/**
* \brief Call this after drawing to mark which buffers need resolving
*
* If the depth buffer was written to and if it has an accompanying HiZ
* buffer, then mark that it needs a depth resolve.
*
* If the color buffer is a multisample window system buffer, then
* mark that it needs a downsample.
*
* Also mark any render targets which will be textured as needing a render
* cache flush.
*/
static void
brw_postdraw_set_buffers_need_resolve(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
struct gl_framebuffer *fb = ctx->DrawBuffer;
struct intel_renderbuffer *front_irb = NULL;
struct intel_renderbuffer *back_irb = intel_get_renderbuffer(fb, BUFFER_BACK_LEFT);
struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencil_irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct gl_renderbuffer_attachment *depth_att = &fb->Attachment[BUFFER_DEPTH];
if (_mesa_is_front_buffer_drawing(fb))
front_irb = intel_get_renderbuffer(fb, BUFFER_FRONT_LEFT);
if (front_irb)
front_irb->need_downsample = true;
if (back_irb)
back_irb->need_downsample = true;
if (depth_irb && ctx->Depth.Mask) {
intel_renderbuffer_att_set_needs_depth_resolve(depth_att);
brw_render_cache_set_add_bo(brw, depth_irb->mt->bo);
}
if (ctx->Extensions.ARB_stencil_texturing &&
stencil_irb && ctx->Stencil._WriteEnabled) {
brw_render_cache_set_add_bo(brw, stencil_irb->mt->bo);
}
for (unsigned i = 0; i < fb->_NumColorDrawBuffers; i++) {
struct intel_renderbuffer *irb =
intel_renderbuffer(fb->_ColorDrawBuffers[i]);
if (irb)
brw_render_cache_set_add_bo(brw, irb->mt->bo);
}
}
bool
brw_blorp_clear_color(struct intel_context *intel, struct gl_framebuffer *fb)
{
struct gl_context *ctx = &intel->ctx;
struct brw_context *brw = brw_context(ctx);
/* The constant color clear code doesn't work for multisampled surfaces, so
* we need to support falling back to other clear mechanisms.
* Unfortunately, our clear code is based on a bitmask that doesn't
* distinguish individual color attachments, so we walk the attachments to
* see if any require fallback, and fall back for all if any of them need
* to.
*/
for (unsigned buf = 0; buf < ctx->DrawBuffer->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[buf];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
if (irb && irb->mt->msaa_layout != INTEL_MSAA_LAYOUT_NONE)
return false;
}
for (unsigned buf = 0; buf < ctx->DrawBuffer->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[buf];
/* If this is an ES2 context or GL_ARB_ES2_compatibility is supported,
* the framebuffer can be complete with some attachments missing. In
* this case the _ColorDrawBuffers pointer will be NULL.
*/
if (rb == NULL)
continue;
brw_blorp_clear_params params(brw, fb, rb, ctx->Color.ColorMask[buf]);
brw_blorp_exec(intel, ¶ms);
}
return true;
}
static void
intelCopyTexSubImage(struct gl_context *ctx, GLuint dims,
struct gl_texture_image *texImage,
GLint xoffset, GLint yoffset, GLint slice,
struct gl_renderbuffer *rb,
GLint x, GLint y,
GLsizei width, GLsizei height)
{
struct intel_context *intel = intel_context(ctx);
/* Try the BLT engine. */
if (intel_copy_texsubimage(intel,
intel_texture_image(texImage),
xoffset, yoffset, slice,
intel_renderbuffer(rb), x, y, width, height)) {
return;
}
/* Otherwise, fall back to meta. This will likely be slow. */
perf_debug("%s - fallback to swrast\n", __FUNCTION__);
_mesa_meta_CopyTexSubImage(ctx, dims, texImage,
xoffset, yoffset, slice,
rb, x, y, width, height);
}
bool
brw_blorp_clear_color(struct brw_context *brw, struct gl_framebuffer *fb,
bool partial_clear)
{
struct gl_context *ctx = &brw->ctx;
/* The constant color clear code doesn't work for multisampled surfaces, so
* we need to support falling back to other clear mechanisms.
* Unfortunately, our clear code is based on a bitmask that doesn't
* distinguish individual color attachments, so we walk the attachments to
* see if any require fallback, and fall back for all if any of them need
* to.
*/
for (unsigned buf = 0; buf < ctx->DrawBuffer->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[buf];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
if (irb && irb->mt->msaa_layout != INTEL_MSAA_LAYOUT_NONE)
return false;
}
for (unsigned buf = 0; buf < ctx->DrawBuffer->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[buf];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
/* If this is an ES2 context or GL_ARB_ES2_compatibility is supported,
* the framebuffer can be complete with some attachments missing. In
* this case the _ColorDrawBuffers pointer will be NULL.
*/
if (rb == NULL)
continue;
brw_blorp_clear_params params(brw, fb, rb, ctx->Color.ColorMask[buf],
partial_clear);
bool is_fast_clear =
(params.fast_clear_op == GEN7_FAST_CLEAR_OP_FAST_CLEAR);
if (is_fast_clear) {
/* Record the clear color in the miptree so that it will be
* programmed in SURFACE_STATE by later rendering and resolve
* operations.
*/
uint32_t new_color_value =
compute_fast_clear_color_bits(&ctx->Color.ClearColor);
if (irb->mt->fast_clear_color_value != new_color_value) {
irb->mt->fast_clear_color_value = new_color_value;
brw->state.dirty.brw |= BRW_NEW_SURFACES;
}
/* If the buffer is already in INTEL_MCS_STATE_CLEAR, the clear is
* redundant and can be skipped.
*/
if (irb->mt->mcs_state == INTEL_MCS_STATE_CLEAR)
continue;
/* If the MCS buffer hasn't been allocated yet, we need to allocate
* it now.
*/
if (!irb->mt->mcs_mt) {
if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt)) {
/* MCS allocation failed--probably this will only happen in
* out-of-memory conditions. But in any case, try to recover
* by falling back to a non-blorp clear technique.
*/
return false;
}
brw->state.dirty.brw |= BRW_NEW_SURFACES;
}
}
DBG("%s to mt %p level %d layer %d\n", __FUNCTION__,
irb->mt, irb->mt_level, irb->mt_layer);
brw_blorp_exec(brw, ¶ms);
if (is_fast_clear) {
/* Now that the fast clear has occurred, put the buffer in
* INTEL_MCS_STATE_CLEAR so that we won't waste time doing redundant
* clears.
*/
irb->mt->mcs_state = INTEL_MCS_STATE_CLEAR;
}
}
return true;
}
/**
* Sets up a surface state structure to point at the given region.
* While it is only used for the front/back buffer currently, it should be
* usable for further buffers when doing ARB_draw_buffer support.
*/
static void
brw_update_renderbuffer_surface(struct brw_context *brw,
struct gl_renderbuffer *rb,
unsigned int unit)
{
struct intel_context *intel = &brw->intel;
GLcontext *ctx = &intel->ctx;
dri_bo *region_bo = NULL;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
struct intel_region *region = irb ? irb->region : NULL;
struct {
unsigned int surface_type;
unsigned int surface_format;
unsigned int width, height, pitch, cpp;
GLubyte color_mask[4];
GLboolean color_blend;
uint32_t tiling;
uint32_t draw_x;
uint32_t draw_y;
} key;
memset(&key, 0, sizeof(key));
if (region != NULL) {
region_bo = region->buffer;
key.surface_type = BRW_SURFACE_2D;
switch (irb->Base.Format) {
/* XRGB and ARGB are treated the same here because the chips in this
* family cannot render to XRGB targets. This means that we have to
* mask writes to alpha (ala glColorMask) and reconfigure the alpha
* blending hardware to use GL_ONE (or GL_ZERO) for cases where
* GL_DST_ALPHA (or GL_ONE_MINUS_DST_ALPHA) is used.
*/
case MESA_FORMAT_ARGB8888:
case MESA_FORMAT_XRGB8888:
key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
break;
case MESA_FORMAT_RGB565:
key.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM;
break;
case MESA_FORMAT_ARGB1555:
key.surface_format = BRW_SURFACEFORMAT_B5G5R5A1_UNORM;
break;
case MESA_FORMAT_ARGB4444:
key.surface_format = BRW_SURFACEFORMAT_B4G4R4A4_UNORM;
break;
default:
_mesa_problem(ctx, "Bad renderbuffer format: %d\n", irb->Base.Format);
}
key.tiling = region->tiling;
if (brw->intel.intelScreen->driScrnPriv->dri2.enabled) {
key.width = rb->Width;
key.height = rb->Height;
} else {
key.width = region->width;
key.height = region->height;
}
key.pitch = region->pitch;
key.cpp = region->cpp;
key.draw_x = region->draw_x;
key.draw_y = region->draw_y;
} else {
key.surface_type = BRW_SURFACE_NULL;
key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
key.tiling = I915_TILING_X;
key.width = 1;
key.height = 1;
key.cpp = 4;
key.draw_x = 0;
key.draw_y = 0;
}
if (intel->gen < 6) {
/* _NEW_COLOR */
memcpy(key.color_mask, ctx->Color.ColorMask[unit],
sizeof(key.color_mask));
/* As mentioned above, disable writes to the alpha component when the
* renderbuffer is XRGB.
*/
if (ctx->DrawBuffer->Visual.alphaBits == 0)
key.color_mask[3] = GL_FALSE;
key.color_blend = (!ctx->Color._LogicOpEnabled &&
(ctx->Color.BlendEnabled & (1 << unit)));
}
dri_bo_unreference(brw->wm.surf_bo[unit]);
brw->wm.surf_bo[unit] = brw_search_cache(&brw->surface_cache,
BRW_SS_SURFACE,
&key, sizeof(key),
®ion_bo, 1,
NULL);
if (brw->wm.surf_bo[unit] == NULL) {
struct brw_surface_state surf;
memset(&surf, 0, sizeof(surf));
surf.ss0.surface_format = key.surface_format;
surf.ss0.surface_type = key.surface_type;
if (key.tiling == I915_TILING_NONE) {
surf.ss1.base_addr = (key.draw_x + key.draw_y * key.pitch) * key.cpp;
} else {
uint32_t tile_base, tile_x, tile_y;
uint32_t pitch = key.pitch * key.cpp;
if (key.tiling == I915_TILING_X) {
tile_x = key.draw_x % (512 / key.cpp);
tile_y = key.draw_y % 8;
tile_base = ((key.draw_y / 8) * (8 * pitch));
tile_base += (key.draw_x - tile_x) / (512 / key.cpp) * 4096;
} else {
/* Y */
tile_x = key.draw_x % (128 / key.cpp);
tile_y = key.draw_y % 32;
tile_base = ((key.draw_y / 32) * (32 * pitch));
tile_base += (key.draw_x - tile_x) / (128 / key.cpp) * 4096;
}
assert(brw->has_surface_tile_offset || (tile_x == 0 && tile_y == 0));
assert(tile_x % 4 == 0);
assert(tile_y % 2 == 0);
/* Note that the low bits of these fields are missing, so
* there's the possibility of getting in trouble.
