static void r200WaitForFrameCompletion( r200ContextPtr rmesa ) { drm_radeon_sarea_t *sarea = rmesa->sarea; if (rmesa->do_irqs) { if (r200GetLastFrame(rmesa) < sarea->last_frame) { if (!rmesa->irqsEmitted) { while (r200GetLastFrame (rmesa) < sarea->last_frame) ; } else { UNLOCK_HARDWARE( rmesa ); r200WaitIrq( rmesa ); LOCK_HARDWARE( rmesa ); } rmesa->irqsEmitted = 10; } if (rmesa->irqsEmitted) { r200EmitIrqLocked( rmesa ); rmesa->irqsEmitted--; } } else { while (r200GetLastFrame (rmesa) < sarea->last_frame) { UNLOCK_HARDWARE( rmesa ); if (rmesa->do_usleeps) DO_USLEEP( 1 ); LOCK_HARDWARE( rmesa ); } } }
int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t, GLuint face ) { const int numLevels = t->base.lastLevel - t->base.firstLevel + 1; if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) { fprintf( stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__, (void *)rmesa->glCtx, (void *)t->base.tObj, t->base.totalSize, t->base.firstLevel, t->base.lastLevel ); } if ( !t || t->base.totalSize == 0 ) return 0; LOCK_HARDWARE( rmesa ); if ( t->base.memBlock == NULL ) { int heap; heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps, (driTextureObject *) t ); if ( heap == -1 ) { UNLOCK_HARDWARE( rmesa ); return -1; } /* Set the base offset of the texture image */ t->bufAddr = rmesa->radeonScreen->texOffset[heap] + t->base.memBlock->ofs; t->pp_txoffset = t->bufAddr; /* Mark this texobj as dirty on all units: */ t->dirty_state = TEX_ALL; } /* Let the world know we've used this memory recently. */ driUpdateTextureLRU( (driTextureObject *) t ); UNLOCK_HARDWARE( rmesa ); /* Upload any images that are new */ if (t->base.dirty_images[face]) { int i; for ( i = 0 ; i < numLevels ; i++ ) { if ( (t->base.dirty_images[face] & (1 << (i+t->base.firstLevel))) != 0 ) { uploadSubImage( rmesa, t, i, 0, 0, t->image[face][i].width, t->image[face][i].height, face ); } } t->base.dirty_images[face] = 0; } return 0; }
static void do_draw_pix( GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height, GLint pitch, const void *pixels, GLuint dest ) { intelContextPtr intel = INTEL_CONTEXT(ctx); __DRIdrawablePrivate *dPriv = intel->driDrawable; drm_clip_rect_t *box = dPriv->pClipRects; int nbox = dPriv->numClipRects; int i; int src_offset = intelAgpOffsetFromVirtual( intel, pixels); int src_pitch = pitch; assert(src_offset != ~0); /* should be caught earlier */ if (INTEL_DEBUG & DEBUG_PIXEL) fprintf(stderr, "%s\n", __FUNCTION__); intelFlush( &intel->ctx ); LOCK_HARDWARE( intel ); if (ctx->DrawBuffer) { y -= height; /* cope with pixel zoom */ if (!clip_pixelrect(ctx, ctx->DrawBuffer, &x, &y, &width, &height)) { UNLOCK_HARDWARE( intel ); return; } y = dPriv->h - y - height; /* convert from gl to hardware coords */ x += dPriv->x; y += dPriv->y; for (i = 0 ; i < nbox ; i++ ) { GLint bx, by, bw, bh; if (intersect_region(box + i, x, y, width, height, &bx, &by, &bw, &bh)) { intelEmitCopyBlitLocked( intel, intel->intelScreen->cpp, src_pitch, src_offset, intel->intelScreen->front.pitch, intel->drawRegion->offset, bx - x, by - y, bx, by, bw, bh ); } } } UNLOCK_HARDWARE( intel ); intelFinish( &intel->ctx ); }
