + check_pbo_format(internalFormat, format, + type, intelImage->base.TexFormat)) { + + DBG("trying pbo upload\n"); + + /* Attempt to texture directly from PBO data (zero copy upload). + * + * Currently disable as it can lead to worse as well as better + * performance (in particular when intel_region_cow() is + * required). + */ + if (intelObj->mt == intelImage->mt && + intelObj->mt->first_level == level && + intelObj->mt->last_level == level) { + + if (try_pbo_zcopy(intel, intelImage, unpack, + internalFormat, + width, height, format, type, pixels)) { + + DBG("pbo zcopy upload succeeded\n"); + return; + } + } + + + /* Otherwise, attempt to use the blitter for PBO image uploads. + */ + if (try_pbo_upload(intel, intelImage, unpack, + internalFormat, + width, height, format, type, pixels)) { + DBG("pbo upload succeeded\n"); + return; + } + + DBG("pbo upload failed\n"); + } + + + + /* intelCopyTexImage calls this function with pixels == NULL, with + * the expectation that the mipmap tree will be set up but nothing + * more will be done. This is where those calls return: + */ + if (compressed) { + pixels = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, pixels, + unpack, + "glCompressedTexImage"); + } else { + pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, 1, + format, type, + pixels, unpack, "glTexImage"); + } + + LOCK_HARDWARE(intel); + + if (intelImage->mt) { + texImage->Data = intel_miptree_image_map(intel, + intelImage->mt, + intelImage->face, + intelImage->level, + &dstRowStride, + intelImage->base.ImageOffsets); + texImage->RowStride = dstRowStride / intelImage->mt->cpp; + } + else { + /* Allocate regular memory and store the image there temporarily. */ + if (texImage->IsCompressed) { + sizeInBytes = texImage->CompressedSize; + dstRowStride = + _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); + assert(dims != 3); + } + else { + dstRowStride = postConvWidth * texelBytes; + sizeInBytes = depth * dstRowStride * postConvHeight; + } + + texImage->Data = _mesa_alloc_texmemory(sizeInBytes); + } + + DBG("Upload image %dx%dx%d row_len %d " + "pitch %d\n", + width, height, depth, width * texelBytes, dstRowStride); + + /* Copy data. Would like to know when it's ok for us to eg. use + * the blitter to copy. Or, use the hardware to do the format + * conversion and copy: + */ + if (pixels) { + if (compressed) { + if (intelImage->mt) { + struct intel_region *dst = intelImage->mt->region; + _mesa_copy_rect(texImage->Data, dst->cpp, dst->pitch, + 0, 0, + intelImage->mt->level[level].width, + intelImage->mt->level[level].height/4, + pixels, + srcRowStride, + 0, 0); + } else + memcpy(texImage->Data, pixels, imageSize); + } else if (!texImage->TexFormat->StoreImage(ctx, dims, + texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */ + dstRowStride, + texImage->ImageOffsets, + width, height, depth, + format, type, pixels, unpack)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); + } + } + + /* GL_SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + intel_generate_mipmap(ctx, target, texObj); + } + + _mesa_unmap_teximage_pbo(ctx, unpack); + + if (intelImage->mt) { + intel_miptree_image_unmap(intel, intelImage->mt); + texImage->Data = NULL; + } + + UNLOCK_HARDWARE(intel); +} + +void +intelTexImage3D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint depth, + GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intelTexImage(ctx, 3, target, level, + internalFormat, width, height, depth, border, + format, type, pixels, unpack, texObj, texImage, 0, 0); +} + + +void +intelTexImage2D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intelTexImage(ctx, 2, target, level, + internalFormat, width, height, 1, border, + format, type, pixels, unpack, texObj, texImage, 0, 0); +} + +void +intelTexImage1D(GLcontext * ctx, + GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint border, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *unpack, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intelTexImage(ctx, 1, target, level, + internalFormat, width, 1, 1, border, + format, type, pixels, unpack, texObj, texImage, 0, 0); +} + +void intelCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage ) +{ + intelTexImage(ctx, 2, target, level, + internalFormat, width, height, 1, border, + 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, 1); +} + +/** + * Need to map texture image into memory before copying image data, + * then unmap it. + */ +static void +intel_get_tex_image(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage, int compressed) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_image *intelImage = intel_texture_image(texImage); + + /* Map */ + if (intelImage->mt) { + /* Image is stored in hardware format in a buffer managed by the + * kernel. Need to explicitly map and unmap it. + */ + intelImage->base.Data = + intel_miptree_image_map(intel, + intelImage->mt, + intelImage->face, + intelImage->level, + &intelImage->base.RowStride, + intelImage->base.ImageOffsets); + intelImage->base.RowStride /= intelImage->mt->cpp; + } + else { + /* Otherwise, the image should actually be stored in + * intelImage->base.Data. This is pretty confusing for + * everybody, I'd much prefer to separate the two functions of + * texImage->Data - storage for texture images in main memory + * and access (ie mappings) of images. In other words, we'd + * create a new texImage->Map field and leave Data simply for + * storage. + */ + assert(intelImage->base.Data); + } + + + if (compressed) { + _mesa_get_compressed_teximage(ctx, target, level, pixels, + texObj, texImage); + } else { + _mesa_get_teximage(ctx, target, level, format, type, pixels, + texObj, texImage); + } + + + /* Unmap */ + if (intelImage->mt) { + intel_miptree_image_unmap(intel, intelImage->mt); + intelImage->base.Data = NULL; + } +} + +void +intelGetTexImage(GLcontext * ctx, GLenum target, GLint level, + GLenum format, GLenum type, GLvoid * pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intel_get_tex_image(ctx, target, level, format, type, pixels, + texObj, texImage, 0); + + +} + +void +intelGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, + GLvoid *pixels, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intel_get_tex_image(ctx, target, level, 0, 0, pixels, + texObj, texImage, 1); +} + +void +intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname, + unsigned long long offset, GLint depth, GLuint pitch) +{ + struct intel_context *intel = pDRICtx->driverPrivate; + struct gl_texture_object *tObj = _mesa_lookup_texture(&intel->ctx, texname); + struct intel_texture_object *intelObj = intel_texture_object(tObj); + + if (!intelObj) + return; + + if (intelObj->mt) + intel_miptree_release(intel, &intelObj->mt); + + intelObj->imageOverride = GL_TRUE; + intelObj->depthOverride = depth; + intelObj->pitchOverride = pitch; + + if (offset) + intelObj->textureOffset = offset; +} + +void +intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv) +{ + struct intel_framebuffer *intel_fb = dPriv->driverPrivate; + struct intel_context *intel = pDRICtx->driverPrivate; + struct intel_texture_object *intelObj; + struct intel_texture_image *intelImage; + struct intel_mipmap_tree *mt; + struct intel_renderbuffer *rb; + struct gl_texture_unit *texUnit; + struct gl_texture_object *texObj; + struct gl_texture_image *texImage; + int level = 0, type, format, internalFormat; + + texUnit = &intel->ctx.Texture.Unit[intel->ctx.Texture.CurrentUnit]; + texObj = _mesa_select_tex_object(&intel->ctx, texUnit, target); + intelObj = intel_texture_object(texObj); + + if (!intelObj) + return; + + intel_update_renderbuffers(pDRICtx, dPriv); + + rb = intel_fb->color_rb[0]; + type = GL_BGRA; + format = GL_UNSIGNED_BYTE; + internalFormat = (rb->region->cpp == 3 ? 