*/
surf.ss1.base_addr = tile_base;
surf.ss5.x_offset = tile_x / 4;
surf.ss5.y_offset = tile_y / 2;
}
if (region_bo != NULL)
surf.ss1.base_addr += region_bo->offset; /* reloc */
surf.ss2.width = key.width - 1;
surf.ss2.height = key.height - 1;
brw_set_surface_tiling(&surf, key.tiling);
surf.ss3.pitch = (key.pitch * key.cpp) - 1;
if (intel->gen < 6) {
/* _NEW_COLOR */
surf.ss0.color_blend = key.color_blend;
surf.ss0.writedisable_red = !key.color_mask[0];
surf.ss0.writedisable_green = !key.color_mask[1];
surf.ss0.writedisable_blue = !key.color_mask[2];
surf.ss0.writedisable_alpha = !key.color_mask[3];
}
/* Key size will never match key size for textures, so we're safe. */
brw->wm.surf_bo[unit] = brw_upload_cache(&brw->surface_cache,
BRW_SS_SURFACE,
&key, sizeof(key),
®ion_bo, 1,
&surf, sizeof(surf));
if (region_bo != NULL) {
/* We might sample from it, and we might render to it, so flag
* them both. We might be able to figure out from other state
* a more restrictive relocation to emit.
*/
drm_intel_bo_emit_reloc(brw->wm.surf_bo[unit],
offsetof(struct brw_surface_state, ss1),
region_bo,
surf.ss1.base_addr - region_bo->offset,
I915_GEM_DOMAIN_RENDER,
I915_GEM_DOMAIN_RENDER);
}
}
/**
* Do additional "completeness" testing of a framebuffer object.
*/
static void
intel_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb)
{
struct intel_context *intel = intel_context(ctx);
const struct intel_renderbuffer *depthRb =
intel_get_renderbuffer(fb, BUFFER_DEPTH);
const struct intel_renderbuffer *stencilRb =
intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL, *stencil_mt = NULL;
int i;
DBG("%s() on fb %p (%s)\n", __FUNCTION__,
fb, (fb == ctx->DrawBuffer ? "drawbuffer" :
(fb == ctx->ReadBuffer ? "readbuffer" : "other buffer")));
if (depthRb)
depth_mt = depthRb->mt;
if (stencilRb) {
stencil_mt = stencilRb->mt;
if (stencil_mt->stencil_mt)
stencil_mt = stencil_mt->stencil_mt;
}
if (depth_mt && stencil_mt) {
if (depth_mt == stencil_mt) {
/* For true packed depth/stencil (not faked on prefers-separate-stencil
* hardware) we need to be sure they're the same level/layer, since
* we'll be emitting a single packet describing the packed setup.
*/
if (depthRb->mt_level != stencilRb->mt_level ||
depthRb->mt_layer != stencilRb->mt_layer) {
DBG("depth image level/layer %d/%d != stencil image %d/%d\n",
depthRb->mt_level,
depthRb->mt_layer,
stencilRb->mt_level,
stencilRb->mt_layer);
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
}
} else {
if (!intel->has_separate_stencil) {
DBG("separate stencil unsupported\n");
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
}
if (stencil_mt->format != MESA_FORMAT_S8) {
DBG("separate stencil is %s instead of S8\n",
_mesa_get_format_name(stencil_mt->format));
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
}
if (intel->gen < 7 && depth_mt->hiz_mt == NULL) {
/* Before Gen7, separate depth and stencil buffers can be used
* only if HiZ is enabled. From the Sandybridge PRM, Volume 2,
* Part 1, Bit 3DSTATE_DEPTH_BUFFER.SeparateStencilBufferEnable:
* [DevSNB]: This field must be set to the same value (enabled
* or disabled) as Hierarchical Depth Buffer Enable.
*/
DBG("separate stencil without HiZ\n");
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED;
}
}
}
for (i = 0; i < Elements(fb->Attachment); i++) {
struct gl_renderbuffer *rb;
struct intel_renderbuffer *irb;
if (fb->Attachment[i].Type == GL_NONE)
continue;
/* A supported attachment will have a Renderbuffer set either
* from being a Renderbuffer or being a texture that got the
* intel_wrap_texture() treatment.
*/
rb = fb->Attachment[i].Renderbuffer;
if (rb == NULL) {
DBG("attachment without renderbuffer\n");
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
continue;
}
if (fb->Attachment[i].Type == GL_TEXTURE) {
const struct gl_texture_image *img =
_mesa_get_attachment_teximage_const(&fb->Attachment[i]);
if (img->Border) {
DBG("texture with border\n");
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
continue;
}
}
irb = intel_renderbuffer(rb);
if (irb == NULL) {
DBG("software rendering renderbuffer\n");
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
continue;
}
if (!intel->vtbl.render_target_supported(intel, rb)) {
DBG("Unsupported HW texture/renderbuffer format attached: %s\n",
_mesa_get_format_name(intel_rb_format(irb)));
fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT;
}
}
}
/**
* 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(struct gl_context * 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;
if (!fb) {
/* this can happen during the initial context initialization */
return;
}
/* Do this here, not core Mesa, since this function is called from
* many places within the driver.
*/
if (ctx->NewState & _NEW_BUFFERS) {
/* 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;
}
/* How many color buffers are we drawing into?
*
* If there are zero buffers or the buffer is too big, don't configure any
* regions for hardware drawing. We'll fallback to software below. Not
* having regions set makes some of the software fallback paths faster.
*/
if ((fb->Width > ctx->Const.MaxRenderbufferSize)
|| (fb->Height > ctx->Const.MaxRenderbufferSize)
|| (fb->_NumColorDrawBuffers == 0)) {
/* writing to 0 */
colorRegions[0] = NULL;
}
else if (fb->_NumColorDrawBuffers > 1) {
int i;
struct intel_renderbuffer *irb;
for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
irb = intel_renderbuffer(fb->_ColorDrawBuffers[i]);
colorRegions[i] = irb ? irb->region : NULL;
}
}
else {
/* Get the intel_renderbuffer for the single colorbuffer we're drawing
* into.
*/
if (fb->Name == 0) {
/* drawing to window system buffer */
if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT)
colorRegions[0] = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
else
colorRegions[0] = intel_get_rb_region(fb, BUFFER_BACK_LEFT);
}
else {
/* drawing to user-created FBO */
struct intel_renderbuffer *irb;
irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]);
colorRegions[0] = (irb && irb->region) ? irb->region : NULL;
}
}
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.
***/
if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) {
irbStencil = intel_renderbuffer(fb->_StencilBuffer->Wrapped);
if (irbStencil && irbStencil->region) {
ASSERT(irbStencil->Base.Format == MESA_FORMAT_S8_Z24);
FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE);
}
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);
}
/* If we have a (packed) stencil buffer attached but no depth buffer,
* we still need to set up the shared depth/stencil state so we can use it.
*/
if (depthRegion == NULL && irbStencil && irbStencil->region)
depthRegion = irbStencil->region;
/*
* Update depth and stencil test state
*/
if (ctx->Driver.Enable) {
ctx->Driver.Enable(ctx, GL_DEPTH_TEST,
(ctx->Depth.Test && fb->Visual.depthBits > 0));
ctx->Driver.Enable(ctx, GL_STENCIL_TEST,
(ctx->Stencil.Enabled && fb->Visual.stencilBits > 0));
}
else {
/* Mesa's Stencil._Enabled field is updated when
* _NEW_BUFFERS | _NEW_STENCIL, but i965 code assumes that the value
* only changes with _NEW_STENCIL (which seems sensible). So flag it
* here since this is the _NEW_BUFFERS path.
*/
intel->NewGLState |= (_NEW_DEPTH | _NEW_STENCIL);
}
intel->vtbl.set_draw_region(intel, colorRegions, depthRegion,
fb->_NumColorDrawBuffers);
intel->NewGLState |= _NEW_BUFFERS;
/* update viewport since it depends on window size */
#ifdef I915
intelCalcViewport(ctx);
#else
intel->NewGLState |= _NEW_VIEWPORT;
#endif
/* 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);
/* Update culling direction which changes depending on the
* orientation of the buffer:
*/
if (ctx->Driver.FrontFace)
ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
else
intel->NewGLState |= _NEW_POLYGON;
}
/**
* CopyPixels with the blitter. Don't support zooming, pixel transfer, etc.
*/
static GLboolean
do_blit_copypixels(struct gl_context * ctx,
GLint srcx, GLint srcy,
GLsizei width, GLsizei height,
GLint dstx, GLint dsty, GLenum type)
{
struct intel_context *intel = intel_context(ctx);
struct gl_framebuffer *fb = ctx->DrawBuffer;
struct gl_framebuffer *read_fb = ctx->ReadBuffer;
GLint orig_dstx;
GLint orig_dsty;
GLint orig_srcx;
GLint orig_srcy;
GLboolean flip = GL_FALSE;
struct intel_renderbuffer *draw_irb = NULL;
struct intel_renderbuffer *read_irb = NULL;
/* Update draw buffer bounds */
_mesa_update_state(ctx);
switch (type) {
case GL_COLOR:
if (fb->_NumColorDrawBuffers != 1) {
fallback_debug("glCopyPixels() fallback: MRT\n");
return GL_FALSE;
}
draw_irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]);
read_irb = intel_renderbuffer(read_fb->_ColorReadBuffer);
break;
case GL_DEPTH_STENCIL_EXT:
draw_irb = intel_renderbuffer(fb->Attachment[BUFFER_DEPTH].Renderbuffer);
read_irb =
intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
break;
case GL_DEPTH:
fallback_debug("glCopyPixels() fallback: GL_DEPTH\n");
return GL_FALSE;
case GL_STENCIL:
fallback_debug("glCopyPixels() fallback: GL_STENCIL\n");
return GL_FALSE;
default:
fallback_debug("glCopyPixels(): Unknown type\n");
return GL_FALSE;
}
if (!draw_irb) {
fallback_debug("glCopyPixels() fallback: missing draw buffer\n");
return GL_FALSE;
}
if (!read_irb) {
fallback_debug("glCopyPixels() fallback: missing read buffer\n");
return GL_FALSE;
}
if (draw_irb->Base.Format != read_irb->Base.Format &&
!(draw_irb->Base.Format == MESA_FORMAT_XRGB8888 &&
read_irb->Base.Format == MESA_FORMAT_ARGB8888)) {
fallback_debug("glCopyPixels() fallback: mismatched formats (%s -> %s\n",
_mesa_get_format_name(read_irb->Base.Format),
_mesa_get_format_name(draw_irb->Base.Format));
return GL_FALSE;
}
/* 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;
intel_prepare_render(intel);
intel_flush(&intel->ctx);
/* Clip to destination buffer. */
orig_dstx = dstx;
orig_dsty = dsty;
if (!_mesa_clip_to_region(fb->_Xmin, fb->_Ymin,
fb->_Xmax, fb->_Ymax,
&dstx, &dsty, &width, &height))
goto out;
/* Adjust src coords for our post-clipped destination origin */
srcx += dstx - orig_dstx;
srcy += dsty - orig_dsty;
/* Clip to source buffer. */
orig_srcx = srcx;
orig_srcy = srcy;
if (!_mesa_clip_to_region(0, 0,
read_fb->Width, read_fb->Height,
&srcx, &srcy, &width, &height))
goto out;
/* Adjust dst coords for our post-clipped source origin */
dstx += srcx - orig_srcx;
dsty += srcy - orig_srcy;
/* Flip dest Y if it's a window system framebuffer. */
if (fb->Name == 0) {
/* copypixels to a window system framebuffer */
dsty = fb->Height - dsty - height;
flip = !flip;
}
/* Flip source Y if it's a window system framebuffer. */
if (read_fb->Name == 0) {
srcy = read_fb->Height - srcy - height;
flip = !flip;
}
srcx += read_irb->draw_x;
srcy += read_irb->draw_y;
dstx += draw_irb->draw_x;
dsty += draw_irb->draw_y;
if (!intel_region_copy(intel,
draw_irb->region, 0, dstx, dsty,
read_irb->region, 0, srcx, srcy,
width, height, flip,
ctx->Color.ColorLogicOpEnabled ?