static void mgaWaitForFrameCompletion( mgaContextPtr mmesa ) { if ( mgaWaitFence( mmesa, mmesa->last_frame_fence, NULL ) == ENOSYS ) { unsigned wait = 0; GLuint last_frame; GLuint last_wrap; LOCK_HARDWARE( mmesa ); last_frame = mmesa->sarea->last_frame.head; last_wrap = mmesa->sarea->last_frame.wrap; /* The DMA routines in the kernel track a couple values in the SAREA * that we use here. The number of times that the primary DMA buffer * has "wrapped" around is tracked in last_wrap. In addition, the * wrap count and the buffer position at the end of the last frame are * stored in last_frame.wrap and last_frame.head. * * By comparing the wrap counts and the current DMA pointer value * (read directly from the hardware) to last_frame.head, we can * determine when the graphics processor has processed all of the * commands for the last frame. * * In this case "last frame" means the frame of the *previous* swap- * buffers call. This is done to prevent queuing a second buffer swap * before the previous swap is executed. */ while ( 1 ) { if ( last_wrap < mmesa->sarea->last_wrap || ( last_wrap == mmesa->sarea->last_wrap && last_frame <= (MGA_READ( MGAREG_PRIMADDRESS ) - mmesa->primary_offset) ) ) { break; } if ( 0 ) { wait++; fprintf( stderr, " last: head=0x%06x wrap=%d\n", last_frame, last_wrap ); fprintf( stderr, " head: head=0x%06lx wrap=%d\n", (long)(MGA_READ( MGAREG_PRIMADDRESS ) - mmesa->primary_offset), mmesa->sarea->last_wrap ); } UPDATE_LOCK( mmesa, DRM_LOCK_FLUSH ); UNLOCK_HARDWARE( mmesa ); DO_USLEEP( 1 ); LOCK_HARDWARE( mmesa ); } if ( wait ) fprintf( stderr, "\n" ); UNLOCK_HARDWARE( mmesa ); } }
/* * Initialize texture memory. * We take care of one or both TMU's here. */ void tdfxTMInit(tdfxContextPtr fxMesa) { if (!fxMesa->glCtx->Shared->DriverData) { const char *extensions; struct tdfxSharedState *shared = CALLOC_STRUCT(tdfxSharedState); if (!shared) return; LOCK_HARDWARE(fxMesa); extensions = fxMesa->Glide.grGetString(GR_EXTENSION); UNLOCK_HARDWARE(fxMesa); if (strstr(extensions, "TEXUMA")) { FxU32 start, end; shared->umaTexMemory = GL_TRUE; LOCK_HARDWARE(fxMesa); fxMesa->Glide.grEnable(GR_TEXTURE_UMA_EXT); start = fxMesa->Glide.grTexMinAddress(0); end = fxMesa->Glide.grTexMaxAddress(0); UNLOCK_HARDWARE(fxMesa); shared->totalTexMem[0] = end - start; shared->totalTexMem[1] = 0; shared->freeTexMem[0] = end - start; shared->freeTexMem[1] = 0; shared->tmFree[0] = NewRangeNode(fxMesa, start, end); shared->tmFree[1] = NULL; /*printf("UMA tex memory: %d\n", (int) (end - start));*/ } else { const int numTMUs = fxMesa->haveTwoTMUs ? 2 : 1; int tmu; shared->umaTexMemory = GL_FALSE; LOCK_HARDWARE(fxMesa); for (tmu = 0; tmu < numTMUs; tmu++) { FxU32 start = fxMesa->Glide.grTexMinAddress(tmu); FxU32 end = fxMesa->Glide.grTexMaxAddress(tmu); shared->totalTexMem[tmu] = end - start; shared->freeTexMem[tmu] = end - start; shared->tmFree[tmu] = NewRangeNode(fxMesa, start, end); /*printf("Split tex memory: %d\n", (int) (end - start));*/ } UNLOCK_HARDWARE(fxMesa); } shared->tmPool = NULL; fxMesa->glCtx->Shared->DriverData = shared; /*printf("Texture memory init UMA: %d\n", shared->umaTexMemory);*/ } }