3 : 4); + + mt = intel_miptree_create_for_region(intel, target, + internalFormat, + 0, 0, rb->region, 1, 0); + if (mt == NULL) + return; + + _mesa_lock_texture(&intel->ctx, texObj); + + if (intelObj->mt) + intel_miptree_release(intel, &intelObj->mt); + + intelObj->mt = mt; + texImage = _mesa_get_tex_image(&intel->ctx, texObj, target, level); + _mesa_init_teximage_fields(&intel->ctx, target, texImage, + rb->region->pitch, rb->region->height, 1, + 0, internalFormat); + + intelImage = intel_texture_image(texImage); + intelImage->face = target_to_face(target); + intelImage->level = level; + texImage->TexFormat = intelChooseTextureFormat(&intel->ctx, internalFormat, + type, format); + _mesa_set_fetch_functions(texImage, 2); + texImage->RowStride = rb->region->pitch; + intel_miptree_reference(&intelImage->mt, intelObj->mt); + + if (!intel_miptree_match_image(intelObj->mt, &intelImage->base, + intelImage->face, intelImage->level)) { + fprintf(stderr, "miptree doesn't match image\n"); + } + + _mesa_unlock_texture(&intel->ctx, texObj); +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.c b/src/mesa/drivers/dri/intel/intel_tex_layout.c index fcb5cc3906..edc3a2eaa4 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_layout.c +++ b/src/mesa/drivers/dri/intel/intel_tex_layout.c @@ -32,15 +32,27 @@ #include "intel_mipmap_tree.h" #include "intel_tex_layout.h" +#include "intel_context.h" #include "macros.h" - -static int align(int value, int alignment) +GLuint intel_compressed_alignment(GLenum internalFormat) { - return (value + alignment - 1) & ~(alignment - 1); + GLuint alignment = 4; + + switch (internalFormat) { + case GL_COMPRESSED_RGB_FXT1_3DFX: + case GL_COMPRESSED_RGBA_FXT1_3DFX: + alignment = 8; + break; + + default: + break; + } + + return alignment; } -void i945_miptree_layout_2d( struct intel_mipmap_tree *mt ) +void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt ) { GLint align_h = 2, align_w = 4; GLuint level; @@ -51,23 +63,36 @@ void i945_miptree_layout_2d( struct intel_mipmap_tree *mt ) mt->pitch = mt->width0; + if (mt->compressed) { + align_w = intel_compressed_alignment(mt->internal_format); + mt->pitch = ALIGN(mt->width0, align_w); + } + /* May need to adjust pitch to accomodate the placement of * the 2nd mipmap. This occurs when the alignment * constraints of mipmap placement push the right edge of the * 2nd mipmap out past the width of its parent. */ if (mt->first_level != mt->last_level) { - GLuint mip1_width = align(minify(mt->width0), align_w) - + minify(minify(mt->width0)); + GLuint mip1_width; + + if (mt->compressed) { + mip1_width = ALIGN(minify(mt->width0), align_w) + + ALIGN(minify(minify(mt->width0)), align_w); + } else { + mip1_width = ALIGN(minify(mt->width0), align_w) + + minify(minify(mt->width0)); + } - if (mip1_width > mt->width0) - mt->pitch = mip1_width; + if (mip1_width > mt->pitch) { + mt->pitch = mip1_width; + } } /* Pitch must be a whole number of dwords, even though we * express it in texels. */ - mt->pitch = align(mt->pitch * mt->cpp, 4) / mt->cpp; + mt->pitch = intel_miptree_pitch_align (intel, mt, mt->pitch); mt->total_height = 0; for ( level = mt->first_level ; level <= mt->last_level ; level++ ) { @@ -79,7 +104,7 @@ void i945_miptree_layout_2d( struct intel_mipmap_tree *mt ) if (mt->compressed) img_height = MAX2(1, height/4); else - img_height = align(height, align_h); + img_height = ALIGN(height, align_h); /* Because the images are packed better, the final offset @@ -90,7 +115,7 @@ void i945_miptree_layout_2d( struct intel_mipmap_tree *mt ) /* Layout_below: step right after second mipmap. */ if (level == mt->first_level + 1) { - x += align(width, align_w); + x += ALIGN(width, align_w); } else { y += img_height; diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.h b/src/mesa/drivers/dri/intel/intel_tex_layout.h index 1e37f8f525..193699d3f7 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_layout.h +++ b/src/mesa/drivers/dri/intel/intel_tex_layout.h @@ -38,4 +38,5 @@ static GLuint minify( GLuint d ) return MAX2(1, d>>1); } -extern void i945_miptree_layout_2d( struct intel_mipmap_tree *mt ); +extern void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt ); +extern GLuint intel_compressed_alignment(GLenum); diff --git a/src/mesa/drivers/dri/intel/intel_tex_obj.h b/src/mesa/drivers/dri/intel/intel_tex_obj.h new file mode 100644 index 0000000000..5a93461525 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_obj.h @@ -0,0 +1,83 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef _INTEL_TEX_OBJ_H +#define _INTEL_TEX_OBJ_H + +struct intel_texture_object +{ + struct gl_texture_object base; /* The "parent" object */ + + /* The mipmap tree must include at least these levels once + * validated: + */ + GLuint firstLevel; + GLuint lastLevel; + + /* Offset for firstLevel image: + */ + GLuint textureOffset; + + /* On validation any active images held in main memory or in other + * regions will be copied to this region and the old storage freed. + */ + struct intel_mipmap_tree *mt; + + GLboolean imageOverride; + GLint depthOverride; + GLuint pitchOverride; +}; + +struct intel_texture_image +{ + struct gl_texture_image base; + + /* These aren't stored in gl_texture_image + */ + GLuint level; + GLuint face; + + /* If intelImage->mt != NULL, image data is stored here. + * Else if intelImage->base.Data != NULL, image is stored there. + * Else there is no image data. + */ + struct intel_mipmap_tree *mt; +}; + +static INLINE struct intel_texture_object * +intel_texture_object(struct gl_texture_object *obj) +{ + return (struct intel_texture_object *) obj; +} + +static INLINE struct intel_texture_image * +intel_texture_image(struct gl_texture_image *img) +{ + return (struct intel_texture_image *) img; +} + +#endif /* _INTEL_TEX_OBJ_H */ diff --git a/src/mesa/drivers/dri/intel/intel_tex_subimage.c b/src/mesa/drivers/dri/intel/intel_tex_subimage.c new file mode 100644 index 0000000000..5428a1d068 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_subimage.c @@ -0,0 +1,186 @@ + +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "mtypes.h" +#include "texobj.h" +#include "texstore.h" +#include "texcompress.h" +#include "enums.h" + +#include "intel_context.h" +#include "intel_tex.h" +#include "intel_mipmap_tree.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +static void +intelTexSubimage(GLcontext * ctx, + GLint dims, + GLenum target, GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLint width, GLint height, GLint depth, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_image *intelImage = intel_texture_image(texImage); + GLuint dstRowStride = 0; + + DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__, + _mesa_lookup_enum_by_nr(target), + level, xoffset, yoffset, width, height); + + intelFlush(ctx); + + pixels = + _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format, + type, pixels, packing, "glTexSubImage2D"); + if (!pixels) + return; + + LOCK_HARDWARE(intel); + + /* Map buffer if necessary. Need to lock to prevent other contexts + * from uploading the buffer under us. + */ + if (intelImage->mt) + texImage->Data = intel_miptree_image_map(intel, + intelImage->mt, + intelImage->face, + intelImage->level, + &dstRowStride, + texImage->ImageOffsets); + else { + if (texImage->IsCompressed) { + dstRowStride = + _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); + assert(dims != 3); + } + else { + dstRowStride = texImage->RowStride * texImage->TexFormat->TexelBytes; + } + } + + assert(dstRowStride); + + if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, + xoffset, yoffset, zoffset, + dstRowStride, + texImage->ImageOffsets, + width, height, depth, + format, type, pixels, packing)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage"); + } + + /* GL_SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + intel_generate_mipmap(ctx, target, texObj); + } + + _mesa_unmap_teximage_pbo(ctx, packing); + + if (intelImage->mt) { + intel_miptree_image_unmap(intel, intelImage->mt); + texImage->Data = NULL; + } + + UNLOCK_HARDWARE(intel); +} + + + + + +void +intelTexSubImage3D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, GLint yoffset, GLint zoffset, + GLsizei width, GLsizei height, GLsizei depth, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + + intelTexSubimage(ctx, 3, + target, level, + xoffset, yoffset, zoffset, + width, height, depth, + format, type, pixels, packing, texObj, texImage); + +} + + + +void +intelTexSubImage2D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + + intelTexSubimage(ctx, 2, + target, level, + xoffset, yoffset, 0, + width, height, 1, + format, type, pixels, packing, texObj, texImage); + +} + + +void +intelTexSubImage1D(GLcontext * ctx, + GLenum target, + GLint level, + GLint xoffset, + GLsizei width, + GLenum format, GLenum type, + const GLvoid * pixels, + const struct gl_pixelstore_attrib *packing, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage) +{ + intelTexSubimage(ctx, 1, + target, level, + xoffset, 0, 0, + width, 1, 1, + format, type, pixels, packing, texObj, texImage); + +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c new file mode 100644 index 0000000000..5763f4ae1f --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c @@ -0,0 +1,300 @@ +#include "mtypes.h" +#include "macros.h" + +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_mipmap_tree.h" +#include "intel_tex.