ctx->Color.LogicOp : GL_COPY)) {
DBG("%s: blit failure\n", __FUNCTION__);
return GL_FALSE;
}
out:
intel_check_front_buffer_rendering(intel);
DBG("%s: success\n", __FUNCTION__);
return GL_TRUE;
}
static void
intel_copy_image_sub_data(struct gl_context *ctx,
struct gl_texture_image *src_image,
struct gl_renderbuffer *src_renderbuffer,
int src_x, int src_y, int src_z,
struct gl_texture_image *dst_image,
struct gl_renderbuffer *dst_renderbuffer,
int dst_x, int dst_y, int dst_z,
int src_width, int src_height)
{
struct brw_context *brw = brw_context(ctx);
struct intel_mipmap_tree *src_mt, *dst_mt;
unsigned src_level, dst_level;
if (src_image) {
src_mt = intel_texture_image(src_image)->mt;
src_level = src_image->Level + src_image->TexObject->MinLevel;
/* Cube maps actually have different images per face */
if (src_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
src_z = src_image->Face;
src_z += src_image->TexObject->MinLayer;
} else {
assert(src_renderbuffer);
src_mt = intel_renderbuffer(src_renderbuffer)->mt;
src_image = src_renderbuffer->TexImage;
src_level = 0;
}
if (dst_image) {
dst_mt = intel_texture_image(dst_image)->mt;
dst_level = dst_image->Level + dst_image->TexObject->MinLevel;
/* Cube maps actually have different images per face */
if (dst_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
dst_z = dst_image->Face;
dst_z += dst_image->TexObject->MinLayer;
} else {
assert(dst_renderbuffer);
dst_mt = intel_renderbuffer(dst_renderbuffer)->mt;
dst_image = dst_renderbuffer->TexImage;
dst_level = 0;
}
copy_miptrees(brw, src_mt, src_x, src_y, src_z, src_level,
dst_mt, dst_x, dst_y, dst_z, dst_level,
src_width, src_height);
/* CopyImage only works for equal formats, texture view equivalence
* classes, and a couple special cases for compressed textures.
*
* Notably, GL_DEPTH_STENCIL does not appear in any equivalence
* classes, so we know the formats must be the same, and thus both
* will either have stencil, or not. They can't be mismatched.
*/
assert((src_mt->stencil_mt != NULL) == (dst_mt->stencil_mt != NULL));
if (dst_mt->stencil_mt) {
copy_miptrees(brw, src_mt->stencil_mt, src_x, src_y, src_z, src_level,
dst_mt->stencil_mt, dst_x, dst_y, dst_z, dst_level,
src_width, src_height);
}
}
/**
* 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 dst_x, dst_y;
(void) fb;
ASSERT(newImage);
intel_image = intel_texture_image(newImage);
if (!intel_image->mt) {
/* Fallback on drawing to a texture that doesn't have a miptree
* (has a border, width/height 0, etc.)
*/
_mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
_mesa_render_texture(ctx, fb, att);
return;
}
else 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 */
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 */
intel_miptree_get_image_offset(intel_image->mt,
att->TextureLevel,
att->CubeMapFace,
att->Zoffset,
&dst_x, &dst_y);
intel_image->mt->region->draw_offset = (dst_y * intel_image->mt->pitch +
dst_x) * intel_image->mt->cpp;
intel_image->mt->region->draw_x = dst_x;
intel_image->mt->region->draw_y = dst_y;
intel_image->used_as_render_target = GL_TRUE;
/* update drawing region, etc */
intel_draw_buffer(ctx, fb);
}
/**
* \brief A fast path for glReadPixels
*
* This fast path is taken when the source format is BGRA, RGBA,
* A or L and when the texture memory is X- or Y-tiled. It downloads
* the source data by directly mapping the memory without a GTT fence.
* This then needs to be de-tiled on the CPU before presenting the data to
* the user in the linear fasion.
*
* This is a performance win over the conventional texture download path.
* In the conventional texture download path, the texture is either mapped
* through the GTT or copied to a linear buffer with the blitter before
* handing off to a software path. This allows us to avoid round-tripping
* through the GPU (in the case where we would be blitting) and do only a
* single copy operation.
*/
static bool
intel_readpixels_tiled_memcpy(struct gl_context * ctx,
GLint xoffset, GLint yoffset,
GLsizei width, GLsizei height,
GLenum format, GLenum type,
GLvoid * pixels,
const struct gl_pixelstore_attrib *pack)
{
struct brw_context *brw = brw_context(ctx);
struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer;
const struct gen_device_info *devinfo = &brw->screen->devinfo;
/* This path supports reading from color buffers only */
if (rb == NULL)
return false;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
int dst_pitch;
/* The miptree's buffer. */
struct brw_bo *bo;
uint32_t cpp;
mem_copy_fn mem_copy = NULL;
/* This fastpath is restricted to specific renderbuffer types:
* a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support
* more types.
*/
if (!devinfo->has_llc ||
!(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) ||
pixels == NULL ||
_mesa_is_bufferobj(pack->BufferObj) ||
pack->Alignment > 4 ||
pack->SkipPixels > 0 ||
pack->SkipRows > 0 ||
(pack->RowLength != 0 && pack->RowLength != width) ||
pack->SwapBytes ||
pack->LsbFirst ||
pack->Invert)
return false;
/* Only a simple blit, no scale, bias or other mapping. */
if (ctx->_ImageTransferState)
return false;
/* It is possible that the renderbuffer (or underlying texture) is
* multisampled. Since ReadPixels from a multisampled buffer requires a
* multisample resolve, we can't handle this here
*/
if (rb->NumSamples > 1)
return false;
/* We can't handle copying from RGBX or BGRX because the tiled_memcpy
* function doesn't set the last channel to 1. Note this checks BaseFormat
* rather than TexFormat in case the RGBX format is being simulated with an
* RGBA format.
*/
if (rb->_BaseFormat == GL_RGB)
return false;
if (!intel_get_memcpy(rb->Format, format, type, &mem_copy, &cpp))
return false;
if (!irb->mt ||
(irb->mt->surf.tiling != ISL_TILING_X &&
irb->mt->surf.tiling != ISL_TILING_Y0)) {
/* The algorithm is written only for X- or Y-tiled memory. */
return false;
}
/* tiled_to_linear() assumes that if the object is swizzled, it is using
* I915_BIT6_SWIZZLE_9_10 for X and I915_BIT6_SWIZZLE_9 for Y. This is only
* true on gen5 and above.
*
* The killer on top is that some gen4 have an L-shaped swizzle mode, where
* parts of the memory aren't swizzled at all. Userspace just can't handle
* that.
*/
if (devinfo->gen < 5 && brw->has_swizzling)
return false;
/* Since we are going to read raw data to the miptree, we need to resolve
* any pending fast color clears before we start.
*/
intel_miptree_access_raw(brw, irb->mt, irb->mt_level, irb->mt_layer, false);
bo = irb->mt->bo;
if (brw_batch_references(&brw->batch, bo)) {
perf_debug("Flushing before mapping a referenced bo.\n");
intel_batchbuffer_flush(brw);
}
void *map = brw_bo_map(brw, bo, MAP_READ | MAP_RAW);
if (map == NULL) {
DBG("%s: failed to map bo\n", __func__);
return false;
}
unsigned slice_offset_x, slice_offset_y;
intel_miptree_get_image_offset(irb->mt, irb->mt_level, irb->mt_layer,
&slice_offset_x, &slice_offset_y);
xoffset += slice_offset_x;
yoffset += slice_offset_y;
dst_pitch = _mesa_image_row_stride(pack, width, format, type);
/* For a window-system renderbuffer, the buffer is actually flipped
* vertically, so we need to handle that. Since the detiling function
* can only really work in the forwards direction, we have to be a
* little creative. First, we compute the Y-offset of the first row of
* the renderbuffer (in renderbuffer coordinates). We then match that
* with the last row of the client's data. Finally, we give
* tiled_to_linear a negative pitch so that it walks through the
* client's data backwards as it walks through the renderbufer forwards.
*/
if (rb->Name == 0) {
yoffset = rb->Height - yoffset - height;
pixels += (ptrdiff_t) (height - 1) * dst_pitch;
dst_pitch = -dst_pitch;
}
/* We postponed printing this message until having committed to executing
* the function.
*/
DBG("%s: x,y=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x "
"mesa_format=0x%x tiling=%d "
"pack=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d)\n",
__func__, xoffset, yoffset, width, height,
format, type, rb->Format, irb->mt->surf.tiling,
pack->Alignment, pack->RowLength, pack->SkipPixels,
pack->SkipRows);
tiled_to_linear(
xoffset * cpp, (xoffset + width) * cpp,
yoffset, yoffset + height,
pixels - (ptrdiff_t) yoffset * dst_pitch - (ptrdiff_t) xoffset * cpp,
map + irb->mt->offset,
dst_pitch, irb->mt->surf.row_pitch,
brw->has_swizzling,
irb->mt->surf.tiling,
mem_copy
);
brw_bo_unmap(bo);
return true;
}
/**
* Do additional "completeness" testing of a framebuffer object.