static void sisRunPipeline( GLcontext *ctx ) { sisContextPtr smesa = SIS_CONTEXT( ctx ); LOCK_HARDWARE(); sisUpdateHWState( ctx ); if (smesa->AGPCmdModeEnabled) { AGP_WritePtr = (GLfloat *)smesa->AGPCmdBufBase + *smesa->pAGPCmdBufNext; AGP_StartPtr = AGP_WritePtr; AGP_ReadPtr = (GLfloat *)((long)MMIO_READ(REG_3D_AGPCmBase) - (long)smesa->AGPCmdBufAddr + (long)smesa->AGPCmdBufBase); sisUpdateAGP( smesa ); } if (!smesa->Fallback && smesa->NewGLState) { if (smesa->NewGLState & _SIS_NEW_VERTEX_STATE) sisChooseVertexState( ctx ); if (smesa->NewGLState & (_SIS_NEW_RENDER_STATE | _NEW_TEXTURE)) sisChooseRenderState( ctx ); smesa->NewGLState = 0; } _tnl_run_pipeline( ctx ); if (smesa->AGPCmdModeEnabled) sisFireVertsAGP( smesa ); else mEndPrimitive(); UNLOCK_HARDWARE(); }
void sisSpanRenderFinish( GLcontext *ctx ) { sisContextPtr smesa = SIS_CONTEXT(ctx); _swrast_flush( ctx ); UNLOCK_HARDWARE(); }
/* 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) { _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); } LOCK_HARDWARE(intel); _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); UNLOCK_HARDWARE(intel); }
/* 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 ok; intel_region_release_pbo(intel, region); assert(region->cpp * region->pitch * region->height == pbo->Base.Size); _DBG("%s %p (%d bytes)\n", __FUNCTION__, region, pbo->Base.Size); /* Now blit from the texture buffer to the new buffer: */ LOCK_HARDWARE(intel); ok = intelEmitCopyBlit(intel, region->cpp, region->pitch, pbo->buffer, 0, region->tiling, region->pitch, region->buffer, 0, region->tiling, 0, 0, 0, 0, region->pitch, region->height, GL_COPY); assert(ok); UNLOCK_HARDWARE(intel); }
static void tdfxEndQuery(GLcontext *ctx, struct gl_query_object *q) { tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx); FxI32 total_pixels; FxI32 z_fail_pixels; if (q->Target == GL_SAMPLES_PASSED_ARB) { LOCK_HARDWARE(fxMesa); fxMesa->Glide.grFinish(); fxMesa->Glide.grGet(GR_STATS_PIXELS_DEPTHFUNC_FAIL, sizeof(FxI32), &z_fail_pixels); fxMesa->Glide.grGet(GR_STATS_PIXELS_IN, sizeof(FxI32), &total_pixels); q->Result = total_pixels - z_fail_pixels; /* Apparently, people have seen z_fail_pixels > total_pixels under * some conditions on some 3Dfx hardware. The occlusion query spec * requires that we clamp to 0. */ if (q->Result < 0) { q->Result = 0; } q->Ready = GL_TRUE; UNLOCK_HARDWARE(fxMesa); } }
void i830WaitAge( i830ContextPtr imesa, int age ) { int i = 0; if (GET_DISPATCH_AGE(imesa) >= age) return; while (1) { drmCommandNone(imesa->driFd, DRM_I830_GETAGE); if (GET_DISPATCH_AGE(imesa) >= age) return; imesa->perf_boxes |= I830_BOX_WAIT; if (imesa->do_irqs) { drmI830IrqEmit ie; drmI830IrqWait iw; int ret; ie.irq_seq = &iw.irq_seq; LOCK_HARDWARE( imesa ); ret = drmCommandWriteRead( imesa->driFd, DRM_I830_IRQ_EMIT, &ie, sizeof(ie) ); if ( ret ) { fprintf( stderr, "%s: drmI830IrqEmit: %d\n", __FUNCTION__, ret ); exit(1); } UNLOCK_HARDWARE(imesa); ret = drmCommandWrite( imesa->driFd, DRM_I830_IRQ_WAIT, &iw, sizeof(iw) ); if ( ret ) { fprintf( stderr, "%s: drmI830IrqWait: %d\n", __FUNCTION__, ret ); exit(1); } } else { if (++i > 5000) usleep(1); } } }
void r128FireBlitLocked( r128ContextPtr rmesa, drmBufPtr buffer, GLint offset, GLint pitch, GLint format, GLint x, GLint y, GLint width, GLint height ) { drm_r128_blit_t blit; GLint ret; blit.idx = buffer->idx; blit.offset = offset; blit.pitch = pitch; blit.format = format; blit.x = x; blit.y = y; blit.width = width; blit.height = height; ret = drmCommandWrite( rmesa->driFd, DRM_R128_BLIT, &blit, sizeof(blit) ); if ( ret ) { UNLOCK_HARDWARE( rmesa ); fprintf( stderr, "DRM_R128_BLIT: return = %d\n", ret ); exit( 1 ); } }
