summaryrefslogtreecommitdiff
path: root/linux-core
diff options
context:
space:
mode:
authorDave Airlie <airlied@redhat.com>2008-08-26 17:44:47 +1000
committerDave Airlie <airlied@redhat.com>2008-08-26 17:44:47 +1000
commit01b2e7368a1f6f14b6d7b6377c804d482bb050cb (patch)
tree42ab383391226f3b39f8b602392b95311a6eb1b4 /linux-core
parent203243eea5c6c91e89534cd01a1e52cc44980bfe (diff)
radeon/ttm: add support for zeroing the contents of VRAM buffers
This uses a solid fill fastpath, falling back to the slow memset path.
Diffstat (limited to 'linux-core')
-rw-r--r--linux-core/radeon_buffer.c277
1 files changed, 224 insertions, 53 deletions
diff --git a/linux-core/radeon_buffer.c b/linux-core/radeon_buffer.c
index 900d450a..96a584f3 100644
--- a/linux-core/radeon_buffer.c
+++ b/linux-core/radeon_buffer.c
@@ -37,7 +37,7 @@ struct drm_ttm_backend *radeon_create_ttm_backend_entry(struct drm_device * dev)
{
drm_radeon_private_t *dev_priv = dev->dev_private;
- if(dev_priv->flags & RADEON_IS_AGP)
+ if (dev_priv->flags & RADEON_IS_AGP)
return drm_agp_init_ttm(dev);
else
return ati_pcigart_init_ttm(dev, &dev_priv->gart_info, radeon_gart_flush);
@@ -58,9 +58,10 @@ int radeon_invalidate_caches(struct drm_device * dev, uint64_t flags)
if (!dev_priv->cp_running)
return 0;
- BEGIN_RING(4);
- RADEON_FLUSH_CACHE();
- RADEON_FLUSH_ZCACHE();
+ BEGIN_RING(6);
+ RADEON_PURGE_CACHE();
+ RADEON_PURGE_ZCACHE();
+ RADEON_WAIT_UNTIL_3D_IDLE();
ADVANCE_RING();
COMMIT_RING();
return 0;
@@ -112,15 +113,17 @@ int radeon_init_mem_type(struct drm_device * dev, uint32_t type,
return 0;
}
-static void radeon_emit_copy_blit(struct drm_device * dev,
- uint32_t src_offset,
- uint32_t dst_offset,
- uint32_t pages, int direction)
+void radeon_emit_copy_blit(struct drm_device * dev,
+ uint32_t src_offset,
+ uint32_t dst_offset,
+ uint32_t pages)
{
uint32_t cur_pages;
- uint32_t stride = PAGE_SIZE;
+ uint32_t stride_bytes = PAGE_SIZE;
drm_radeon_private_t *dev_priv = dev->dev_private;
- uint32_t format, height;
+ uint32_t format, pitch;
+ const uint32_t clip = (0x1fff) | (0x1fff << 16);
+ uint32_t stride_pixels;
RING_LOCALS;
if (!dev_priv)
@@ -130,67 +133,171 @@ static void radeon_emit_copy_blit(struct drm_device * dev,
format = RADEON_COLOR_FORMAT_ARGB8888;
/* radeon limited to 16k stride */
- stride &= 0x3fff;
+ stride_bytes &= 0x3fff;
+ /* radeon pitch is /64 */
+ pitch = stride_bytes / 64;
+
+ stride_pixels = stride_bytes / 4;
+
while(pages > 0) {
cur_pages = pages;
- if (cur_pages > 2048)
- cur_pages = 2048;
+ if (cur_pages > 8191)
+ cur_pages = 8191;
pages -= cur_pages;
- /* needs verification */
- BEGIN_RING(7);
- OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
+ /* pages are in Y direction - height