*/
static void
intel_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb)
{
struct brw_context *brw = brw_context(ctx);
struct intel_renderbuffer *depthRb =
intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencilRb =
intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL, *stencil_mt = NULL;
int i;
DBG("%s() on fb %p (%s)\n", __FUNCTION__,
fb, (fb == ctx->DrawBuffer ? "drawbuffer" :
(fb == ctx->ReadBuffer ? "readbuffer" : "other buffer")));
if (depthRb)
depth_mt = depthRb->mt;
if (stencilRb) {
stencil_mt = stencilRb->mt;
if (stencil_mt->stencil_mt)
stencil_mt = stencil_mt->stencil_mt;
}
if (depth_mt && stencil_mt) {
if (brw->gen >= 7) {
/* For gen >= 7, we are using the lod/minimum-array-element fields
* and supportting layered rendering. This means that we must restrict
* the depth & stencil attachments to match in various more retrictive
* ways. (width, height, depth, LOD and layer)
*/
if (depth_mt->physical_width0 != stencil_mt->physical_width0 ||
depth_mt->physical_height0 != stencil_mt->physical_height0 ||
depth_mt->physical_depth0 != stencil_mt->physical_depth0 ||
depthRb->mt_level != stencilRb->mt_level ||
depthRb->mt_layer != stencilRb->mt_layer) {
fbo_incomplete(fb,
"FBO incomplete: depth and stencil must match in"
"width, height, depth, LOD and layer\n");
}
}
if (depth_mt == stencil_mt) {
/* For true packed depth/stencil (not faked on prefers-separate-stencil
* hardware) we need to be sure they're the same level/layer, since
* we'll be emitting a single packet describing the packed setup.
*/
if (depthRb->mt_level != stencilRb->mt_level ||
depthRb->mt_layer != stencilRb->mt_layer) {
fbo_incomplete(fb,
"FBO incomplete: depth image level/layer %d/%d != "
"stencil image %d/%d\n",
depthRb->mt_level,
depthRb->mt_layer,
stencilRb->mt_level,
stencilRb->mt_layer);
}
} else {
if (!brw->has_separate_stencil) {
fbo_incomplete(fb, "FBO incomplete: separate stencil "
"unsupported\n");
}
if (stencil_mt->format != MESA_FORMAT_S8) {
fbo_incomplete(fb, "FBO incomplete: separate stencil is %s "
"instead of S8\n",
_mesa_get_format_name(stencil_mt->format));
}
if (brw->gen < 7 && !intel_renderbuffer_has_hiz(depthRb)) {
/* Before Gen7, separate depth and stencil buffers can be used
* only if HiZ is enabled. From the Sandybridge PRM, Volume 2,
* Part 1, Bit 3DSTATE_DEPTH_BUFFER.SeparateStencilBufferEnable:
* [DevSNB]: This field must be set to the same value (enabled
* or disabled) as Hierarchical Depth Buffer Enable.
*/
fbo_incomplete(fb, "FBO incomplete: separate stencil "
"without HiZ\n");
}
}
}
for (i = 0; i < Elements(fb->Attachment); i++) {
struct gl_renderbuffer *rb;
struct intel_renderbuffer *irb;
if (fb->Attachment[i].Type == GL_NONE)
continue;
/* A supported attachment will have a Renderbuffer set either
* from being a Renderbuffer or being a texture that got the
* intel_wrap_texture() treatment.
*/
rb = fb->Attachment[i].Renderbuffer;
if (rb == NULL) {
fbo_incomplete(fb, "FBO incomplete: attachment without "
"renderbuffer\n");
continue;
}
if (fb->Attachment[i].Type == GL_TEXTURE) {
if (rb->TexImage->Border) {
fbo_incomplete(fb, "FBO incomplete: texture with border\n");
continue;
}
}
irb = intel_renderbuffer(rb);
if (irb == NULL) {
fbo_incomplete(fb, "FBO incomplete: software rendering "
"renderbuffer\n");
continue;
}
if (!brw_render_target_supported(brw, rb)) {
fbo_incomplete(fb, "FBO incomplete: Unsupported HW "
"texture/renderbuffer format attached: %s\n",
_mesa_get_format_name(intel_rb_format(irb)));
}
}
}
static void
intel_image_target_renderbuffer_storage(struct gl_context *ctx,
struct gl_renderbuffer *rb,
void *image_handle)
{
struct brw_context *brw = brw_context(ctx);
struct intel_renderbuffer *irb;
__DRIscreen *screen;
__DRIimage *image;
screen = brw->intelScreen->driScrnPriv;
image = screen->dri2.image->lookupEGLImage(screen, image_handle,
screen->loaderPrivate);
if (image == NULL)
return;
if (image->planar_format && image->planar_format->nplanes > 1) {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glEGLImageTargetRenderbufferStorage(planar buffers are not "
"supported as render targets.");
return;
}
/* Buffers originating from outside are for read-only. */
if (image->dma_buf_imported) {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glEGLImageTargetRenderbufferStorage(dma buffers are read-only)");
return;
}
/* __DRIimage is opaque to the core so it has to be checked here */
switch (image->format) {
case MESA_FORMAT_RGBA8888_REV:
_mesa_error(ctx, GL_INVALID_OPERATION,
"glEGLImageTargetRenderbufferStorage(unsupported image format");
return;
break;
default:
break;
}
irb = intel_renderbuffer(rb);
intel_miptree_release(&irb->mt);
irb->mt = intel_miptree_create_for_bo(brw,
image->region->bo,
image->format,
image->offset,
image->region->width,
image->region->height,
image->region->pitch,
image->region->tiling);
if (!irb->mt)
return;
rb->InternalFormat = image->internal_format;
rb->Width = image->region->width;
rb->Height = image->region->height;
rb->Format = image->format;
rb->_BaseFormat = _mesa_base_fbo_format(ctx, image->internal_format);
rb->NeedsFinishRenderTexture = true;
}
static GLboolean do_check_fallback(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
GLcontext *ctx = &brw->intel.ctx;
GLuint i;
if (brw->intel.no_rast) {
DBG("FALLBACK: rasterization disabled\n");
return GL_TRUE;
}
/* _NEW_RENDERMODE
*/
if (ctx->RenderMode != GL_RENDER) {
DBG("FALLBACK: render mode\n");
return GL_TRUE;
}
/* _NEW_TEXTURE:
*/
for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
if (texUnit->_ReallyEnabled) {
struct intel_texture_object *intelObj = intel_texture_object(texUnit->_Current);
struct gl_texture_image *texImage = intelObj->base.Image[0][intelObj->firstLevel];
if (texImage->Border) {
DBG("FALLBACK: texture border\n");
return GL_TRUE;
}
}
}
/* _NEW_STENCIL
*/
if (ctx->Stencil._Enabled &&
(ctx->DrawBuffer->Name == 0 && !brw->intel.hw_stencil)) {
DBG("FALLBACK: stencil\n");
return GL_TRUE;
}
/* _NEW_BUFFERS */
if (IS_965(intel->intelScreen->deviceID) &&
!IS_G4X(intel->intelScreen->deviceID)) {
for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
/* The original gen4 hardware couldn't set up WM surfaces pointing
* at an offset within a tile, which can happen when rendering to
* anything but the base level of a texture or the +X face/0 depth.
* This was fixed with the 4 Series hardware.
*
* For these original chips, you would have to make the depth and
* color destination surfaces include information on the texture
* type, LOD, face, and various limits to use them as a destination.
* I would have done this, but there's also a nasty requirement that
* the depth and the color surfaces all be of the same LOD, which
* may be a worse requirement than this alignment. (Also, we may
* want to just demote the texture to untiled, instead).
*/
if (irb->region && irb->region->tiling != I915_TILING_NONE &&
(irb->region->draw_offset & 4095)) {
DBG("FALLBACK: non-tile-aligned destination for tiled FBO\n");
return GL_TRUE;
}
}
}
return GL_FALSE;
}
/**
* Called via glRenderbufferStorageEXT() to set the format and allocate
* storage for a user-created renderbuffer.
*/
GLboolean
intel_alloc_renderbuffer_storage(struct gl_context * 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);
rb->NumSamples = quantize_num_samples(intel, rb->NumSamples);
ASSERT(rb->Name != 0);
switch (internalFormat) {
default:
/* Use the same format-choice logic as for textures.
* Renderbuffers aren't any different from textures for us,
* except they're less useful because you can't texture with
* them.
*/
rb->Format = intel->ctx.Driver.ChooseTextureFormat(ctx, internalFormat,
GL_NONE, GL_NONE);
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:
/* These aren't actual texture formats, so force them here. */
if (intel->has_separate_stencil) {
rb->Format = MESA_FORMAT_S8;
} else {
assert(!intel->must_use_separate_stencil);
rb->Format = MESA_FORMAT_S8_Z24;
}
break;
}
rb->Width = width;
rb->Height = height;
rb->_BaseFormat = _mesa_base_fbo_format(ctx, internalFormat);
intel_miptree_release(&irb->mt);
DBG("%s: %s: %s (%dx%d)\n", __FUNCTION__,
_mesa_lookup_enum_by_nr(internalFormat),
_mesa_get_format_name(rb->Format), width, height);
if (width == 0 || height == 0)
return true;
irb->mt = intel_miptree_create_for_renderbuffer(intel, rb->Format,
width, height,
rb->NumSamples);
if (!irb->mt)
return false;
if (intel->vtbl.is_hiz_depth_format(intel, rb->Format)) {
bool ok = intel_miptree_alloc_hiz(intel, irb->mt, rb->NumSamples);
if (!ok) {
intel_miptree_release(&irb->mt);
return false;
}
}
return true;
}
/**
* Try to do a glBlitFramebuffer using glCopyTexSubImage2D
* We can do this when the dst renderbuffer is actually a texture and
* there is no scaling, mirroring or scissoring.
*
* \return new buffer mask indicating the buffers left to blit using the
* normal path.
*/
static GLbitfield
intel_blit_framebuffer_copy_tex_sub_image(struct gl_context *ctx,
GLint srcX0, GLint srcY0,
GLint srcX1, GLint srcY1,
GLint dstX0, GLint dstY0,
GLint dstX1, GLint dstY1,
GLbitfield mask, GLenum filter)
{
if (mask & GL_COLOR_BUFFER_BIT) {
GLint i;
const struct gl_framebuffer *drawFb = ctx->DrawBuffer;
const struct gl_framebuffer *readFb = ctx->ReadBuffer;
const struct gl_renderbuffer_attachment *drawAtt;
struct intel_renderbuffer *srcRb =
intel_renderbuffer(readFb->_ColorReadBuffer);
/* If the source and destination are the same size with no mirroring,
* the rectangles are within the size of the texture and there is no
* scissor then we can use glCopyTexSubimage2D to implement the blit.
* This will end up as a fast hardware blit on some drivers.
*/
const GLboolean use_intel_copy_texsubimage =
srcX0 - srcX1 == dstX0 - dstX1 &&
srcY0 - srcY1 == dstY0 - dstY1 &&
srcX1 >= srcX0 &&
srcY1 >= srcY0 &&
srcX0 >= 0 && srcX1 <= readFb->Width &&
srcY0 >= 0 && srcY1 <= readFb->Height &&
dstX0 >= 0 && dstX1 <= drawFb->Width &&
dstY0 >= 0 && dstY1 <= drawFb->Height &&
!ctx->Scissor.Enabled;
/* Verify that all the draw buffers can be blitted using
* intel_copy_texsubimage().