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +/** + * Compute which mipmap levels that really need to be sent to the hardware. + * This depends on the base image size, GL_TEXTURE_MIN_LOD, + * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL. + */ +static void +intel_calculate_first_last_level(struct intel_texture_object *intelObj) +{ + struct gl_texture_object *tObj = &intelObj->base; + const struct gl_texture_image *const baseImage = + tObj->Image[0][tObj->BaseLevel]; + + /* These must be signed values. MinLod and MaxLod can be negative numbers, + * and having firstLevel and lastLevel as signed prevents the need for + * extra sign checks. + */ + int firstLevel; + int lastLevel; + + /* Yes, this looks overly complicated, but it's all needed. + */ + switch (tObj->Target) { + case GL_TEXTURE_1D: + case GL_TEXTURE_2D: + case GL_TEXTURE_3D: + case GL_TEXTURE_CUBE_MAP: + if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) { + /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL. + */ + firstLevel = lastLevel = tObj->BaseLevel; + } + else { +#ifdef I915 + firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5); + firstLevel = MAX2(firstLevel, tObj->BaseLevel); + firstLevel = MIN2(firstLevel, tObj->BaseLevel + baseImage->MaxLog2); + lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5); + lastLevel = MAX2(lastLevel, tObj->BaseLevel); + lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2); + lastLevel = MIN2(lastLevel, tObj->MaxLevel); + lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */ +#else + /* Currently not taking min/max lod into account here, those + * values are programmed as sampler state elsewhere and we + * upload the same mipmap levels regardless. Not sure if + * this makes sense as it means it isn't possible for the app + * to use min/max lod to reduce texture memory pressure: + */ + firstLevel = tObj->BaseLevel; + lastLevel = MIN2(tObj->BaseLevel + baseImage->MaxLog2, + tObj->MaxLevel); + lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */ +#endif + } + break; + case GL_TEXTURE_RECTANGLE_NV: + case GL_TEXTURE_4D_SGIS: + firstLevel = lastLevel = 0; + break; + default: + return; + } + + /* save these values */ + intelObj->firstLevel = firstLevel; + intelObj->lastLevel = lastLevel; +} + +/** + * Copies the image's contents at its level into the object's miptree, + * and updates the image to point at the object's miptree. + */ +static void +copy_image_data_to_tree(struct intel_context *intel, + struct intel_texture_object *intelObj, + struct intel_texture_image *intelImage) +{ + if (intelImage->mt) { + /* Copy potentially with the blitter: + */ + intel_miptree_image_copy(intel, + intelObj->mt, + intelImage->face, + intelImage->level, intelImage->mt); + + intel_miptree_release(intel, &intelImage->mt); + } + else { + assert(intelImage->base.Data != NULL); + + /* More straightforward upload. + */ + intel_miptree_image_data(intel, + intelObj->mt, + intelImage->face, + intelImage->level, + intelImage->base.Data, + intelImage->base.RowStride, + intelImage->base.RowStride * + intelImage->base.Height); + _mesa_align_free(intelImage->base.Data); + intelImage->base.Data = NULL; + } + + intel_miptree_reference(&intelImage->mt, intelObj->mt); +} + + +/* + */ +GLuint +intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit) +{ + struct gl_texture_object *tObj = intel->ctx.Texture.Unit[unit]._Current; + struct intel_texture_object *intelObj = intel_texture_object(tObj); + int comp_byte = 0; + int cpp; + GLuint face, i; + GLuint nr_faces = 0; + struct intel_texture_image *firstImage; + + /* We know/require this is true by now: + */ + assert(intelObj->base._Complete); + + /* What levels must the tree include at a minimum? + */ + intel_calculate_first_last_level(intelObj); + firstImage = + intel_texture_image(intelObj->base.Image[0][intelObj->firstLevel]); + + /* Fallback case: + */ + if (firstImage->base.Border || + ((firstImage->base._BaseFormat == GL_DEPTH_COMPONENT) && + ((tObj->WrapS == GL_CLAMP_TO_BORDER) || + (tObj->WrapT == GL_CLAMP_TO_BORDER)))) { + if (intelObj->mt) { + intel_miptree_release(intel, &intelObj->mt); + } + return GL_FALSE; + } + + + /* If both firstImage and intelObj have a tree which can contain + * all active images, favour firstImage. Note that because of the + * completeness requirement, we know that the image dimensions + * will match. + */ + if (firstImage->mt && + firstImage->mt != intelObj->mt && + firstImage->mt->first_level <= intelObj->firstLevel && + firstImage->mt->last_level >= intelObj->lastLevel) { + + if (intelObj->mt) + intel_miptree_release(intel, &intelObj->mt); + + intel_miptree_reference(&intelObj->mt, firstImage->mt); + } + + if (firstImage->base.IsCompressed) { + comp_byte = intel_compressed_num_bytes(firstImage->base.TexFormat->MesaFormat); + cpp = comp_byte; + } + else cpp = firstImage->base.TexFormat->TexelBytes; + + /* Check tree can hold all active levels. Check tree matches + * target, imageFormat, etc. + * + * XXX: For some layouts (eg i945?), the test might have to be + * first_level == firstLevel, as the tree isn't valid except at the + * original start level. Hope to get around this by + * programming minLod, maxLod, baseLevel into the hardware and + * leaving the tree alone. + */ + if (intelObj->mt && + (intelObj->mt->target != intelObj->base.Target || + intelObj->mt->internal_format != firstImage->base.InternalFormat || + intelObj->mt->first_level != intelObj->firstLevel || + intelObj->mt->last_level != intelObj->lastLevel || + intelObj->mt->width0 != firstImage->base.Width || + intelObj->mt->height0 != firstImage->base.Height || + intelObj->mt->depth0 != firstImage->base.Depth || + intelObj->mt->cpp != cpp || + intelObj->mt->compressed != firstImage->base.IsCompressed)) { + intel_miptree_release(intel, &intelObj->mt); + } + + + /* May need to create a new tree: + */ + if (!intelObj->mt) { + intelObj->mt = intel_miptree_create(intel, + intelObj->base.Target, + firstImage->base.InternalFormat, + intelObj->firstLevel, + intelObj->lastLevel, + firstImage->base.Width, + firstImage->base.Height, + firstImage->base.Depth, + cpp, + comp_byte); + } + + /* Pull in any images not in the object's tree: + */ + nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + for (face = 0; face < nr_faces; face++) { + for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) { + struct intel_texture_image *intelImage = + intel_texture_image(intelObj->base.Image[face][i]); + + /* Need to import images in main memory or held in other trees. + */ + if (intelObj->mt != intelImage->mt) { + copy_image_data_to_tree(intel, intelObj, intelImage); + } + } + } + + return GL_TRUE; +} + +void +intel_tex_map_level_images(struct intel_context *intel, + struct intel_texture_object *intelObj, + int level) +{ + GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + GLuint face; + + for (face = 0; face < nr_faces; face++) { + struct intel_texture_image *intelImage = + intel_texture_image(intelObj->base.Image[face][level]); + + if (intelImage->mt) { + intelImage->base.Data = + intel_miptree_image_map(intel, + intelImage->mt, + intelImage->face, + intelImage->level, + &intelImage->base.RowStride, + intelImage->base.ImageOffsets); + /* convert stride to texels, not bytes */ + intelImage->base.RowStride /= intelImage->mt->cpp; + /* intelImage->base.ImageStride /= intelImage->mt->cpp; */ + } + } +} + +void +intel_tex_unmap_level_images(struct intel_context *intel, + struct intel_texture_object *intelObj, + int level) +{ + GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + GLuint face; + + for (face = 0; face < nr_faces; face++) { + struct intel_texture_image *intelImage = + intel_texture_image(intelObj->base.Image[face][level]); + + if (intelImage->mt) { + intel_miptree_image_unmap(intel, intelImage->mt); + intelImage->base.Data = NULL; + } + } +} + +void +intel_tex_map_images(struct intel_context *intel, + struct intel_texture_object *intelObj) +{ + int i; + + DBG("%s\n", __FUNCTION__); + + for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) + intel_tex_map_level_images(intel, intelObj, i); +} + +void +intel_tex_unmap_images(struct intel_context *intel, + struct intel_texture_object *intelObj) +{ + int i; + + for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) + intel_tex_unmap_level_images(intel, intelObj, i); +} diff --git a/src/mesa/drivers/dri/intel/server/i830_dri.h b/src/mesa/drivers/dri/intel/server/i830_dri.h new file mode 100644 index 