static void ffbSWRenderFinish(GLcontext *ctx) { ffbContextPtr fmesa = FFB_CONTEXT(ctx); UNLOCK_HARDWARE(fmesa); fmesa->hw_locked = 0; }
/** * 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); /* Now unmap the framebuffer: */ #if 0 intel_region_unmap(intel, intel->front_region); intel_region_unmap(intel, intel->back_region); intel_region_unmap(intel, intel->intelScreen->depth_region); #endif 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); }
/* 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); }
void *intelAllocateAGP( intelContextPtr intel, GLsizei size ) { int region_offset; drmI830MemAlloc alloc; int ret; if (0) fprintf(stderr, "%s: %d bytes\n", __FUNCTION__, size); alloc.region = I830_MEM_REGION_AGP; alloc.alignment = 0; alloc.size = size; alloc.region_offset = ®ion_offset; LOCK_HARDWARE(intel); /* Make sure the global heap is initialized */ if (intel->texture_heaps[0]) driAgeTextures( intel->texture_heaps[0] ); ret = drmCommandWriteRead( intel->driFd, DRM_I830_ALLOC, &alloc, sizeof(alloc)); if (ret) { fprintf(stderr, "%s: DRM_I830_ALLOC ret %d\n", __FUNCTION__, ret); UNLOCK_HARDWARE(intel); return NULL; } if (0) fprintf(stderr, "%s: allocated %d bytes\n", __FUNCTION__, size); /* Need to propogate this information (agp memory in use) to our * local texture lru. The kernel has already updated the global * lru. An alternative would have been to allocate memory the * usual way and then notify the kernel to pin the allocation. */ if (intel->texture_heaps[0]) driAgeTextures( intel->texture_heaps[0] ); UNLOCK_HARDWARE(intel); return (void *)((char *)intel->intelScreen->tex.map + region_offset); }
static void ffbDDFinish(GLcontext *ctx) { ffbContextPtr fmesa = FFB_CONTEXT(ctx); LOCK_HARDWARE(fmesa); FFBWait(fmesa, fmesa->regs); UNLOCK_HARDWARE(fmesa); }
void i830FlushPrims( i830ContextPtr imesa ) { if (imesa->vertex_buffer) { LOCK_HARDWARE( imesa ); i830FlushPrimsLocked( imesa ); UNLOCK_HARDWARE( imesa ); } }
static void r128SpanRenderFinish( GLcontext *ctx ) { r128ContextPtr rmesa = R128_CONTEXT(ctx); _swrast_flush( ctx ); r128WaitForIdleLocked( rmesa ); UNLOCK_HARDWARE( rmesa ); }
static void nouveauSpanRenderFinish( GLcontext *ctx ) { nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); _swrast_flush( ctx ); nouveauWaitForIdleLocked( nmesa ); UNLOCK_HARDWARE( nmesa ); }
/* This waits for *everybody* to finish rendering -- overkill. */ void i810DmaFinish( i810ContextPtr imesa ) { I810_FIREVERTICES( imesa ); LOCK_HARDWARE( imesa ); i810RegetLockQuiescent( imesa ); UNLOCK_HARDWARE( imesa ); }
/* Flip the front & back buffes */ void i830PageFlip( const __DRIdrawablePrivate *dPriv ) { i830ContextPtr imesa; int tmp, ret; if (I830_DEBUG & DEBUG_IOCTL) fprintf(stderr, "%s\n", __FUNCTION__); assert(dPriv); assert(dPriv->driContextPriv); assert(dPriv->driContextPriv->driverPrivate); imesa = (i830ContextPtr) dPriv->driContextPriv->driverPrivate; I830_FIREVERTICES( imesa ); LOCK_HARDWARE( imesa ); imesa->sarea->perf_boxes |= imesa->perf_boxes; imesa->perf_boxes = 0; if (dPriv->pClipRects) { *(XF86DRIClipRectRec *)imesa->sarea->boxes = dPriv->pClipRects[0]; imesa->sarea->nbox = 1; } ret = drmCommandNone(imesa->driFd, DRM_I830_FLIP); if (ret) { fprintf(stderr, "%s: %d\n", __FUNCTION__, ret); UNLOCK_HARDWARE( imesa ); exit(1); } tmp = GET_ENQUEUE_AGE(imesa); UNLOCK_HARDWARE( imesa ); /* multiarb will suck the life out of the server without this throttle: */ if (GET_DISPATCH_AGE(imesa) < imesa->lastSwap) { i830WaitAge(imesa, imesa->lastSwap); } i830SetDrawBuffer( imesa->glCtx, imesa->glCtx->Color.DriverDrawBuffer ); imesa->upload_cliprects = GL_TRUE; imesa->lastSwap = tmp; }
/* Make sure all commands have been sent to the hardware and have * completed processing. */ static void sisFinish( GLcontext *ctx ) { sisContextPtr smesa = SIS_CONTEXT(ctx); SIS_FIREVERTICES(smesa); LOCK_HARDWARE(); WaitEngIdle( smesa ); UNLOCK_HARDWARE(); }
static void ffbBufferSize(GLframebuffer *buffer, GLuint *width, GLuint *height) { GET_CURRENT_CONTEXT(ctx); ffbContextPtr fmesa = FFB_CONTEXT(ctx); LOCK_HARDWARE(fmesa); *width = fmesa->driDrawable->w; *height = fmesa->driDrawable->h; UNLOCK_HARDWARE(fmesa); }
static void tdfxFlush( GLcontext *ctx ) { tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx); FLUSH_BATCH( fxMesa ); LOCK_HARDWARE( fxMesa ); fxMesa->Glide.grFlush(); UNLOCK_HARDWARE( fxMesa ); }
/* Send all commands to the hardware. If vertex buffers or indirect * buffers are in use, then we need to make sure they are sent to the * hardware. All commands that are normally sent to the ring are * already considered `flushed'. */ static void mach64DDFlush( GLcontext *ctx ) { mach64ContextPtr mmesa = MACH64_CONTEXT(ctx); LOCK_HARDWARE( mmesa ); FLUSH_DMA_LOCKED( mmesa ); UNLOCK_HARDWARE( mmesa ); #if ENABLE_PERF_BOXES if ( mmesa->boxes ) { LOCK_HARDWARE( mmesa ); mach64PerformanceBoxesLocked( mmesa ); UNLOCK_HARDWARE( mmesa ); } /* Log the performance counters if necessary */ mach64PerformanceCounters( mmesa ); #endif }
void mgaWaitAge( mgaContextPtr mmesa, int age ) { if (GET_DISPATCH_AGE(mmesa) < age) { LOCK_HARDWARE(mmesa); if (GET_DISPATCH_AGE(mmesa) < age) { UPDATE_LOCK( mmesa, DRM_LOCK_FLUSH ); } UNLOCK_HARDWARE(mmesa); } }
/* 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); { struct _DriBufferObject *src_buffer = intel_bufferobj_buffer(intel, pbo, INTEL_READ); struct _DriBufferObject *dst_buffer = intel_region_buffer(intel->intelScreen, intelImage->mt->region, INTEL_WRITE_FULL); intelEmitCopyBlit(intel, intelImage->mt->cpp, src_stride, src_buffer, src_offset, dst_stride, dst_buffer, dst_offset, 0, 0, 0, 0, width, height, GL_COPY); intel_batchbuffer_flush(intel->batch); } UNLOCK_HARDWARE(intel); return GL_TRUE; }
/* Return the width and height of the current color buffer. */ static void tridentDDGetBufferSize( GLframebuffer *framebuffer, GLuint *width, GLuint *height ) { GET_CURRENT_CONTEXT(ctx); tridentContextPtr tmesa = TRIDENT_CONTEXT(ctx); LOCK_HARDWARE(tmesa); *width = tmesa->driDrawable->w; *height = tmesa->driDrawable->h; UNLOCK_HARDWARE(tmesa); }
/* Return the current color buffer size. */ static void mach64DDGetBufferSize( GLframebuffer *buffer, GLuint *width, GLuint *height ) { GET_CURRENT_CONTEXT(ctx); mach64ContextPtr mmesa = MACH64_CONTEXT(ctx); LOCK_HARDWARE( mmesa ); *width = mmesa->driDrawable->w; *height = mmesa->driDrawable->h; UNLOCK_HARDWARE( mmesa ); }