*/
for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
int idx = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
if (idx == -1)
continue;
drawAtt = &drawFb->Attachment[idx];
if (srcRb && drawAtt && drawAtt->Texture &&
use_intel_copy_texsubimage)
continue;
else
return mask;
}
/* Blit to all active draw buffers */
for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
int idx = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
if (idx == -1)
continue;
drawAtt = &drawFb->Attachment[idx];
{
const struct gl_texture_object *texObj = drawAtt->Texture;
const GLuint dstLevel = drawAtt->TextureLevel;
const GLenum target = texObj->Target;
struct gl_texture_image *texImage =
_mesa_select_tex_image(ctx, texObj, target, dstLevel);
if (!intel_copy_texsubimage(intel_context(ctx),
intel_texture_image(texImage),
dstX0, dstY0,
srcRb,
srcX0, srcY0,
srcX1 - srcX0, /* width */
srcY1 - srcY0))
return mask;
}
}
mask &= ~GL_COLOR_BUFFER_BIT;
}
return mask;
}
/**
* 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);
int cpp;
GLuint pitch;
ASSERT(rb->Name != 0);
switch (internalFormat) {
case GL_R3_G3_B2:
case GL_RGB4:
case GL_RGB5:
rb->Format = MESA_FORMAT_RGB565;
rb->DataType = GL_UNSIGNED_BYTE;
break;
case GL_RGB:
case GL_RGB8:
case GL_RGB10:
case GL_RGB12:
case GL_RGB16:
rb->Format = MESA_FORMAT_XRGB8888;
rb->DataType = GL_UNSIGNED_BYTE;
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->Format = MESA_FORMAT_ARGB8888;
rb->DataType = GL_UNSIGNED_BYTE;
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->Format = MESA_FORMAT_S8_Z24;
rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
break;
case GL_DEPTH_COMPONENT16:
rb->Format = MESA_FORMAT_Z16;
rb->DataType = GL_UNSIGNED_SHORT;
break;
case GL_DEPTH_COMPONENT:
case GL_DEPTH_COMPONENT24:
case GL_DEPTH_COMPONENT32:
rb->Format = MESA_FORMAT_S8_Z24;
rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
break;
case GL_DEPTH_STENCIL_EXT:
case GL_DEPTH24_STENCIL8_EXT:
rb->Format = MESA_FORMAT_S8_Z24;
rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
break;
default:
_mesa_problem(ctx,
"Unexpected format in intel_alloc_renderbuffer_storage");
return GL_FALSE;
}
rb->_BaseFormat = _mesa_base_fbo_format(ctx, internalFormat);
cpp = _mesa_get_format_bytes(rb->Format);
intelFlush(ctx);
/* free old region */
if (irb->region) {
intel_region_release(&irb->region);
}
/* allocate new memory region/renderbuffer */
/* Choose a pitch to match hardware requirements:
*/
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, I915_TILING_NONE, cpp,
width, height, pitch, GL_TRUE);
if (!irb->region)
return GL_FALSE; /* out of memory? */
ASSERT(irb->region->buffer);
rb->Width = width;
rb->Height = height;
return GL_TRUE;
}
/**
* Do additional "completeness" testing of a framebuffer object.
*/
static void
intel_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb)
{
struct intel_context *intel = intel_context(ctx);
struct intel_renderbuffer *depthRb =
intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencilRb =
intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL, *stencil_mt = NULL;
int i;
DBG("%s() on fb %p (%s)\n", __FUNCTION__,
fb, (fb == ctx->DrawBuffer ? "drawbuffer" :
(fb == ctx->ReadBuffer ? "readbuffer" : "other buffer")));
if (depthRb)
depth_mt = depthRb->mt;
if (stencilRb)
stencil_mt = stencilRb->mt;
if (depth_mt && stencil_mt) {
/* Make sure that the depth and stencil buffers are actually the same
* slice of the same miptree, since we only support packed
* depth/stencil.
*/
if (depth_mt == stencil_mt) {
if (depthRb->mt_level != stencilRb->mt_level ||
depthRb->mt_layer != stencilRb->mt_layer) {
fbo_incomplete(fb,
"FBO incomplete: depth image level/layer %d/%d != "
"stencil image %d/%d\n",
depthRb->mt_level,
depthRb->mt_layer,
stencilRb->mt_level,
stencilRb->mt_layer);
}
} else {
fbo_incomplete(fb, "FBO incomplete: separate stencil unsupported\n");
}
}
for (i = 0; i < Elements(fb->Attachment); i++) {
struct gl_renderbuffer *rb;
struct intel_renderbuffer *irb;
if (fb->Attachment[i].Type == GL_NONE)
continue;
/* A supported attachment will have a Renderbuffer set either
* from being a Renderbuffer or being a texture that got the
* intel_wrap_texture() treatment.
*/
rb = fb->Attachment[i].Renderbuffer;
if (rb == NULL) {
fbo_incomplete(fb, "FBO incomplete: attachment without "
"renderbuffer\n");
continue;
}
if (fb->Attachment[i].Type == GL_TEXTURE) {
if (rb->TexImage->Border) {
fbo_incomplete(fb, "FBO incomplete: texture with border\n");
continue;
}
}
irb = intel_renderbuffer(rb);
if (irb == NULL) {
fbo_incomplete(fb, "FBO incomplete: software rendering "
"renderbuffer\n");
continue;
}
if (!intel->vtbl.render_target_supported(intel, rb)) {
fbo_incomplete(fb, "FBO incomplete: Unsupported HW "
"texture/renderbuffer format attached: %s\n",
_mesa_get_format_name(intel_rb_format(irb)));
}
}
}
static bool
do_blit_drawpixels(struct gl_context * ctx,
GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *unpack,
const GLvoid * pixels)
{
struct brw_context *brw = brw_context(ctx);
struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
GLuint src_offset;
drm_intel_bo *src_buffer;
DBG("%s\n", __FUNCTION__);
if (!intel_check_blit_fragment_ops(ctx, false))
return false;
if (ctx->DrawBuffer->_NumColorDrawBuffers != 1) {
DBG("%s: fallback due to MRT\n", __FUNCTION__);
return false;
}
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
if (!_mesa_format_matches_format_and_type(irb->mt->format, format, type,
false)) {
DBG("%s: bad format for blit\n", __FUNCTION__);
return false;
}
if (unpack->SwapBytes || unpack->LsbFirst ||
unpack->SkipPixels || unpack->SkipRows) {
DBG("%s: bad packing params\n", __FUNCTION__);
return false;
}
int src_stride = _mesa_image_row_stride(unpack, width, format, type);
bool src_flip = false;
/* Mesa flips the src_stride for unpack->Invert, but we want our mt to have
* a normal src_stride.
*/
if (unpack->Invert) {
src_stride = -src_stride;
src_flip = true;
}
src_offset = (GLintptr)pixels;
src_offset += _mesa_image_offset(2, unpack, width, height,
format, type, 0, 0, 0);
intel_prepare_render(brw);
src_buffer = intel_bufferobj_buffer(brw, src,
src_offset, width * height *
irb->mt->cpp);
struct intel_mipmap_tree *pbo_mt =
intel_miptree_create_for_bo(brw,
src_buffer,
irb->mt->format,
src_offset,
width, height,
src_stride, I915_TILING_NONE);
if (!pbo_mt)
return false;
if (!intel_miptree_blit(brw,
pbo_mt, 0, 0,
0, 0, src_flip,
irb->mt, irb->mt_level, irb->mt_layer,
x, y, _mesa_is_winsys_fbo(ctx->DrawBuffer),
width, height, GL_COPY)) {
DBG("%s: blit failed\n", __FUNCTION__);
intel_miptree_release(&pbo_mt);
return false;
}
intel_miptree_release(&pbo_mt);
if (ctx->Query.CurrentOcclusionObject)
ctx->Query.CurrentOcclusionObject->Result += width * height;
intel_check_front_buffer_rendering(brw);
DBG("%s: success\n", __FUNCTION__);
return true;
}
/**
* Sets up a surface state structure to point at the given region.
* While it is only used for the front/back buffer currently, it should be
* usable for further buffers when doing ARB_draw_buffer support.
*/
static void
gen6_update_renderbuffer_surface(struct brw_context *brw,
struct gl_renderbuffer *rb,
bool layered,
unsigned int unit)
{
struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
struct intel_mipmap_tree *mt = irb->mt;
uint32_t *surf;
uint32_t format = 0;
/* _NEW_BUFFERS */
mesa_format rb_format = _mesa_get_render_format(ctx, intel_rb_format(irb));
uint32_t surftype;
int depth = MAX2(irb->layer_count, 1);
const GLenum gl_target =
rb->TexImage ? rb->TexImage->TexObject->Target : GL_TEXTURE_2D;
uint32_t surf_index =
brw->wm.prog_data->binding_table.render_target_start + unit;
intel_miptree_used_for_rendering(irb->mt);
surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE, 6 * 4, 32,
&brw->wm.base.surf_offset[surf_index]);
format = brw->render_target_format[rb_format];
if (unlikely(!brw->format_supported_as_render_target[rb_format])) {
_mesa_problem(ctx, "%s: renderbuffer format %s unsupported\n",
__func__, _mesa_get_format_name(rb_format));
}
switch (gl_target) {
case GL_TEXTURE_CUBE_MAP_ARRAY:
case GL_TEXTURE_CUBE_MAP:
surftype = BRW_SURFACE_2D;
depth *= 6;
break;
case GL_TEXTURE_3D:
depth = MAX2(irb->mt->logical_depth0, 1);
/* fallthrough */
default:
surftype = translate_tex_target(gl_target);
break;
}
const int min_array_element = layered ? 0 : irb->mt_layer;
surf[0] = SET_FIELD(surftype, BRW_SURFACE_TYPE) |
SET_FIELD(format, BRW_SURFACE_FORMAT);
/* reloc */
assert(mt->offset % mt->cpp == 0);
surf[1] = mt->bo->offset64 + mt->offset;
/* In the gen6 PRM Volume 1 Part 1: Graphics Core, Section 7.18.3.7.1
* (Surface Arrays For all surfaces other than separate stencil buffer):
*
* "[DevSNB] Errata: Sampler MSAA Qpitch will be 4 greater than the value
* calculated in the equation above , for every other odd Surface Height
* starting from 1 i.e. 1,5,9,13"
*
* Since this Qpitch errata only impacts the sampler, we have to adjust the
* input for the rendering surface to achieve the same qpitch. For the
* affected heights, we increment the height by 1 for the rendering
* surface.
*/
int height0 = irb->mt->logical_height0;
if (brw->gen == 6 && irb->mt->num_samples > 1 && (height0 % 4) == 1)
height0++;
surf[2] = SET_FIELD(mt->logical_width0 - 1, BRW_SURFACE_WIDTH) |
SET_FIELD(height0 - 1, BRW_SURFACE_HEIGHT) |
SET_FIELD(irb->mt_level - irb->mt->first_level, BRW_SURFACE_LOD);
surf[3] = brw_get_surface_tiling_bits(mt->tiling) |
SET_FIELD(depth - 1, BRW_SURFACE_DEPTH) |
SET_FIELD(mt->pitch - 1, BRW_SURFACE_PITCH);
surf[4] = brw_get_surface_num_multisamples(mt->num_samples) |
SET_FIELD(min_array_element, BRW_SURFACE_MIN_ARRAY_ELEMENT) |
SET_FIELD(depth - 1, BRW_SURFACE_RENDER_TARGET_VIEW_EXTENT);
surf[5] = (mt->align_h == 4 ? BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0);
drm_intel_bo_emit_reloc(brw->batch.bo,
brw->wm.base.surf_offset[surf_index] + 4,
mt->bo,
surf[1] - mt->bo->offset64,
I915_GEM_DOMAIN_RENDER,
I915_GEM_DOMAIN_RENDER);
}
/**
* Try to do a glBlitFramebuffer using glCopyTexSubImage2D
* We can do this when the dst renderbuffer is actually a texture and
* there is no scaling, mirroring or scissoring.