0000000000..def049e7a6 --- /dev/null +++ b/src/mesa/drivers/dri/intel/server/i830_dri.h @@ -0,0 +1,62 @@ +/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/i810/i830_dri.h,v 1.6 2003/09/28 20:15:59 alanh Exp $ */ + +#ifndef _I830_DRI_H +#define _I830_DRI_H + +#include "xf86drm.h" + +#define I830_MAX_DRAWABLES 256 + +#define I830_MAJOR_VERSION 1 +#define I830_MINOR_VERSION 9 +#define I830_PATCHLEVEL 0 + +#define I830_REG_SIZE 0x80000 + +typedef struct _I830DRIRec { + drm_handle_t regs; + drmSize regsSize; + + drmSize unused1; /* backbufferSize */ + drm_handle_t unused2; /* backbuffer */ + + drmSize unused3; /* depthbufferSize */ + drm_handle_t unused4; /* depthbuffer */ + + drmSize unused5; /* rotatedSize */ + drm_handle_t unused6; /* rotatedbuffer */ + + drm_handle_t unused7; /* textures */ + int unused8; /* textureSize */ + + drm_handle_t unused9; /* agp_buffers */ + drmSize unused10; /* agp_buf_size */ + + int deviceID; + int width; + int height; + int mem; + int cpp; + int bitsPerPixel; + + int unused11[8]; /* was front/back/depth/rotated offset/pitch */ + + int unused12; /* logTextureGranularity */ + int unused13; /* textureOffset */ + + int irq; + int sarea_priv_offset; +} I830DRIRec, *I830DRIPtr; + +typedef struct { + /* Nothing here yet */ + int dummy; +} I830ConfigPrivRec, *I830ConfigPrivPtr; + +typedef struct { + /* Nothing here yet */ + int dummy; +} I830DRIContextRec, *I830DRIContextPtr; + + +#endif diff --git a/src/mesa/drivers/dri/intel/server/intel.h b/src/mesa/drivers/dri/intel/server/intel.h new file mode 100644 index 0000000000..6ea72499c1 --- /dev/null +++ b/src/mesa/drivers/dri/intel/server/intel.h @@ -0,0 +1,331 @@ +#ifndef _INTEL_H_ +#define _INTEL_H_ + +#include "xf86drm.h" /* drm_handle_t, etc */ + +/* Intel */ +#ifndef PCI_CHIP_I810 +#define PCI_CHIP_I810 0x7121 +#define PCI_CHIP_I810_DC100 0x7123 +#define PCI_CHIP_I810_E 0x7125 +#define PCI_CHIP_I815 0x1132 +#define PCI_CHIP_I810_BRIDGE 0x7120 +#define PCI_CHIP_I810_DC100_BRIDGE 0x7122 +#define PCI_CHIP_I810_E_BRIDGE 0x7124 +#define PCI_CHIP_I815_BRIDGE 0x1130 +#endif + +#define PCI_CHIP_845_G 0x2562 +#define PCI_CHIP_I830_M 0x3577 + +#ifndef PCI_CHIP_I855_GM +#define PCI_CHIP_I855_GM 0x3582 +#define PCI_CHIP_I855_GM_BRIDGE 0x3580 +#endif + +#ifndef PCI_CHIP_I865_G +#define PCI_CHIP_I865_G 0x2572 +#define PCI_CHIP_I865_G_BRIDGE 0x2570 +#endif + +#ifndef PCI_CHIP_I915_G +#define PCI_CHIP_I915_G 0x2582 +#define PCI_CHIP_I915_G_BRIDGE 0x2580 +#endif + +#ifndef PCI_CHIP_I915_GM +#define PCI_CHIP_I915_GM 0x2592 +#define PCI_CHIP_I915_GM_BRIDGE 0x2590 +#endif + +#ifndef PCI_CHIP_E7221_G +#define PCI_CHIP_E7221_G 0x258A +/* Same as I915_G_BRIDGE */ +#define PCI_CHIP_E7221_G_BRIDGE 0x2580 +#endif + +#ifndef PCI_CHIP_I945_G +#define PCI_CHIP_I945_G 0x2772 +#define PCI_CHIP_I945_G_BRIDGE 0x2770 +#endif + +#ifndef PCI_CHIP_I945_GM +#define PCI_CHIP_I945_GM 0x27A2 +#define PCI_CHIP_I945_GM_BRIDGE 0x27A0 +#endif + +#define IS_I810(pI810) (pI810->Chipset == PCI_CHIP_I810 || \ + pI810->Chipset == PCI_CHIP_I810_DC100 || \ + pI810->Chipset == PCI_CHIP_I810_E) +#define IS_I815(pI810) (pI810->Chipset == PCI_CHIP_I815) +#define IS_I830(pI810) (pI810->Chipset == PCI_CHIP_I830_M) +#define IS_845G(pI810) (pI810->Chipset == PCI_CHIP_845_G) +#define IS_I85X(pI810) (pI810->Chipset == PCI_CHIP_I855_GM) +#define IS_I852(pI810) (pI810->Chipset == PCI_CHIP_I855_GM && (pI810->variant == I852_GM || pI810->variant == I852_GME)) +#define IS_I855(pI810) (pI810->Chipset == PCI_CHIP_I855_GM && (pI810->variant == I855_GM || pI810->variant == I855_GME)) +#define IS_I865G(pI810) (pI810->Chipset == PCI_CHIP_I865_G) + +#define IS_I915G(pI810) (pI810->Chipset == PCI_CHIP_I915_G || pI810->Chipset == PCI_CHIP_E7221_G) +#define IS_I915GM(pI810) (pI810->Chipset == PCI_CHIP_I915_GM) +#define IS_I945G(pI810) (pI810->Chipset == PCI_CHIP_I945_G) +#define IS_I945GM(pI810) (pI810->Chipset == PCI_CHIP_I945_GM) +#define IS_I9XX(pI810) (IS_I915G(pI810) || IS_I915GM(pI810) || IS_I945G(pI810) || IS_I945GM(pI810)) + +#define IS_MOBILE(pI810) (IS_I830(pI810) || IS_I85X(pI810) || IS_I915GM(pI810) || IS_I945GM(pI810)) + +#define I830_GMCH_CTRL 0x52 + +#define I830_GMCH_MEM_MASK 0x1 +#define I830_GMCH_MEM_64M 0x1 +#define I830_GMCH_MEM_128M 0 + +#define I830_GMCH_GMS_MASK 0x70 +#define I830_GMCH_GMS_DISABLED 0x00 +#define I830_GMCH_GMS_LOCAL 0x10 +#define I830_GMCH_GMS_STOLEN_512 0x20 +#define I830_GMCH_GMS_STOLEN_1024 0x30 +#define I830_GMCH_GMS_STOLEN_8192 0x40 + +#define I855_GMCH_GMS_MASK (0x7 << 4) +#define I855_GMCH_GMS_DISABLED 0x00 +#define I855_GMCH_GMS_STOLEN_1M (0x1 << 4) +#define I855_GMCH_GMS_STOLEN_4M (0x2 << 4) +#define I855_GMCH_GMS_STOLEN_8M (0x3 << 4) +#define I855_GMCH_GMS_STOLEN_16M (0x4 << 4) +#define I855_GMCH_GMS_STOLEN_32M (0x5 << 4) +#define I915G_GMCH_GMS_STOLEN_48M (0x6 << 4) +#define I915G_GMCH_GMS_STOLEN_64M (0x7 << 4) + +typedef unsigned char Bool; +#define TRUE 1 +#define FALSE 0 + +#define PIPE_NONE 0<<0 +#define PIPE_CRT 1<<0 +#define PIPE_TV 1<<1 +#define PIPE_DFP 1<<2 +#define PIPE_LFP 1<<3 +#define PIPE_CRT2 1<<4 +#define PIPE_TV2 1<<5 +#define PIPE_DFP2 1<<6 +#define PIPE_LFP2 1<<7 + +typedef struct _I830MemPool *I830MemPoolPtr; +typedef struct _I830MemRange *I830MemRangePtr; +typedef struct _I830MemRange { + long Start; + long End; + long Size; + unsigned long Physical; + unsigned long Offset; /* Offset of AGP-allocated portion */ + unsigned long Alignment; + drm_handle_t Key; + unsigned long Pitch; // add pitch + I830MemPoolPtr Pool; +} I830MemRange; + +typedef struct _I830MemPool { + I830MemRange Total; + I830MemRange Free; + I830MemRange Fixed; + I830MemRange Allocated; +} I830MemPool; + +typedef struct { + int tail_mask; + I830MemRange mem; + unsigned char *virtual_start; + int head; + int tail; + int space; +} I830RingBuffer; + +typedef struct _I830Rec { + unsigned char *MMIOBase; + unsigned char *FbBase; + int cpp; + uint32_t aper_size; + unsigned int bios_version; + + /* These are set in PreInit and never changed. */ + long FbMapSize; + long TotalVideoRam; + I830MemRange StolenMemory; /* pre-allocated memory */ + long BIOSMemorySize; /* min stolen pool size */ + int BIOSMemSizeLoc; + + /* These change according to what has been allocated. */ + long FreeMemory; + I830MemRange MemoryAperture; + I830MemPool StolenPool; + long allocatedMemory; + + /* Regions allocated either from the above pools, or from agpgart. */ + /* for single and dual head configurations */ + I830MemRange FrontBuffer; + I830MemRange FrontBuffer2; + I830MemRange Scratch; + I830MemRange Scratch2; + + I830RingBuffer *LpRing; + + I830MemRange BackBuffer; + I830MemRange DepthBuffer; + I830MemRange TexMem; + int TexGranularity; + I830MemRange ContextMem; + int drmMinor; + Bool have3DWindows; + + Bool NeedRingBufferLow; + Bool allowPageFlip; + Bool disableTiling; + + int Chipset; + unsigned long LinearAddr; + unsigned long MMIOAddr; + + drmSize registerSize; /**< \brief MMIO register map size */ + drm_handle_t registerHandle; /**< \brief MMIO register map handle */ + // IOADDRESS ioBase; + int irq; /**< \brief IRQ number */ + int GttBound; + + drm_handle_t ring_map; + unsigned int Fence[8]; + +} I830Rec; + +/* + * 12288 is set as the maximum, chosen because it is enough for + * 1920x1440@32bpp with a 2048 pixel line pitch with some to spare. + */ +#define I830_MAXIMUM_VBIOS_MEM 12288 +#define I830_DEFAULT_VIDEOMEM_2D (MB(32) / 1024) +#define I830_DEFAULT_VIDEOMEM_3D (MB(64) / 1024) + +/* Flags for memory allocation function */ +#define FROM_ANYWHERE 0x00000000 +#define FROM_POOL_ONLY 0x00000001 +#define FROM_NEW_ONLY 0x00000002 +#define FROM_MASK 0x0000000f + +#define ALLOCATE_AT_TOP 0x00000010 +#define ALLOCATE_AT_BOTTOM 0x00000020 +#define FORCE_GAPS 0x00000040 + +#define NEED_PHYSICAL_ADDR 0x00000100 +#define ALIGN_BOTH_ENDS 0x00000200 +#define FORCE_LOW 0x00000400 + +#define ALLOC_NO_TILING 0x00001000 +#define ALLOC_INITIAL 0x00002000 + +#define ALLOCATE_DRY_RUN 0x80000000 + +/* Chipset registers for VIDEO BIOS memory RW access */ +#define _855_DRAM_RW_CONTROL 0x58 +#define _845_DRAM_RW_CONTROL 0x90 +#define DRAM_WRITE 0x33330000 + +#define KB(x) ((x) * 1024) +#define MB(x) ((x) * KB(1024)) + +#define GTT_PAGE_SIZE KB(4) +#define ROUND_TO(x, y) (((x) + (y) - 1) / (y) * (y)) +#define ROUND_DOWN_TO(x, y) ((x) / (y) * (y)) +#define ROUND_TO_PAGE(x) ROUND_TO((x), GTT_PAGE_SIZE) +#define ROUND_TO_MB(x) ROUND_TO((x), MB(1)) +#define PRIMARY_RINGBUFFER_SIZE KB(128) + + +/* Ring buffer registers, p277, overview p19 + */ +#define LP_RING 0x2030 +#define HP_RING 