+ page width in X direction - width */
+ BEGIN_RING(10);
+ OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 8));
OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
RADEON_GMC_DST_PITCH_OFFSET_CNTL |
+ RADEON_GMC_SRC_CLIPPING | RADEON_GMC_DST_CLIPPING |
RADEON_GMC_BRUSH_NONE |
(format << 8) |
RADEON_GMC_SRC_DATATYPE_COLOR |
RADEON_ROP3_S |
RADEON_DP_SRC_SOURCE_MEMORY |
RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
- if (direction) {
- OUT_RING((stride << 22) | (src_offset >> 10));
- OUT_RING((stride << 22) | (dst_offset >> 10));
- } else {
- OUT_RING((stride << 22) | (dst_offset >> 10));
- OUT_RING((stride << 22) | (src_offset >> 10));
- }
- OUT_RING(0);
+ OUT_RING((pitch << 22) | (src_offset >> 10));
+ OUT_RING((pitch << 22) | (dst_offset >> 10));
+ OUT_RING(clip); // SRC _SC BOT_RITE
+ OUT_RING(0); // SC_TOP_LEFT
+ OUT_RING(clip); // SC_BOT_RITE
+
+ OUT_RING(pages);
OUT_RING(pages); /* x - y */
- OUT_RING((stride << 16) | cur_pages);
+ OUT_RING(cur_pages | (stride_pixels << 16));
ADVANCE_RING();
}
- BEGIN_RING(2);
+ BEGIN_RING(4);
+ OUT_RING(CP_PACKET0(RADEON_RB2D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(RADEON_RB2D_DC_FLUSH_ALL);
RADEON_WAIT_UNTIL_2D_IDLE();
ADVANCE_RING();
+ COMMIT_RING();
return;
}
-static int radeon_move_blit(struct drm_buffer_object * bo,
- int evict, int no_wait, struct drm_bo_mem_reg *new_mem)
+int radeon_move_blit(struct drm_buffer_object * bo,
+ int evict, int no_wait, struct drm_bo_mem_reg *new_mem,
+ struct drm_bo_mem_reg *old_mem)
{
- struct drm_bo_mem_reg *old_mem = &bo->mem;
- int dir = 0;
+ struct drm_device *dev = bo->dev;
+ drm_radeon_private_t *dev_priv = dev->dev_private;
+ uint32_t old_start, new_start;
- if ((old_mem->mem_type == new_mem->mem_type) &&
- (new_mem->mm_node->start <
- old_mem->mm_node->start + old_mem->mm_node->size)) {
- dir = 1;
- }
+ old_start = old_mem->mm_node->start << PAGE_SHIFT;
+ new_start = new_mem->mm_node->start << PAGE_SHIFT;
+
+ if (old_mem->mem_type == DRM_BO_MEM_VRAM)
+ old_start += dev_priv->fb_location;
+ if (old_mem->mem_type == DRM_BO_MEM_TT)
+ old_start += dev_priv->gart_vm_start;
+
+ if (new_mem->mem_type == DRM_BO_MEM_VRAM)
+ new_start += dev_priv->fb_location;
+ if (new_mem->mem_type == DRM_BO_MEM_TT)
+ new_start += dev_priv->gart_vm_start;
radeon_emit_copy_blit(bo->dev,
- old_mem->mm_node->start << PAGE_SHIFT,
- new_mem->mm_node->start << PAGE_SHIFT,
- new_mem->num_pages, dir);
+ old_start,
+ new_start,
+ new_mem->num_pages);
+
+ /* invalidate the chip caches */
-
return drm_bo_move_accel_cleanup(bo, evict, no_wait, 0,
DRM_FENCE_TYPE_EXE, 0,
new_mem);
}
+void radeon_emit_solid_fill(struct drm_device * dev,
+ uint32_t dst_offset,