*
* \return new buffer mask indicating the buffers left to blit using the
* normal path.
*/
static GLbitfield
intel_blit_framebuffer_with_blitter(struct gl_context *ctx,
GLint srcX0, GLint srcY0,
GLint srcX1, GLint srcY1,
GLint dstX0, GLint dstY0,
GLint dstX1, GLint dstY1,
GLbitfield mask, GLenum filter)
{
struct brw_context *brw = brw_context(ctx);
/* Sync up the state of window system buffers. We need to do this before
* we go looking for the buffers.
*/
intel_prepare_render(brw);
if (mask & GL_COLOR_BUFFER_BIT) {
GLint i;
const struct gl_framebuffer *drawFb = ctx->DrawBuffer;
const struct gl_framebuffer *readFb = ctx->ReadBuffer;
struct gl_renderbuffer *src_rb = readFb->_ColorReadBuffer;
struct intel_renderbuffer *src_irb = intel_renderbuffer(src_rb);
if (!src_irb) {
perf_debug("glBlitFramebuffer(): missing src renderbuffer. "
"Falling back to software rendering.\n");
return mask;
}
/* If the source and destination are the same size with no mirroring,
* the rectangles are within the size of the texture and there is no
* scissor, then we can probably use the blit engine.
*/
if (!(srcX0 - srcX1 == dstX0 - dstX1 &&
srcY0 - srcY1 == dstY0 - dstY1 &&
srcX1 >= srcX0 &&
srcY1 >= srcY0 &&
srcX0 >= 0 && srcX1 <= readFb->Width &&
srcY0 >= 0 && srcY1 <= readFb->Height &&
dstX0 >= 0 && dstX1 <= drawFb->Width &&
dstY0 >= 0 && dstY1 <= drawFb->Height &&
!ctx->Scissor.Enabled)) {
perf_debug("glBlitFramebuffer(): non-1:1 blit. "
"Falling back to software rendering.\n");
return mask;
}
/* Blit to all active draw buffers. We don't do any pre-checking,
* because we assume that copying to MRTs is rare, and failure midway
* through copying is even more rare. Even if it was to occur, it's
* safe to let meta start the copy over from scratch, because
* glBlitFramebuffer completely overwrites the destination pixels, and
* results are undefined if any destination pixels have a dependency on
* source pixels.
*/
for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
struct gl_renderbuffer *dst_rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
struct intel_renderbuffer *dst_irb = intel_renderbuffer(dst_rb);
if (!dst_irb) {
perf_debug("glBlitFramebuffer(): missing dst renderbuffer. "
"Falling back to software rendering.\n");
return mask;
}
gl_format src_format = _mesa_get_srgb_format_linear(src_rb->Format);
gl_format dst_format = _mesa_get_srgb_format_linear(dst_rb->Format);
if (src_format != dst_format) {
perf_debug("glBlitFramebuffer(): unsupported blit from %s to %s. "
"Falling back to software rendering.\n",
_mesa_get_format_name(src_format),
_mesa_get_format_name(dst_format));
return mask;
}
if (!intel_miptree_blit(brw,
src_irb->mt,
src_irb->mt_level, src_irb->mt_layer,
srcX0, srcY0, src_rb->Name == 0,
dst_irb->mt,
dst_irb->mt_level, dst_irb->mt_layer,
dstX0, dstY0, dst_rb->Name == 0,
dstX1 - dstX0, dstY1 - dstY0, GL_COPY)) {
perf_debug("glBlitFramebuffer(): unknown blit failure. "
"Falling back to software rendering.\n");
return mask;
}
}
mask &= ~GL_COLOR_BUFFER_BIT;
}
return mask;
}
static bool
do_single_blorp_clear(struct brw_context *brw, struct gl_framebuffer *fb,
struct gl_renderbuffer *rb, unsigned buf,
bool partial_clear, bool encode_srgb, unsigned layer)
{
struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
mesa_format format = irb->mt->format;
struct brw_blorp_params params;
brw_blorp_params_init(¶ms);
if (!encode_srgb && _mesa_get_format_color_encoding(format) == GL_SRGB)
format = _mesa_get_srgb_format_linear(format);
brw_blorp_surface_info_init(brw, ¶ms.dst, irb->mt, irb->mt_level,
layer, format, true);
/* Override the surface format according to the context's sRGB rules. */
params.dst.brw_surfaceformat = brw->render_target_format[format];
params.x0 = fb->_Xmin;
params.x1 = fb->_Xmax;
if (rb->Name != 0) {
params.y0 = fb->_Ymin;
params.y1 = fb->_Ymax;
} else {
params.y0 = rb->Height - fb->_Ymax;
params.y1 = rb->Height - fb->_Ymin;
}
memcpy(¶ms.wm_inputs, ctx->Color.ClearColor.f, sizeof(float) * 4);
bool use_simd16_replicated_data = true;
/* From the SNB PRM (Vol4_Part1):
*
* "Replicated data (Message Type = 111) is only supported when
* accessing tiled memory. Using this Message Type to access linear
* (untiled) memory is UNDEFINED."
*/
if (irb->mt->tiling == I915_TILING_NONE)
use_simd16_replicated_data = false;
/* Constant color writes ignore everyting in blend and color calculator
* state. This is not documented.
*/
if (set_write_disables(irb, ctx->Color.ColorMask[buf],
params.color_write_disable))
use_simd16_replicated_data = false;
if (irb->mt->fast_clear_state != INTEL_FAST_CLEAR_STATE_NO_MCS &&
!partial_clear && use_simd16_replicated_data &&
brw_is_color_fast_clear_compatible(brw, irb->mt,
&ctx->Color.ClearColor)) {
memset(¶ms.wm_inputs, 0xff, 4*sizeof(float));
params.fast_clear_op = GEN7_PS_RENDER_TARGET_FAST_CLEAR_ENABLE;
brw_get_fast_clear_rect(brw, fb, irb->mt, ¶ms.x0, ¶ms.y0,
¶ms.x1, ¶ms.y1);
} else {
brw_meta_get_buffer_rect(fb, ¶ms.x0, ¶ms.y0,
¶ms.x1, ¶ms.y1);
}
brw_blorp_params_get_clear_kernel(brw, ¶ms, use_simd16_replicated_data);
const bool is_fast_clear =
params.fast_clear_op == GEN7_PS_RENDER_TARGET_FAST_CLEAR_ENABLE;
if (is_fast_clear) {
/* Record the clear color in the miptree so that it will be
* programmed in SURFACE_STATE by later rendering and resolve
* operations.
*/
const bool color_updated = brw_meta_set_fast_clear_color(
brw, irb->mt, &ctx->Color.ClearColor);
/* If the buffer is already in INTEL_FAST_CLEAR_STATE_CLEAR, the clear
* is redundant and can be skipped.
*/
if (!color_updated &&
irb->mt->fast_clear_state == INTEL_FAST_CLEAR_STATE_CLEAR)
return true;
/* If the MCS buffer hasn't been allocated yet, we need to allocate
* it now.
*/
if (!irb->mt->mcs_mt) {
if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt)) {
/* MCS allocation failed--probably this will only happen in
* out-of-memory conditions. But in any case, try to recover
* by falling back to a non-blorp clear technique.
*/
return false;
}
}
}
const char *clear_type;
if (is_fast_clear)
clear_type = "fast";
else if (use_simd16_replicated_data)
clear_type = "replicated";
else
clear_type = "slow";
DBG("%s (%s) to mt %p level %d layer %d\n", __FUNCTION__, clear_type,
irb->mt, irb->mt_level, irb->mt_layer);
brw_blorp_exec(brw, ¶ms);
if (is_fast_clear) {
/* Now that the fast clear has occurred, put the buffer in
* INTEL_FAST_CLEAR_STATE_CLEAR so that we won't waste time doing
* redundant clears.
*/
irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_CLEAR;
} else if (intel_miptree_is_lossless_compressed(brw, irb->mt)) {
/* Compressed buffers can be cleared also using normal rep-clear. In
* such case they bahave such as if they were drawn using normal 3D
* render pipeline, and we simply mark the mcs as dirty.
*/
assert(partial_clear);
irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_UNRESOLVED;
}
return true;
}
/*
* Render a bitmap.
*/
static bool
do_blit_bitmap( struct gl_context *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 gl_framebuffer *fb = ctx->DrawBuffer;
struct intel_renderbuffer *irb;
GLfloat tmpColor[4];
GLubyte ubcolor[4];
GLuint color;
GLsizei bitmap_width = width;
GLsizei bitmap_height = height;
GLint px, py;
GLuint stipple[32];
GLint orig_dstx = dstx;
GLint orig_dsty = dsty;
/* Update draw buffer bounds */
_mesa_update_state(ctx);
if (ctx->Depth.Test) {
/* The blit path produces incorrect results when depth testing is on.
* It seems the blit Z coord is always 1.0 (the far plane) so fragments
* will likely be obscured by other, closer geometry.
*/
return false;
}
intel_prepare_render(intel);
if (fb->_NumColorDrawBuffers != 1) {
perf_debug("accelerated glBitmap() only supports rendering to a "
"single color buffer\n");
return false;
}
irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]);
if (_mesa_is_bufferobj(unpack->BufferObj)) {
bitmap = map_pbo(ctx, width, height, unpack, bitmap);
if (bitmap == NULL)
return true; /* even though this is an error, we're done */
}
COPY_4V(tmpColor, ctx->Current.RasterColor);
if (_mesa_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]);
switch (irb->mt->format) {
case MESA_FORMAT_B8G8R8A8_UNORM:
case MESA_FORMAT_B8G8R8X8_UNORM:
color = PACK_COLOR_8888(ubcolor[3], ubcolor[0], ubcolor[1], ubcolor[2]);
break;
case MESA_FORMAT_B5G6R5_UNORM:
color = PACK_COLOR_565(ubcolor[0], ubcolor[1], ubcolor[2]);
break;
default:
perf_debug("Unsupported format %s in accelerated glBitmap()\n",
_mesa_get_format_name(irb->mt->format));
return false;
}
if (!intel_check_blit_fragment_ops(ctx, tmpColor[3] == 1.0F))
return false;
/* Clip to buffer bounds and scissor. */
if (!_mesa_clip_to_region(fb->_Xmin, fb->_Ymin,
fb->_Xmax, fb->_Ymax,
&dstx, &dsty, &width, &height))
goto out;
dsty = y_flip(fb, dsty, height);
#define DY 32
#define DX 32
/* Chop it all into chunks that can be digested by hardware: */
for (py = 0; py < height; py += DY) {
for (px = 0; px < width; px += DX) {
int h = MIN2(DY, height - py);
int w = MIN2(DX, width - 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:
*
* Have to translate destination coordinates back into source
* coordinates.