0x2040 + +#define RING_TAIL 0x00 +#define TAIL_ADDR 0x000FFFF8 +#define I830_TAIL_MASK 0x001FFFF8 + +#define RING_HEAD 0x04 +#define HEAD_WRAP_COUNT 0xFFE00000 +#define HEAD_WRAP_ONE 0x00200000 +#define HEAD_ADDR 0x001FFFFC +#define I830_HEAD_MASK 0x001FFFFC + +#define RING_START 0x08 +#define START_ADDR 0x03FFFFF8 +#define I830_RING_START_MASK 0xFFFFF000 + +#define RING_LEN 0x0C +#define RING_NR_PAGES 0x001FF000 +#define I830_RING_NR_PAGES 0x001FF000 +#define RING_REPORT_MASK 0x00000006 +#define RING_REPORT_64K 0x00000002 +#define RING_REPORT_128K 0x00000004 +#define RING_NO_REPORT 0x00000000 +#define RING_VALID_MASK 0x00000001 +#define RING_VALID 0x00000001 +#define RING_INVALID 0x00000000 + + +/* Fence/Tiling ranges [0..7] + */ +#define FENCE 0x2000 +#define FENCE_NR 8 + +#define I915G_FENCE_START_MASK 0x0ff00000 + +#define I830_FENCE_START_MASK 0x07f80000 + +#define FENCE_START_MASK 0x03F80000 +#define FENCE_X_MAJOR 0x00000000 +#define FENCE_Y_MAJOR 0x00001000 +#define FENCE_SIZE_MASK 0x00000700 +#define FENCE_SIZE_512K 0x00000000 +#define FENCE_SIZE_1M 0x00000100 +#define FENCE_SIZE_2M 0x00000200 +#define FENCE_SIZE_4M 0x00000300 +#define FENCE_SIZE_8M 0x00000400 +#define FENCE_SIZE_16M 0x00000500 +#define FENCE_SIZE_32M 0x00000600 +#define FENCE_SIZE_64M 0x00000700 +#define I915G_FENCE_SIZE_1M 0x00000000 +#define I915G_FENCE_SIZE_2M 0x00000100 +#define I915G_FENCE_SIZE_4M 0x00000200 +#define I915G_FENCE_SIZE_8M 0x00000300 +#define I915G_FENCE_SIZE_16M 0x00000400 +#define I915G_FENCE_SIZE_32M 0x00000500 +#define I915G_FENCE_SIZE_64M 0x00000600 +#define I915G_FENCE_SIZE_128M 0x00000700 +#define FENCE_PITCH_1 0x00000000 +#define FENCE_PITCH_2 0x00000010 +#define FENCE_PITCH_4 0x00000020 +#define FENCE_PITCH_8 0x00000030 +#define FENCE_PITCH_16 0x00000040 +#define FENCE_PITCH_32 0x00000050 +#define FENCE_PITCH_64 0x00000060 +#define FENCE_VALID 0x00000001 + +#include <mmio.h> + +# define MMIO_IN8(base, offset) \ + *(volatile unsigned char *)(((unsigned char*)(base)) + (offset)) +# define MMIO_IN32(base, offset) \ + read_MMIO_LE32(base, offset) +# define MMIO_OUT8(base, offset, val) \ + *(volatile unsigned char *)(((unsigned char*)(base)) + (offset)) = (val) +# define MMIO_OUT32(base, offset, val) \ + *(volatile unsigned int *)(void *)(((unsigned char*)(base)) + (offset)) = CPU_TO_LE32(val) + + + /* Memory mapped register access macros */ +#define INREG8(addr) MMIO_IN8(MMIO, addr) +#define INREG(addr) MMIO_IN32(MMIO, addr) +#define OUTREG8(addr, val) MMIO_OUT8(MMIO, addr, val) +#define OUTREG(addr, val) MMIO_OUT32(MMIO, addr, val) + +#define DSPABASE 0x70184 + +#endif diff --git a/src/mesa/drivers/dri/intel/server/intel_dri.c b/src/mesa/drivers/dri/intel/server/intel_dri.c new file mode 100644 index 0000000000..e49c4214ad --- /dev/null +++ b/src/mesa/drivers/dri/intel/server/intel_dri.c @@ -0,0 +1,1306 @@ +/** + * \file server/intel_dri.c + * \brief File to perform the device-specific initialization tasks typically + * done in the X server. + * + * Here they are converted to run in the client (or perhaps a standalone + * process), and to work with the frame buffer device rather than the X + * server infrastructure. + * + * Copyright (C) 2006 Dave Airlie (airlied@linux.ie) + + Permission is hereby granted, free of charge, to any person obtaining a + copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sub license, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial portions + of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT HOLDERS AND/OR THEIR SUPPLIERS BE LIABLE FOR + ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +*/ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <errno.h> +#include <unistd.h> + +#include "driver.h" +#include "drm.h" + +#include "intel.h" +#include "i830_dri.h" + +#include "memops.h" +#include "pciaccess.h" + +static size_t drm_page_size; +static int nextTile = 0; +#define xf86DrvMsg(...) do {} while(0) + +static const int pitches[] = { + 128 * 8, + 128 * 16, + 128 * 32, + 128 * 64, + 0 +}; + +static Bool I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea); + +static unsigned long +GetBestTileAlignment(unsigned long size) +{ + unsigned long i; + + for (i = KB(512); i < size; i <<= 1) + ; + + if (i > MB(64)) + i = MB(64); + + return i; +} + +static void SetFenceRegs(const DRIDriverContext *ctx, I830Rec *pI830) +{ + int i; + unsigned char *MMIO = ctx->MMIOAddress; + + for (i = 0; i < 8; i++) { + OUTREG(FENCE + i * 4, pI830->Fence[i]); + // if (I810_DEBUG & DEBUG_VERBOSE_VGA) + fprintf(stderr,"Fence Register : %x\n", pI830->Fence[i]); + } +} + +/* Tiled memory is good... really, really good... + * + * Need to make it less likely that we miss out on this - probably + * need to move the frontbuffer away from the 'guarenteed' alignment + * of the first memory segment, or perhaps allocate a discontigous + * framebuffer to get more alignment 'sweet spots'. + */ +static void +SetFence(const DRIDriverContext *ctx, I830Rec *pI830, + int nr, unsigned int start, unsigned int pitch, + unsigned int size) +{ + unsigned int val; + unsigned int fence_mask = 0; + unsigned int fence_pitch; + + if (nr < 0 || nr > 7) { + fprintf(stderr, + "SetFence: fence %d out of range\n",nr); + return; + } + + pI830->Fence[nr] = 0; + + if (IS_I9XX(pI830)) + fence_mask = ~I915G_FENCE_START_MASK; + else + fence_mask = ~I830_FENCE_START_MASK; + + if (start & fence_mask) { + fprintf(stderr, + "SetFence: %d: start (0x%08x) is not %s aligned\n", + nr, start, (IS_I9XX(pI830)) ? "1MB" : "512k"); + return; + } + + if (start % size) { + fprintf(stderr, + "SetFence: %d: start (0x%08x) is not size (%dk) aligned\n", + nr, start, size / 1024); + return; + } + + if (pitch & 127) { + fprintf(stderr, + "SetFence: %d: pitch (%d) not a multiple of 128 bytes\n", + nr, pitch); + return; + } + + val = (start | FENCE_X_MAJOR | FENCE_VALID); + + if (IS_I9XX(pI830)) { + switch (size) { + case MB(1): + val |= I915G_FENCE_SIZE_1M; + break; + case MB(2): + val |= I915G_FENCE_SIZE_2M; + break; + case MB(4): + val |= I915G_FENCE_SIZE_4M; + break; + case MB(8): + val |= I915G_FENCE_SIZE_8M; + break; + case MB(16): + val |= I915G_FENCE_SIZE_16M; + break; + case MB(32): + val |= I915G_FENCE_SIZE_32M; + break; + case MB(64): + val |= I915G_FENCE_SIZE_64M; + break; + default: + fprintf(stderr, + "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024); + return; + } + } else { + switch (size) { + case KB(512): + val |= FENCE_SIZE_512K; + break; + case MB(1): + val |= FENCE_SIZE_1M; + break; + case MB(2): + val |= FENCE_SIZE_2M; + break; + case MB(4): + val |= FENCE_SIZE_4M; + break; + case MB(8): + val |= FENCE_SIZE_8M; + break; + case MB(16): + val |= FENCE_SIZE_16M; + break; + case MB(32): + val |= FENCE_SIZE_32M; + break; + case MB(64): + val |= FENCE_SIZE_64M; + break; + default: + fprintf(stderr, + "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024); + return; + } + } + + if (IS_I9XX(pI830)) + fence_pitch = pitch / 512; + else + fence_pitch = pitch / 128; + + switch (fence_pitch) { + case 1: + val |= FENCE_PITCH_1; + break; + case 2: + val |= FENCE_PITCH_2; + break; + case 4: + val |= FENCE_PITCH_4; + break; + case 8: + val |= FENCE_PITCH_8; + break; + case 16: + val |= FENCE_PITCH_16; + break; + case 32: + val |= FENCE_PITCH_32; + break; + case 64: + val |= FENCE_PITCH_64; + break; + default: + fprintf(stderr, + "SetFence: %d: illegal pitch (%d)\n", nr, pitch); + return; + } + + pI830->Fence[nr] = val; +} + +static Bool +MakeTiles(const DRIDriverContext *ctx, I830Rec *pI830, I830MemRange *pMem) +{ + int pitch, ntiles, i; + + pitch = pMem->Pitch * ctx->cpp; + /* + * Simply try to break the region up into at most four pieces of size + * equal to the alignment. + */ + ntiles = ROUND_TO(pMem->Size, pMem->Alignment) / pMem->Alignment; + if (ntiles >= 4) { + return FALSE; + } + + for (i = 0; i < ntiles; i++, nextTile++) { + SetFence(ctx, pI830, nextTile, pMem->Start + i * pMem->Alignment, + pitch, pMem->Alignment); + } + return TRUE; +} + +static void I830SetupMemoryTiling(const DRIDriverContext *ctx, I830Rec *pI830) +{ + int i; + + /* Clear out */ + for (i = 0; i < 8; i++) + pI830->Fence[i] = 0; + + nextTile = 0; + + if (pI830->BackBuffer.Alignment >= KB(512)) { + if (MakeTiles(ctx, pI830, &(pI830->BackBuffer))) { + fprintf(stderr, + "Activating tiled memory for the back buffer.\n"); + } else { + fprintf(stderr, + "MakeTiles failed for the back buffer.\n"); + pI830->allowPageFlip = FALSE; + } + } + + if (pI830->DepthBuffer.Alignment >= KB(512)) { + if (MakeTiles(ctx, pI830, &(pI830->DepthBuffer))) { + fprintf(stderr, + "Activating tiled memory for the depth buffer.