+ uint32_t pages, uint8_t value)
+{
+ uint32_t cur_pages;
+ uint32_t stride_bytes = PAGE_SIZE;
+ drm_radeon_private_t *dev_priv = dev->dev_private;
+ uint32_t format, pitch;
+ const uint32_t clip = (0x1fff) | (0x1fff << 16);
+ uint32_t stride_pixels;
+ RING_LOCALS;
+
+ if (!dev_priv)
+ return;
+
+ /* 32-bit copy format */
+ format = RADEON_COLOR_FORMAT_ARGB8888;
+
+ /* radeon limited to 16k stride */
+ stride_bytes &= 0x3fff;
+ /* radeon pitch is /64 */
+ pitch = stride_bytes / 64;
+
+ stride_pixels = stride_bytes / 4;
+
+ while(pages > 0) {
+ cur_pages = pages;
+ if (cur_pages > 8191)
+ cur_pages = 8191;
+ pages -= cur_pages;
+
+ /* pages are in Y direction - height
+ page width in X direction - width */
+ BEGIN_RING(8);
+ OUT_RING(CP_PACKET3(RADEON_CNTL_PAINT_MULTI, 6));
+ OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
+ RADEON_GMC_DST_CLIPPING |
+ RADEON_GMC_BRUSH_SOLID_COLOR |
+ (format << 8) |
+ RADEON_ROP3_S |
+ RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
+ OUT_RING((pitch << 22) | (dst_offset >> 10)); // PITCH
+ OUT_RING(0); // SC_TOP_LEFT // DST CLIPPING
+ OUT_RING(clip); // SC_BOT_RITE
+
+ OUT_RING(0); // COLOR
+
+ OUT_RING(pages); /* x - y */
+ OUT_RING(cur_pages | (stride_pixels << 16));
+ ADVANCE_RING();
+ }
+
+ BEGIN_RING(4);
+ OUT_RING(CP_PACKET0(RADEON_RB2D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(RADEON_RB2D_DC_FLUSH_ALL);
+ RADEON_WAIT_UNTIL_2D_IDLE();
+ ADVANCE_RING();
+
+ COMMIT_RING();
+ return;
+}
+
+int radeon_move_zero_fill(struct drm_buffer_object * bo,
+ int evict, int no_wait, struct drm_bo_mem_reg *new_mem)
+{
+ struct drm_device *dev = bo->dev;
+ drm_radeon_private_t *dev_priv = dev->dev_private;
+ uint32_t new_start;
+
+ new_start = new_mem->mm_node->start << PAGE_SHIFT;
+
+ if (new_mem->mem_type == DRM_BO_MEM_VRAM)
+ new_start += dev_priv->fb_location;
+
+ radeon_emit_solid_fill(bo->dev,
+ new_start,
+ new_mem->num_pages, 0);
+
+ /* invalidate the chip caches */
+
+ return drm_bo_move_accel_cleanup(bo, 1, no_wait, 0,
+ DRM_FENCE_TYPE_EXE, 0,
+ new_mem);
+}
+
static int radeon_move_flip(struct drm_buffer_object * bo,
int evict, int no_wait, struct drm_bo_mem_reg * new_mem)
{
@@ -200,8 +307,7 @@ static int radeon_move_flip(struct drm_buffer_object * bo,
tmp_mem = *new_mem;
tmp_mem.mm_node = NULL;
- // tmp_mem.mask = DRM_BO_FLAG_MEM_TT |
- // DRM_BO_FLAG_CACHED | DRM_BO_FLAG_FORCE_CACHING;
+ tmp_mem.proposed_flags = DRM_BO_FLAG_MEM_TT;
ret = drm_bo_mem_space(bo, &tmp_mem, no_wait);
if (ret)
@@ -211,7 +317,7 @@ static int radeon_move_flip(struct drm_buffer_object * bo,
if (ret)
goto out_cleanup;
- ret = radeon_move_blit(bo, 1, no_wait, &tmp_mem);
+ ret = radeon_move_blit(bo, 1, no_wait, &tmp_mem, &bo->mem);