*/
int count = get_bitmap_rect(bitmap_width, bitmap_height, unpack,
bitmap,
-orig_dstx + (dstx + px),
-orig_dsty + y_flip(fb, dsty + py, h),
w, h,
(GLubyte *)stipple,
8,
_mesa_is_winsys_fbo(fb));
if (count == 0)
continue;
if (!intelEmitImmediateColorExpandBlit(intel,
irb->mt->cpp,
(GLubyte *)stipple,
sz,
color,
irb->mt->region->pitch,
irb->mt->region->bo,
0,
irb->mt->region->tiling,
dstx + px,
dsty + py,
w, h,
logic_op)) {
return false;
}
if (ctx->Query.CurrentOcclusionObject)
ctx->Query.CurrentOcclusionObject->Result += count;
}
}
out:
if (unlikely(INTEL_DEBUG & DEBUG_SYNC))
intel_batchbuffer_flush(intel);
if (_mesa_is_bufferobj(unpack->BufferObj)) {
/* done with PBO so unmap it now */
ctx->Driver.UnmapBuffer(ctx, unpack->BufferObj, MAP_INTERNAL);
}
intel_check_front_buffer_rendering(intel);
return true;
}
brw_blorp_clear_params::brw_blorp_clear_params(struct brw_context *brw,
struct gl_framebuffer *fb,
struct gl_renderbuffer *rb,
GLubyte *color_mask,
bool partial_clear,
unsigned layer)
{
struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
dst.set(brw, irb->mt, irb->mt_level, layer, true);
/* Override the surface format according to the context's sRGB rules. */
mesa_format format = _mesa_get_render_format(ctx, irb->mt->format);
dst.brw_surfaceformat = brw->render_target_format[format];
x0 = fb->_Xmin;
x1 = fb->_Xmax;
if (rb->Name != 0) {
y0 = fb->_Ymin;
y1 = fb->_Ymax;
} else {
y0 = rb->Height - fb->_Ymax;
y1 = rb->Height - fb->_Ymin;
}
float *push_consts = (float *)&wm_push_consts;
push_consts[0] = ctx->Color.ClearColor.f[0];
push_consts[1] = ctx->Color.ClearColor.f[1];
push_consts[2] = ctx->Color.ClearColor.f[2];
push_consts[3] = ctx->Color.ClearColor.f[3];
use_wm_prog = true;
memset(&wm_prog_key, 0, sizeof(wm_prog_key));
wm_prog_key.use_simd16_replicated_data = true;
/* From the SNB PRM (Vol4_Part1):
*
* "Replicated data (Message Type = 111) is only supported when
* accessing tiled memory. Using this Message Type to access linear
* (untiled) memory is UNDEFINED."
*/
if (irb->mt->tiling == I915_TILING_NONE)
wm_prog_key.use_simd16_replicated_data = false;
/* Constant color writes ignore everyting in blend and color calculator
* state. This is not documented.
*/
for (int i = 0; i < 4; i++) {
if (_mesa_format_has_color_component(irb->mt->format, i) &&
!color_mask[i]) {
color_write_disable[i] = true;
wm_prog_key.use_simd16_replicated_data = false;
}
}
/* If we can do this as a fast color clear, do so.
*
* Note that the condition "!partial_clear" means we only try to do full
* buffer clears using fast color clear logic. This is necessary because
* the fast color clear alignment requirements mean that we typically have
* to clear a larger rectangle than (x0, y0) to (x1, y1). Restricting fast
* color clears to the full-buffer condition guarantees that the extra
* memory locations that get written to are outside the image boundary (and
* hence irrelevant). Note that the rectangle alignment requirements are
* never larger than the size of a tile, so there is no danger of
* overflowing beyond the memory belonging to the region.
*/
if (irb->mt->fast_clear_state != INTEL_FAST_CLEAR_STATE_NO_MCS &&
!partial_clear && wm_prog_key.use_simd16_replicated_data &&
is_color_fast_clear_compatible(brw, format, &ctx->Color.ClearColor)) {
memset(push_consts, 0xff, 4*sizeof(float));
fast_clear_op = GEN7_FAST_CLEAR_OP_FAST_CLEAR;
/* Figure out what the clear rectangle needs to be aligned to, and how
* much it needs to be scaled down.
*/
unsigned x_align, y_align, x_scaledown, y_scaledown;
if (irb->mt->msaa_layout == INTEL_MSAA_LAYOUT_NONE) {
/* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
* Target(s)", beneath the "Fast Color Clear" bullet (p327):
*
* Clear pass must have a clear rectangle that must follow
* alignment rules in terms of pixels and lines as shown in the
* table below. Further, the clear-rectangle height and width
* must be multiple of the following dimensions. If the height
* and width of the render target being cleared do not meet these
* requirements, an MCS buffer can be created such that it
* follows the requirement and covers the RT.
*
* The alignment size in the table that follows is related to the
* alignment size returned by intel_get_non_msrt_mcs_alignment(), but
* with X alignment multiplied by 16 and Y alignment multiplied by 32.
*/
intel_get_non_msrt_mcs_alignment(brw, irb->mt, &x_align, &y_align);
x_align *= 16;
y_align *= 32;
/* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
* Target(s)", beneath the "Fast Color Clear" bullet (p327):
*
* In order to optimize the performance MCS buffer (when bound to
* 1X RT) clear similarly to MCS buffer clear for MSRT case,
* clear rect is required to be scaled by the following factors
* in the horizontal and vertical directions:
*
* The X and Y scale down factors in the table that follows are each
* equal to half the alignment value computed above.
*/
x_scaledown = x_align / 2;
y_scaledown = y_align / 2;
/* From BSpec: 3D-Media-GPGPU Engine > 3D Pipeline > Pixel > Pixel
* Backend > MCS Buffer for Render Target(s) [DevIVB+] > Table "Color
* Clear of Non-MultiSampled Render Target Restrictions":
*
* Clear rectangle must be aligned to two times the number of
* pixels in the table shown below due to 16x16 hashing across the
* slice.
*/
x_align *= 2;
y_align *= 2;
} else {
/* From the Ivy Bridge PRM, Vol2 Part1 11.7 "MCS Buffer for Render
* Target(s)", beneath the "MSAA Compression" bullet (p326):
*
* Clear pass for this case requires that scaled down primitive
* is sent down with upper left co-ordinate to coincide with
* actual rectangle being cleared. For MSAA, clear rectangle’s
* height and width need to as show in the following table in
* terms of (width,height) of the RT.
*
* MSAA Width of Clear Rect Height of Clear Rect
* 4X Ceil(1/8*width) Ceil(1/2*height)
* 8X Ceil(1/2*width) Ceil(1/2*height)
*
* The text "with upper left co-ordinate to coincide with actual
* rectangle being cleared" is a little confusing--it seems to imply
* that to clear a rectangle from (x,y) to (x+w,y+h), one needs to
* feed the pipeline using the rectangle (x,y) to
* (x+Ceil(w/N),y+Ceil(h/2)), where N is either 2 or 8 depending on
* the number of samples. Experiments indicate that this is not
* quite correct; actually, what the hardware appears to do is to
* align whatever rectangle is sent down the pipeline to the nearest
* multiple of 2x2 blocks, and then scale it up by a factor of N
* horizontally and 2 vertically. So the resulting alignment is 4
* vertically and either 4 or 16 horizontally, and the scaledown
* factor is 2 vertically and either 2 or 8 horizontally.
*/
switch (irb->mt->num_samples) {
case 4:
x_scaledown = 8;
break;
case 8:
x_scaledown = 2;
break;
default:
assert(!"Unexpected sample count for fast clear");
break;
}
y_scaledown = 2;
x_align = x_scaledown * 2;
y_align = y_scaledown * 2;
}
/* Do the alignment and scaledown. */
x0 = ROUND_DOWN_TO(x0, x_align) / x_scaledown;
y0 = ROUND_DOWN_TO(y0, y_align) / y_scaledown;
x1 = ALIGN(x1, x_align) / x_scaledown;
y1 = ALIGN(y1, y_align) / y_scaledown;
}
}
/**
* 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(struct gl_context * ctx,
struct gl_framebuffer *fb,
struct gl_renderbuffer_attachment *att)
{
struct intel_context *intel = intel_context(ctx);
struct gl_texture_image *image = _mesa_get_attachment_teximage(att);
struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
struct intel_texture_image *intel_image = intel_texture_image(image);
struct intel_mipmap_tree *mt = intel_image->mt;
int layer;
(void) fb;
if (att->CubeMapFace > 0) {
assert(att->Zoffset == 0);
layer = att->CubeMapFace;
} else {
layer = att->Zoffset;
}
if (!intel_image->mt) {
/* Fallback on drawing to a texture that doesn't have a miptree
* (has a border, width/height 0, etc.)
*/
_mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
_swrast_render_texture(ctx, fb, att);
return;
}
else if (!irb) {
intel_miptree_check_level_layer(mt, att->TextureLevel, layer);
irb = (struct intel_renderbuffer *)intel_new_renderbuffer(ctx, ~0);
if (irb) {
/* bind the wrapper to the attachment point */
_mesa_reference_renderbuffer(&att->Renderbuffer, &irb->Base.Base);
}
else {
/* fallback to software rendering */
_swrast_render_texture(ctx, fb, att);
return;
}
}
if (!intel_renderbuffer_update_wrapper(intel, irb, image, layer)) {
_mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
_swrast_render_texture(ctx, fb, att);
return;
}
irb->tex_image = image;
DBG("Begin render %s texture tex=%u w=%d h=%d refcount=%d\n",
_mesa_get_format_name(image->TexFormat),
att->Texture->Name, image->Width, image->Height,
irb->Base.Base.RefCount);
/* update drawing region, etc */
intel_draw_buffer(ctx);
}
bool
do_single_blorp_clear(struct brw_context *brw, struct gl_framebuffer *fb,
struct gl_renderbuffer *rb, unsigned buf,
bool partial_clear, unsigned layer)
{
struct gl_context *ctx = &brw->ctx;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
brw_blorp_clear_params params(brw, fb, rb, ctx->Color.ColorMask[buf],
partial_clear, layer);
bool is_fast_clear =
(params.fast_clear_op == GEN7_FAST_CLEAR_OP_FAST_CLEAR);
if (is_fast_clear) {
/* Record the clear color in the miptree so that it will be
* programmed in SURFACE_STATE by later rendering and resolve
* operations.