\n"); + } else { + fprintf(stderr, + "MakeTiles failed for the depth buffer.\n"); + } + } + + return; +} + +static int I830DetectMemory(const DRIDriverContext *ctx, I830Rec *pI830) +{ + struct pci_device host_bridge, ig_dev; + uint32_t gmch_ctrl; + int memsize = 0; + int range; + uint32_t aper_size; + uint32_t membase2 = 0; + + memset(&host_bridge, 0, sizeof(host_bridge)); + memset(&ig_dev, 0, sizeof(ig_dev)); + + ig_dev.dev = 2; + + pci_device_cfg_read_u32(&host_bridge, &gmch_ctrl, I830_GMCH_CTRL); + + if (IS_I830(pI830) || IS_845G(pI830)) { + if ((gmch_ctrl & I830_GMCH_MEM_MASK) == I830_GMCH_MEM_128M) { + aper_size = 0x80000000; + } else { + aper_size = 0x40000000; + } + } else { + if (IS_I9XX(pI830)) { + int ret; + ret = pci_device_cfg_read_u32(&ig_dev, &membase2, 0x18); + if (membase2 & 0x08000000) + aper_size = 0x8000000; + else + aper_size = 0x10000000; + + fprintf(stderr,"aper size is %08X %08x %d\n", aper_size, membase2, ret); + } else + aper_size = 0x8000000; + } + + pI830->aper_size = aper_size; + + + /* We need to reduce the stolen size, by the GTT and the popup. + * The GTT varying according the the FbMapSize and the popup is 4KB */ + range = (ctx->shared.fbSize / (1024*1024)) + 4; + + if (IS_I85X(pI830) || IS_I865G(pI830) || IS_I9XX(pI830)) { + switch (gmch_ctrl & I830_GMCH_GMS_MASK) { + case I855_GMCH_GMS_STOLEN_1M: + memsize = MB(1) - KB(range); + break; + case I855_GMCH_GMS_STOLEN_4M: + memsize = MB(4) - KB(range); + break; + case I855_GMCH_GMS_STOLEN_8M: + memsize = MB(8) - KB(range); + break; + case I855_GMCH_GMS_STOLEN_16M: + memsize = MB(16) - KB(range); + break; + case I855_GMCH_GMS_STOLEN_32M: + memsize = MB(32) - KB(range); + break; + case I915G_GMCH_GMS_STOLEN_48M: + if (IS_I9XX(pI830)) + memsize = MB(48) - KB(range); + break; + case I915G_GMCH_GMS_STOLEN_64M: + if (IS_I9XX(pI830)) + memsize = MB(64) - KB(range); + break; + } + } else { + switch (gmch_ctrl & I830_GMCH_GMS_MASK) { + case I830_GMCH_GMS_STOLEN_512: + memsize = KB(512) - KB(range); + break; + case I830_GMCH_GMS_STOLEN_1024: + memsize = MB(1) - KB(range); + break; + case I830_GMCH_GMS_STOLEN_8192: + memsize = MB(8) - KB(range); + break; + case I830_GMCH_GMS_LOCAL: + memsize = 0; + xf86DrvMsg(pScrn->scrnIndex, X_WARNING, + "Local memory found, but won't be used.\n"); + break; + } + } + if (memsize > 0) { + fprintf(stderr, + "detected %d kB stolen memory.\n", memsize / 1024); + } else { + fprintf(stderr, + "no video memory detected.\n"); + } + return memsize; +} + +static int AgpInit(const DRIDriverContext *ctx, I830Rec *info) +{ + unsigned long mode = 0x4; + + if (drmAgpAcquire(ctx->drmFD) < 0) { + fprintf(stderr, "[gart] AGP not available\n"); + return 0; + } + + if (drmAgpEnable(ctx->drmFD, mode) < 0) { + fprintf(stderr, "[gart] AGP not enabled\n"); + drmAgpRelease(ctx->drmFD); + return 0; + } + else + fprintf(stderr, "[gart] AGP enabled at %dx\n", ctx->agpmode); + + return 1; +} + +/* + * Allocate memory from the given pool. Grow the pool if needed and if + * possible. + */ +static unsigned long +AllocFromPool(const DRIDriverContext *ctx, I830Rec *pI830, + I830MemRange *result, I830MemPool *pool, + long size, unsigned long alignment, int flags) +{ + long needed, start, end; + + if (!result || !pool || !size) + return 0; + + /* Calculate how much space is needed. */ + if (alignment <= GTT_PAGE_SIZE) + needed = size; + else { + start = ROUND_TO(pool->Free.Start, alignment); + end = ROUND_TO(start + size, alignment); + needed = end - pool->Free.Start; + } + if (needed > pool->Free.Size) { + return 0; + } + + result->Start = ROUND_TO(pool->Free.Start, alignment); + pool->Free.Start += needed; + result->End = pool->Free.Start; + + pool->Free.Size = pool->Free.End - pool->Free.Start; + result->Size = result->End - result->Start; + result->Pool = pool; + result->Alignment = alignment; + return needed; +} + +static unsigned long AllocFromAGP(const DRIDriverContext *ctx, I830Rec *pI830, long size, unsigned long alignment, I830MemRange *result) +{ + unsigned long start, end; + unsigned long newApStart, newApEnd; + int ret; + if (!result || !size) + return 0; + + if (!alignment) + alignment = 4; + + start = ROUND_TO(pI830->MemoryAperture.Start, alignment); + end = ROUND_TO(start + size, alignment); + newApStart = end; + newApEnd = pI830->MemoryAperture.End; + + ret=drmAgpAlloc(ctx->drmFD, size, 0, &(result->Physical), (drm_handle_t *)&(result->Key)); + + if (ret) + { + fprintf(stderr,"drmAgpAlloc failed %d\n", ret); + return 0; + } + pI830->allocatedMemory += size; + pI830->MemoryAperture.Start = newApStart; + pI830->MemoryAperture.End = newApEnd; + pI830->MemoryAperture.Size = newApEnd - newApStart; + // pI830->FreeMemory -= size; + result->Start = start; + result->End = start + size; + result->Size = size; + result->Offset = start; + result->Alignment = alignment; + result->Pool = NULL; + + return size; +} + +unsigned long +I830AllocVidMem(const DRIDriverContext *ctx, I830Rec *pI830, + I830MemRange *result, I830MemPool *pool, long size, + unsigned long alignment, int flags) +{ + unsigned long ret; + + if (!result) + return 0; + + /* Make sure these are initialised. */ + result->Size = 0; + result->Key = -1; + + if (!size) { + return 0; + } + + if (pool->Free.Size < size) { + ret = AllocFromAGP(ctx, pI830, size, alignment, result); + } + else { + ret = AllocFromPool(ctx, pI830, result, pool, size, alignment, flags); + if (ret == 0) + ret = AllocFromAGP(ctx, pI830, size, alignment, result); + } + return ret; +} + +static Bool BindAgpRange(const DRIDriverContext *ctx, I830MemRange *mem) +{ + if (!mem) + return FALSE; + + if (mem->Key == -1) + return TRUE; + + return !drmAgpBind(ctx->drmFD, mem->Key, mem->Offset); +} + +/* simple memory allocation routines needed */ +/* put ring buffer in low memory */ +/* need to allocate front, back, depth buffers aligned correctly, + allocate ring buffer, +*/ + +/* */ +static Bool +I830AllocateMemory(const DRIDriverContext *ctx, I830Rec *pI830) +{ + unsigned long size, ret; + unsigned long lines, lineSize, align; + + /* allocate ring buffer */ + memset(pI830->LpRing, 0, sizeof(I830RingBuffer)); + pI830->LpRing->mem.Key = -1; + + size = PRIMARY_RINGBUFFER_SIZE; + + ret = I830AllocVidMem(ctx, pI830, &pI830->LpRing->mem, &pI830->StolenPool, size, 0x1000, 0); + + if (ret != size) + { + fprintf(stderr,"unable to allocate ring buffer %ld\n", ret); + return FALSE; + } + + pI830->LpRing->tail_mask = pI830->LpRing->mem.Size - 1; + + + /* allocate front buffer */ + memset(&(pI830->FrontBuffer), 0, sizeof(pI830->FrontBuffer)); + pI830->FrontBuffer.Key = -1; + pI830->FrontBuffer.Pitch = ctx->shared.virtualWidth; + + align = KB(512); + + lineSize = ctx->shared.virtualWidth * ctx->cpp; + lines = (ctx->shared.virtualHeight + 15) / 16 * 16; + size = lineSize * lines; + size = ROUND_TO_PAGE(size); + + align = GetBestTileAlignment(size); + + ret = I830AllocVidMem(ctx, pI830, &pI830->FrontBuffer, &pI830->StolenPool, size, align, 0); + if (ret < size) + { + fprintf(stderr,"unable to allocate front buffer %ld\n", ret); + return FALSE; + } + + memset(&(pI830->BackBuffer), 0, sizeof(pI830->BackBuffer)); + pI830->BackBuffer.Key = -1; + pI830->BackBuffer.Pitch = ctx->shared.virtualWidth; + + ret = I830AllocVidMem(ctx, pI830, &pI830->BackBuffer, &pI830->StolenPool, size, align, 0); + if (ret < size) + { + fprintf(stderr,"unable to allocate back buffer %ld\n", ret); + return FALSE; + } + + memset(&(pI830->DepthBuffer), 0, sizeof(pI830->DepthBuffer)); + pI830->DepthBuffer.Key = -1; + pI830->DepthBuffer.Pitch = ctx->shared.virtualWidth; + + ret = I830AllocVidMem(ctx, pI830, &pI830->DepthBuffer, &pI830->StolenPool, size, align, 0); + if (ret < size) + { + fprintf(stderr,"unable to allocate depth buffer %ld\n", ret); + return FALSE; + } + + memset(&(pI830->ContextMem), 0, sizeof(pI830->ContextMem)); + pI830->ContextMem.Key = -1; + size = KB(32); + + ret = I830AllocVidMem(ctx, pI830, &pI830->ContextMem, &pI830->StolenPool, size, align, 0); + if (ret < size) + { + fprintf(stderr,"unable to allocate context buffer %ld\n", ret); + return FALSE; + } + +#if 0 + memset(&(pI830->TexMem), 0, sizeof(pI830->TexMem)); + pI830->TexMem.Key = -1; + + size = 32768 * 1024; + ret = AllocFromAGP(ctx, pI830, size, align, &pI830->TexMem); + if (ret < size) + { + fprintf(stderr,"unable to allocate texture memory %ld\n", ret); + return FALSE; + } +#endif + + return TRUE; +} + +static Bool +I830BindMemory(const DRIDriverContext *ctx, I830Rec *pI830) +{ + if (!BindAgpRange(ctx, &pI830->LpRing->mem)) + return FALSE; + if (!BindAgpRange(ctx, &pI830->FrontBuffer)) + return FALSE; + if (!BindAgpRange(ctx, &pI830->BackBuffer)) + return FALSE; + if (!BindAgpRange(ctx, &pI830->DepthBuffer)) + return FALSE; + if (!BindAgpRange(ctx, &pI830->ContextMem)) + return FALSE; +#if 0 + if (!BindAgpRange(ctx, &pI830->TexMem)) + return FALSE; +#endif + return TRUE; +} + +static void SetupDRIMM(const DRIDriverContext *ctx, I830Rec *pI830) +{ + unsigned long aperEnd = ROUND_DOWN_TO(pI830->aper_size, GTT_PAGE_SIZE) / GTT_PAGE_SIZE; + unsigned long aperStart = ROUND_TO(pI830->aper_size - KB(32768), GTT_PAGE_SIZE) / GTT_PAGE_SIZE; + + fprintf(stderr, "aper size is %08X\n", ctx->shared.fbSize); + if (drmMMInit(ctx->drmFD, aperStart, aperEnd - aperStart, DRM_BO_MEM_TT)) { + fprintf(stderr, + "DRM MM Initialization Failed\n"); + } else { + fprintf(stderr, + "DRM MM Initialized at offset 0x%lx length %d page\n", aperStart, aperEnd-aperStart); + } + +} + +static Bool +I830CleanupDma(const DRIDriverContext *ctx) +{ + drmI830Init info; + + memset(&info, 0, sizeof(drmI830Init)); + info.func = I830_CLEANUP_DMA; + + if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT, + &info, sizeof(drmI830Init))) { + fprintf(stderr, "I830 Dma Cleanup Failed\n"); + return FALSE; + } + + return TRUE; +} + +static Bool +I830InitDma(const DRIDriverContext *ctx, I830Rec *pI830) +{ + I830RingBuffer *ring = pI830->LpRing; + drmI830Init info; + + memset(&info, 0, sizeof(drmI830Init)); + info.func = I830_INIT_DMA; + + info.ring_start = ring->mem.Start + pI830->LinearAddr; + info.ring_end = ring->mem.End + pI830->LinearAddr; + info.ring_size = ring->mem.Size; + + info.mmio_offset = (unsigned int)ctx->MMIOStart; + + info.sarea_priv_offset = sizeof(drm_sarea_t); + + info.front_offset = pI830->FrontBuffer.Start; + info.back_offset = pI830->BackBuffer.Start; + info.depth_offset = pI830->DepthBuffer.Start; + info.w = ctx->shared.virtualWidth; + info.h = ctx->shared.virtualHeight; + info.pitch = ctx->shared.virtualWidth; + info.back_pitch = pI830->BackBuffer.Pitch; + info.depth_pitch = pI830->DepthBuffer.Pitch; + info.cpp = ctx->cpp; + + if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT, + &info, sizeof(drmI830Init))) { + fprintf(stderr, + "I830 Dma Initialization Failed\n"); + return FALSE; + } + + return TRUE; +} + +static int I830CheckDRMVersion( const DRIDriverContext *ctx, + I830Rec *pI830 ) +{ + drmVersionPtr version; + + version = drmGetVersion(ctx->drmFD); + + if (version) { + int req_minor, req_patch; + + req_minor = 4; + req_patch = 0; + + if (version->version_major != 1 || + version->version_minor < req_minor || + (version->version_minor == req_minor && + version->version_patchlevel < req_patch)) { + /* Incompatible drm version */ + fprintf(stderr, + "[dri] I830DRIScreenInit failed because of a version " + "mismatch.\n" + "[dri] i915.o kernel module version is %d.%d.%d " + "but version 1.%d.%d or newer is needed.\n" + "[dri] Disabling DRI.\n", + version->version_major, + version->version_minor, + version->version_patchlevel, + req_minor, + req_patch); + drmFreeVersion(version); + return 0; + } + + pI830->drmMinor = version->version_minor; + drmFreeVersion(version); + } + return 1; +} + +static void +I830SetRingRegs(const DRIDriverContext *ctx, I830Rec *pI830) +{ + unsigned int itemp; + unsigned char *MMIO = ctx->MMIOAddress; + + OUTREG(LP_RING + RING_LEN, 0); + OUTREG(LP_RING + RING_TAIL, 0); + OUTREG(LP_RING + RING_HEAD, 0); + + if ((long)(pI830->LpRing->mem.Start & I830_RING_START_MASK) != + pI830->LpRing->mem.Start) { + fprintf(stderr, + "I830SetRingRegs: Ring buffer start (%lx) violates its " + "mask (%x)\n", pI830->LpRing->mem.Start, I830_RING_START_MASK); + } + /* Don't care about the old value. Reserved bits must be zero anyway. */ + itemp = pI830->LpRing->mem.Start & I830_RING_START_MASK; + OUTREG(LP_RING + RING_START, itemp); + + if (((pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES) != + pI830->LpRing->mem.Size - 4096) { + fprintf(stderr, + "I830SetRingRegs: Ring buffer size - 4096 (%lx) violates its " + "mask (%x)\n", pI830->LpRing->mem.Size - 4096, + I830_RING_NR_PAGES); + } + /* Don't care about the old value. Reserved bits must be zero anyway. */ + itemp = (pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES; + itemp |= (RING_NO_REPORT | RING_VALID); + OUTREG(LP_RING + RING_LEN, itemp); + + pI830->LpRing->head = INREG(LP_RING + RING_HEAD) & I830_HEAD_MASK; + pI830->LpRing->tail = INREG(LP_RING + RING_TAIL); + pI830->LpRing->space = pI830->LpRing->head - (pI830->LpRing->tail + 8); + if (pI830->LpRing->space < 0) + pI830->LpRing->space += pI830->LpRing->mem.Size; + + SetFenceRegs(ctx, pI830); + + /* RESET THE DISPLAY PIPE TO POINT TO THE FRONTBUFFER - hacky + hacky hacky */ + OUTREG(DSPABASE, pI830->FrontBuffer.Start + pI830->LinearAddr); + +} + +static Bool +I830SetParam(const DRIDriverContext *ctx, int param, int value) +{ + drmI830SetParam sp; + + memset(&sp, 0, sizeof(sp)); + sp.param = param; + sp.value = value; + + if (drmCommandWrite(ctx->drmFD, DRM_I830_SETPARAM, &sp, sizeof(sp))) { + fprintf(stderr, "I830 SetParam Failed\n"); + return FALSE; + } + + return TRUE; +} + +static Bool +I830DRIMapScreenRegions(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea) +{ + fprintf(stderr, + "[drm] Mapping front buffer\n"); + + if (drmAddMap(ctx->drmFD, + (drm_handle_t)(sarea->front_offset + pI830->LinearAddr), + sarea->front_size, + DRM_FRAME_BUFFER, /*DRM_AGP,*/ + 0, + &sarea->front_handle) < 0) { + fprintf(stderr, + "[drm] drmAddMap(front_handle) failed. Disabling DRI\n"); + return FALSE; + } + ctx->shared.hFrameBuffer = sarea->front_handle; + ctx->shared.fbSize = sarea->front_size; + fprintf(stderr, "[drm] Front Buffer = 0x%08x\n", + sarea->front_handle); + + if (drmAddMap(ctx->drmFD, + (drm_handle_t)(sarea->back_offset), + sarea->back_size, DRM_AGP, 0, + &sarea->back_handle) < 0) { + fprintf(stderr, + "[drm] drmAddMap(back_handle) failed. Disabling DRI\n"); + return FALSE; + } + fprintf(stderr, "[drm] Back Buffer = 0x%08x\n", + sarea->back_handle); + + if (drmAddMap(ctx->drmFD, + (drm_handle_t)sarea->depth_offset, + sarea->depth_size, DRM_AGP, 0, + &sarea->depth_handle) < 0) { + fprintf(stderr, + "[drm] drmAddMap(depth_handle) failed. Disabling DRI\n"); + return FALSE; + } + fprintf(stderr, "[drm] Depth Buffer = 0x%08x\n", + sarea->depth_handle); + +#if 0 + if (drmAddMap(ctx->drmFD, + (drm_handle_t)sarea->tex_offset, + sarea->tex_size, DRM_AGP, 0, + &sarea->tex_handle) < 0) { + fprintf(stderr, + "[drm] drmAddMap(tex_handle) failed. Disabling DRI\n"); + return FALSE; + } + fprintf(stderr, "[drm] textures = 0x%08x\n", + sarea->tex_handle); +#endif + return TRUE; +} + + +static void +I830DRIUnmapScreenRegions(const DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea) +{ +#if 1 + if (sarea->front_handle) { + drmRmMap(ctx->drmFD, sarea->front_handle); + sarea->front_handle = 0; + } +#endif + if (sarea->back_handle) { + drmRmMap(ctx->drmFD, sarea->back_handle); + sarea->back_handle = 0; + } + if (sarea->depth_handle) { + drmRmMap(ctx->drmFD, sarea->depth_handle); + sarea->depth_handle = 0; + } + if (sarea->tex_handle) { + drmRmMap(ctx->drmFD, sarea->tex_handle); + sarea->tex_handle = 0; + } +} + +static Bool +I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea) +{ + if (drmAddMap(ctx->drmFD, + (drm_handle_t)pI830->LpRing->mem.Start, + pI830->LpRing->mem.Size, DRM_AGP, 0, + &pI830->ring_map) < 0) { + fprintf(stderr, + "[drm] drmAddMap(ring_map) failed. Disabling DRI\n"); + return FALSE; + } + fprintf(stderr, "[drm] ring buffer = 0x%08x\n", + pI830->ring_map); + + if (I830InitDma(ctx, pI830) == FALSE) { + return FALSE; + } + + /* init to zero to be safe */ + + I830DRIMapScreenRegions(ctx, pI830, sarea); + SetupDRIMM(ctx, pI830); + + if (ctx->pciDevice != PCI_CHIP_845_G && + ctx->pciDevice != PCI_CHIP_I830_M) { + I830SetParam(ctx, I830_SETPARAM_USE_MI_BATCHBUFFER_START, 1 ); + } + + /* Okay now initialize the dma engine */ + { + pI830->irq = drmGetInterruptFromBusID(ctx->drmFD, + ctx->pciBus, + ctx->pciDevice, + ctx->pciFunc); + + if (drmCtlInstHandler(ctx->drmFD, pI830->irq)) { + fprintf(stderr, + "[drm] failure adding irq handler\n"); + pI830->irq = 0; + return FALSE; + } + else + fprintf(stderr, + "[drm] dma control initialized, using IRQ %d\n", + pI830->irq); + } + + fprintf(stderr, "[dri] visual configs initialized\n"); + + return TRUE; +} + +static Bool +I830ClearScreen(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea) +{ + /* need to drmMap front and back buffers and zero them */ + drmAddress map_addr; + int ret; + + ret = drmMap(ctx->drmFD, + sarea->front_handle, + sarea->front_size, + &map_addr); + + if (ret) + { + fprintf(stderr, "Unable to map front buffer\n"); + return FALSE; + } + + drimemsetio((char *)map_addr, + 0, + sarea->front_size); + drmUnmap(map_addr, sarea->front_size); + + + ret = drmMap(ctx->drmFD, + sarea->back_handle, + sarea->back_size, + &map_addr); + + if (ret) + { + fprintf(stderr, "Unable to map back buffer\n"); + return FALSE; + } + + drimemsetio((char *)map_addr, + 0, + sarea->back_size); + drmUnmap(map_addr, sarea->back_size); + + return TRUE; +} + +static Bool +I830ScreenInit(DRIDriverContext *ctx, I830Rec *pI830) + +{ + I830DRIPtr pI830DRI; + drmI830Sarea *pSAREAPriv; + int err; + + drm_page_size = getpagesize(); + + pI830->registerSize = ctx->MMIOSize; + /* This is a hack for now. We have to have more than a 4k page here + * because of the size of the state. However, the state should be + * in a per-context mapping. This will be added in the Mesa 3.5 port + * of the I830 driver. + */ + ctx->shared.SAREASize = SAREA_MAX; + + /* Note that drmOpen will try to load the kernel module, if needed. */ + ctx->drmFD = drmOpen("i915", NULL ); + if (ctx->drmFD < 0) { + fprintf(stderr, "[drm] drmOpen failed\n"); + return 0; + } + + if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) { + fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n", + ctx->drmFD, ctx->pciBusID, strerror(-err)); + return 0; + } + + if (drmAddMap( ctx->drmFD, + 0, + ctx->shared.SAREASize, + DRM_SHM, + DRM_CONTAINS_LOCK, + &ctx->shared.hSAREA) < 0) + { + fprintf(stderr, "[drm] drmAddMap failed\n"); + return 0; + } + + fprintf(stderr, "[drm] added %d byte SAREA at 0x%08x\n", + ctx->shared.SAREASize, ctx->shared.hSAREA); + + if (drmMap( ctx->drmFD, + ctx->shared.hSAREA, + ctx->shared.SAREASize, + (drmAddressPtr)(&ctx->pSAREA)) < 0) + { + fprintf(stderr, "[drm] drmMap failed\n"); + return 0; + + } + + memset(ctx->pSAREA, 0, ctx->shared.SAREASize); + fprintf(stderr, "[drm] mapped SAREA 0x%08x to %p, size %d\n", + ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize); + + + if (drmAddMap(ctx->drmFD, + ctx->MMIOStart, + ctx->MMIOSize, + DRM_REGISTERS, + DRM_READ_ONLY, + &pI830->registerHandle) < 0) { + fprintf(stderr, "[drm] drmAddMap mmio failed\n"); + return 0; + } + fprintf(stderr, + "[drm] register handle = 0x%08x\n", pI830->registerHandle); + + + if (!I830CheckDRMVersion(ctx, pI830)) { + return FALSE; + } + + /* Create a 'server' context so we can grab the lock for + * initialization ioctls. + */ + if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) { + fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err); + return 0; + } + + DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); + + /* Initialize the SAREA private data structure */ + pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + + sizeof(drm_sarea_t)); + memset(pSAREAPriv, 0, sizeof(*pSAREAPriv)); + + pI830->StolenMemory.Size = I830DetectMemory(ctx, pI830); + pI830->StolenMemory.Start = 0; + pI830->StolenMemory.End = pI830->StolenMemory.Size; + + pI830->MemoryAperture.Start = pI830->StolenMemory.End; + pI830->MemoryAperture.End = KB(40000); + pI830->MemoryAperture.Size = pI830->MemoryAperture.End - pI830->MemoryAperture.Start; + + pI830->StolenPool.Fixed = pI830->StolenMemory; + pI830->StolenPool.Total = pI830->StolenMemory; + pI830->StolenPool.Free = pI830->StolenPool.Total; + pI830->FreeMemory = pI830->StolenPool.Total.Size; + + if (!AgpInit(ctx, pI830)) + return FALSE; + + if (I830AllocateMemory(ctx, pI830) == FALSE) + { + return FALSE; + } + + if (I830BindMemory(ctx, pI830) == FALSE) + { + return FALSE; + } + + pSAREAPriv->rotated_offset = -1; + pSAREAPriv->rotated_size = 0; + pSAREAPriv->rotated_pitch = ctx->shared.virtualWidth; + + pSAREAPriv->front_offset = pI830->FrontBuffer.Start; + pSAREAPriv->front_size = pI830->FrontBuffer.Size; + pSAREAPriv->width = ctx->shared.virtualWidth; + pSAREAPriv->height = ctx->shared.virtualHeight; + pSAREAPriv->pitch = ctx->shared.virtualWidth; + pSAREAPriv->virtualX = ctx->shared.virtualWidth; + pSAREAPriv->virtualY = ctx->shared.virtualHeight; + pSAREAPriv->back_offset = pI830->BackBuffer.Start; + pSAREAPriv->back_size = pI830->BackBuffer.Size; + pSAREAPriv->depth_offset = pI830->DepthBuffer.Start; + pSAREAPriv->depth_size = pI830->DepthBuffer.Size; +#if 0 + pSAREAPriv->tex_offset = pI830->TexMem.Start; + pSAREAPriv->tex_size = pI830->TexMem.Size; +#endif + pSAREAPriv->log_tex_granularity = pI830->TexGranularity; + + ctx->driverClientMsg = malloc(sizeof(I830DRIRec)); + ctx->driverClientMsgSize = sizeof(I830DRIRec); + pI830DRI = (I830DRIPtr)ctx->driverClientMsg; + pI830DRI->deviceID = pI830->Chipset; + pI830DRI->regsSize = I830_REG_SIZE; + pI830DRI->width = ctx->shared.virtualWidth; + pI830DRI->height = ctx->shared.virtualHeight; + pI830DRI->mem = ctx->shared.fbSize; + pI830DRI->cpp = ctx->cpp; + + pI830DRI->bitsPerPixel = ctx->bpp; + pI830DRI->sarea_priv_offset = sizeof(drm_sarea_t); + + err = I830DRIDoMappings(ctx, pI830, pSAREAPriv); + if (err == FALSE) + return FALSE; + + I830SetupMemoryTiling(ctx, pI830); + + /* Quick hack to clear the front & back buffers. Could also use + * the clear ioctl to do this, but would need to setup hw state + * first. + */ + I830ClearScreen(ctx, pI830, pSAREAPriv); + + I830SetRingRegs(ctx, pI830); + + return TRUE; +} + + +/** + * \brief Validate the fbdev mode. + * + * \param ctx display handle. + * + * \return one on success, or zero on failure. + * + * Saves some registers and returns 1. + * + * \sa radeonValidateMode(). + */ +static int i830ValidateMode( const DRIDriverContext *ctx ) +{ + return 1; +} + +/** + * \brief Examine mode returned by fbdev. + * + * \param ctx display handle. + * + * \return one on success, or zero on failure. + * + * Restores registers that fbdev has clobbered and returns 1. + * + * \sa i810ValidateMode(). + */ +static int i830PostValidateMode( const DRIDriverContext *ctx ) +{ + I830Rec *pI830 = ctx->driverPrivate; + + I830SetRingRegs(ctx, pI830); + return 1; +} + + +/** + * \brief Initialize the framebuffer device mode + * + * \param ctx display handle. + * + * \return one on success, or zero on failure. + * + * Fills in \p info with some default values and some information from \p ctx + * and then calls I810ScreenInit() for the screen initialization. + * + * Before exiting clears the framebuffer memory accessing it directly. + */ +static int i830InitFBDev( DRIDriverContext *ctx ) +{ + I830Rec *pI830 = calloc(1, sizeof(I830Rec)); + int i; + + { + int dummy = ctx->shared.virtualWidth; + + switch (ctx->bpp / 8) { + case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break; + case 2: dummy = (ctx->shared.virtualWidth + 31) & ~31; break; + case 3: + case 4: dummy = (ctx->shared.virtualWidth + 15) & ~15; break; + } + + ctx->shared.virtualWidth = dummy; + ctx->shared.Width = ctx->shared.virtualWidth; + } + + + for (i = 0; pitches[i] != 0; i++) { + if (pitches[i] >= ctx->shared.virtualWidth) { + ctx->shared.virtualWidth = pitches[i]; + break; + } + } + + ctx->driverPrivate = (void *)pI830; + + pI830->LpRing = calloc(1, sizeof(I830RingBuffer)); + pI830->Chipset = ctx->chipset; + pI830->LinearAddr = ctx->FBStart; + + if (!I830ScreenInit( ctx, pI830 )) + return 0; + + + return 1; +} + + +/** + * \brief The screen is being closed, so clean up any state and free any + * resources used by the DRI. + * + * \param ctx display handle. + * + * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver + * private data. + */ +static void i830HaltFBDev( DRIDriverContext *ctx ) +{ + drmI830Sarea *pSAREAPriv; + I830Rec *pI830 = ctx->driverPrivate; + + if (pI830->irq) { + drmCtlUninstHandler(ctx->drmFD); + pI830->irq = 0; } + + I830CleanupDma(ctx); + + pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + + sizeof(drm_sarea_t)); + + I830DRIUnmapScreenRegions(ctx, pI830, pSAREAPriv); + drmUnmap( ctx->pSAREA, ctx->shared.SAREASize ); + drmClose(ctx->drmFD); + + if (ctx->driverPrivate) { + free(ctx->driverPrivate); + ctx->driverPrivate = 0; + } +} + + +extern void i810NotifyFocus( int ); + +/** + * \brief Exported driver interface for Mini GLX. + * + * \sa DRIDriverRec. + */ +const struct DRIDriverRec __driDriver = { + i830ValidateMode, + i830PostValidateMode, + i830InitFBDev, + i830HaltFBDev, + NULL,//I830EngineShutdown, + NULL, //I830EngineRestore, +#ifndef _EMBEDDED + 0, +#else + i810NotifyFocus, +#endif +}; |