if (ret)
goto out_cleanup;
@@ -227,25 +333,90 @@ out_cleanup:
return ret;
}
+static int radeon_move_vram(struct drm_buffer_object * bo,
+ int evict, int no_wait, struct drm_bo_mem_reg * new_mem)
+{
+ struct drm_device *dev = bo->dev;
+ struct drm_bo_mem_reg tmp_mem;
+ struct drm_bo_mem_reg *old_mem = &bo->mem;
+ int ret;
+ bool was_local = false;
+
+ /* old - LOCAL memory node bo->mem
+ tmp - TT type memory node
+ new - VRAM memory node */
+
+ tmp_mem = *old_mem;
+ tmp_mem.mm_node = NULL;
+
+ if (old_mem->mem_type == DRM_BO_MEM_LOCAL) {
+ tmp_mem.proposed_flags = DRM_BO_FLAG_MEM_TT;
+
+ ret = drm_bo_mem_space(bo, &tmp_mem, no_wait);
+ if (ret)
+ return ret;
+ }
+
+ if (!bo->ttm) {
+ ret = drm_bo_add_ttm(bo);
+ if (ret)
+ goto out_cleanup;
+ }
+
+ if (old_mem->mem_type == DRM_BO_MEM_LOCAL) {
+ ret = drm_bo_move_ttm(bo, evict, no_wait, &tmp_mem);
+ if (ret)
+ return ret;
+ }
+
+ ret = radeon_move_blit(bo, 1, no_wait, new_mem, &bo->mem);
+ if (ret)
+ goto out_cleanup;
+
+out_cleanup:
+ if (tmp_mem.mm_node) {
+ mutex_lock(&dev->struct_mutex);
+ if (tmp_mem.mm_node != bo->pinned_node)
+ drm_mm_put_block(tmp_mem.mm_node);
+ tmp_mem.mm_node = NULL;
+ mutex_unlock(&dev->struct_mutex);
+ }
+ return ret;
+}
+
int radeon_move(struct drm_buffer_object * bo,
- int evict, int no_wait, struct drm_bo_mem_reg * new_mem)
+ int evict, int no_wait, struct drm_bo_mem_reg *new_mem)
{
+ struct drm_device *dev = bo->dev;
struct drm_bo_mem_reg *old_mem = &bo->mem;
+ drm_radeon_private_t *dev_priv = dev->dev_private;
- return drm_bo_move_memcpy(bo, evict, no_wait, new_mem);
-#if 0
- DRM_DEBUG("\n");
- if (old_mem->mem_type == DRM_BO_MEM_LOCAL) {
- return drm_bo_move_memcpy(bo, evict, no_wait, new_mem);
- } else if (new_mem->mem_type == DRM_BO_MEM_LOCAL) {
+ if (!dev_priv->cp_running)
+ goto fallback;
+
+ if (bo->mem.flags & DRM_BO_FLAG_CLEAN) /* need to implement solid fill */
+ {
+ if (radeon_move_zero_fill(bo, evict, no_wait, new_mem))
+ return drm_bo_move_zero(bo, evict, no_wait, new_mem);
+ return 0;
+ }
+
+ if (new_mem->mem_type == DRM_BO_MEM_VRAM) {
+ if (radeon_move_vram(bo, evict, no_wait, new_mem))
+ goto fallback;
+ } else if (new_mem->mem_type == DRM_BO_MEM_LOCAL){
if (radeon_move_flip(bo, evict, no_wait, new_mem))
- return drm_bo_move_memcpy(bo, evict, no_wait, new_mem);
+ goto fallback;
} else {
- if (radeon_move_blit(bo, evict, no_wait, new_mem))
- return drm_bo_move_memcpy(bo, evict, no_wait, new_mem);
+ if (radeon_move_flip(bo, evict, no_wait, new_mem))
+ goto fallback;
}
return 0;
-#endif
+fallback:
+ if (bo->mem.flags & DRM_BO_FLAG_CLEAN)
+ return drm_bo_move_zero(bo, evict, no_wait, new_mem);
+ else
+ return drm_bo_move_memcpy(bo, evict, no_wait, new_mem);
}