*/
uint32_t new_color_value =
compute_fast_clear_color_bits(&ctx->Color.ClearColor);
if (irb->mt->fast_clear_color_value != new_color_value) {
irb->mt->fast_clear_color_value = new_color_value;
brw->state.dirty.brw |= BRW_NEW_SURFACES;
}
/* If the buffer is already in INTEL_FAST_CLEAR_STATE_CLEAR, the clear
* is redundant and can be skipped.
*/
if (irb->mt->fast_clear_state == INTEL_FAST_CLEAR_STATE_CLEAR)
return true;
/* If the MCS buffer hasn't been allocated yet, we need to allocate
* it now.
*/
if (!irb->mt->mcs_mt) {
if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt)) {
/* MCS allocation failed--probably this will only happen in
* out-of-memory conditions. But in any case, try to recover
* by falling back to a non-blorp clear technique.
*/
return false;
}
brw->state.dirty.brw |= BRW_NEW_SURFACES;
}
}
const char *clear_type;
if (is_fast_clear)
clear_type = "fast";
else if (params.wm_prog_key.use_simd16_replicated_data)
clear_type = "replicated";
else
clear_type = "slow";
DBG("%s (%s) to mt %p level %d layer %d\n", __FUNCTION__, clear_type,
irb->mt, irb->mt_level, irb->mt_layer);
brw_blorp_exec(brw, ¶ms);
if (is_fast_clear) {
/* Now that the fast clear has occurred, put the buffer in
* INTEL_FAST_CLEAR_STATE_CLEAR so that we won't waste time doing
* redundant clears.
*/
irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_CLEAR;
}
return true;
}
/**
* Use blitting to clear the renderbuffers named by 'flags'.
* Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferMask 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, GLboolean all,
GLint cx, GLint cy, GLint cw, GLint ch)
{
struct intel_context *intel = intel_context( ctx );
GLuint clear_depth;
GLbitfield skipBuffers = 0;
BATCH_LOCALS;
if (INTEL_DEBUG & DEBUG_DRI)
_mesa_printf("%s %x\n", __FUNCTION__, mask);
/*
* Compute values for clearing the buffers.
*/
clear_depth = 0;
if (mask & BUFFER_BIT_DEPTH) {
clear_depth = (GLuint) (ctx->DrawBuffer->_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)
{
drm_clip_rect_t clear;
int i;
if (intel->ctx.DrawBuffer->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;
/* adjust for page flipping */
if ( intel->sarea->pf_current_page == 1 ) {
const GLuint tmp = mask;
mask &= ~(BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_BACK_LEFT);
if ( tmp & BUFFER_BIT_FRONT_LEFT ) mask |= BUFFER_BIT_BACK_LEFT;
if ( tmp & BUFFER_BIT_BACK_LEFT ) mask |= BUFFER_BIT_FRONT_LEFT;
}
}
else {
/* clearing FBO */
ASSERT(intel->numClipRects == 1);
ASSERT(intel->pClipRects == &intel->fboRect);
clear.x1 = cx;
clear.y1 = intel->ctx.DrawBuffer->Height - cy - ch;
clear.x2 = clear.y1 + 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 */
if (!all) {
intel_intersect_cliprects(&b, &clear, box);
} else {
b = *box;
}
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 */
const struct intel_renderbuffer *irb
= intel_renderbuffer(ctx->DrawBuffer->
Attachment[buf].Renderbuffer);
GLuint clearVal;
GLint pitch, cpp;
GLuint BR13, CMD;
ASSERT(irb);
ASSERT(irb->region);
pitch = irb->region->pitch;
cpp = irb->region->cpp;
/* Setup the blit command */
if (cpp == 4) {
BR13 = (0xF0 << 16) | (pitch * cpp) | (1<<24) | (1<<25);
if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
CMD = XY_COLOR_BLT_CMD;
if (clearMask & BUFFER_BIT_DEPTH)
CMD |= XY_COLOR_BLT_WRITE_RGB;
if (clearMask & BUFFER_BIT_STENCIL)
CMD |= XY_COLOR_BLT_WRITE_ALPHA;
}
else {
/* clearing RGBA */
CMD = (XY_COLOR_BLT_CMD |
XY_COLOR_BLT_WRITE_ALPHA |
XY_COLOR_BLT_WRITE_RGB);
}
}
else {
ASSERT(cpp == 2 || cpp == 0);
BR13 = (0xF0 << 16) | (pitch * cpp) | (1<<24);
CMD = XY_COLOR_BLT_CMD;
}
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);
*/
BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
OUT_BATCH( CMD );
OUT_BATCH( BR13 );
OUT_BATCH( (b.y1 << 16) | b.x1 );
OUT_BATCH( (b.y2 << 16) | b.x2 );
OUT_RELOC( irb->region->buffer, DRM_MM_TT|DRM_MM_WRITE,
irb->region->draw_offset );
OUT_BATCH( clearVal );
ADVANCE_BATCH();
clearMask &= ~bufBit; /* turn off bit, for faster loop exit */
}
}
}
intel_batchbuffer_flush( intel->batch );
}
UNLOCK_HARDWARE( intel );
}
/**
* CopyPixels with the blitter. Don't support zooming, pixel transfer, etc.
*/
static bool
do_blit_copypixels(struct gl_context * ctx,
GLint srcx, GLint srcy,
GLsizei width, GLsizei height,
GLint dstx, GLint dsty, GLenum type)
{
struct brw_context *brw = brw_context(ctx);
struct gl_framebuffer *fb = ctx->DrawBuffer;
struct gl_framebuffer *read_fb = ctx->ReadBuffer;
GLint orig_dstx;
GLint orig_dsty;
GLint orig_srcx;
GLint orig_srcy;
struct intel_renderbuffer *draw_irb = NULL;
struct intel_renderbuffer *read_irb = NULL;
/* Update draw buffer bounds */
_mesa_update_state(ctx);
intel_prepare_render(brw);
switch (type) {
case GL_COLOR:
if (fb->_NumColorDrawBuffers != 1) {
perf_debug("glCopyPixels() fallback: MRT\n");
return false;
}
draw_irb = intel_renderbuffer(fb->_ColorDrawBuffers[0]);
read_irb = intel_renderbuffer(read_fb->_ColorReadBuffer);
break;
case GL_DEPTH_STENCIL_EXT:
draw_irb = intel_renderbuffer(fb->Attachment[BUFFER_DEPTH].Renderbuffer);
read_irb =
intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
break;
case GL_DEPTH:
perf_debug("glCopyPixels() fallback: GL_DEPTH\n");
return false;
case GL_STENCIL:
perf_debug("glCopyPixels() fallback: GL_STENCIL\n");
return false;
default:
perf_debug("glCopyPixels(): Unknown type\n");
return false;
}
if (!draw_irb) {
perf_debug("glCopyPixels() fallback: missing draw buffer\n");
return false;
}
if (!read_irb) {
perf_debug("glCopyPixels() fallback: missing read buffer\n");
return false;
}
if (draw_irb->mt->num_samples > 1 || read_irb->mt->num_samples > 1) {
perf_debug("glCopyPixels() fallback: multisampled buffers\n");
return false;
}
if (ctx->_ImageTransferState) {
perf_debug("glCopyPixels(): Unsupported image transfer state\n");
return false;
}
if (ctx->Depth.Test) {
perf_debug("glCopyPixels(): Unsupported depth test state\n");
return false;
}
if (ctx->Stencil._Enabled) {
perf_debug("glCopyPixels(): Unsupported stencil test state\n");
return false;
}
if (ctx->Fog.Enabled ||
ctx->Texture._MaxEnabledTexImageUnit != -1 ||
ctx->FragmentProgram._Enabled) {
perf_debug("glCopyPixels(): Unsupported fragment shader state\n");
return false;
}
if (ctx->Color.AlphaEnabled ||
ctx->Color.BlendEnabled) {
perf_debug("glCopyPixels(): Unsupported blend state\n");
return false;
}
if (!ctx->Color.ColorMask[0][0] ||
!ctx->Color.ColorMask[0][1] ||
!ctx->Color.ColorMask[0][2] ||
!ctx->Color.ColorMask[0][3]) {
perf_debug("glCopyPixels(): Unsupported color mask state\n");
return false;
}
if (ctx->Pixel.ZoomX != 1.0F || ctx->Pixel.ZoomY != 1.0F) {
perf_debug("glCopyPixles(): Unsupported pixel zoom\n");
return false;
}
intel_batchbuffer_flush(brw);
/* Clip to destination buffer. */
orig_dstx = dstx;
orig_dsty = dsty;
if (!_mesa_clip_to_region(fb->_Xmin, fb->_Ymin,
fb->_Xmax, fb->_Ymax,
&dstx, &dsty, &width, &height))
goto out;
/* Adjust src coords for our post-clipped destination origin */
srcx += dstx - orig_dstx;
srcy += dsty - orig_dsty;
/* Clip to source buffer. */
orig_srcx = srcx;
orig_srcy = srcy;
if (!_mesa_clip_to_region(0, 0,
read_fb->Width, read_fb->Height,
&srcx, &srcy, &width, &height))
goto out;
/* Adjust dst coords for our post-clipped source origin */
dstx += srcx - orig_srcx;
dsty += srcy - orig_srcy;
if (!intel_miptree_blit(brw,
read_irb->mt, read_irb->mt_level, read_irb->mt_layer,
srcx, srcy, _mesa_is_winsys_fbo(read_fb),
draw_irb->mt, draw_irb->mt_level, draw_irb->mt_layer,
dstx, dsty, _mesa_is_winsys_fbo(fb),
width, height,
(ctx->Color.ColorLogicOpEnabled ?
ctx->Color.LogicOp : GL_COPY))) {
DBG("%s: blit failure\n", __FUNCTION__);
return false;
}
if (ctx->Query.CurrentOcclusionObject)
ctx->Query.CurrentOcclusionObject->Result += width * height;
out:
DBG("%s: success\n", __FUNCTION__);
return true;
}
bool
brw_meta_fast_clear(struct brw_context *brw, struct gl_framebuffer *fb,
GLbitfield buffers, bool partial_clear)
{
struct gl_context *ctx = &brw->ctx;
mesa_format format;
enum { FAST_CLEAR, REP_CLEAR, PLAIN_CLEAR } clear_type;
GLbitfield plain_clear_buffers, meta_save, rep_clear_buffers, fast_clear_buffers;
struct rect fast_clear_rect, clear_rect;
int layers;
fast_clear_buffers = rep_clear_buffers = plain_clear_buffers = 0;
/* First we loop through the color draw buffers and determine which ones
* can be fast cleared, which ones can use the replicated write and which
* ones have to fall back to regular color clear.
*/
for (unsigned buf = 0; buf < fb->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = fb->_ColorDrawBuffers[buf];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
int index = fb->_ColorDrawBufferIndexes[buf];
/* Only clear the buffers present in the provided mask */
if (((1 << index) & buffers) == 0)
continue;
/* If this is an ES2 context or GL_ARB_ES2_compatibility is supported,
* the framebuffer can be complete with some attachments missing. In
* this case the _ColorDrawBuffers pointer will be NULL.
*/
if (rb == NULL)
continue;
clear_type = FAST_CLEAR;
/* We don't have fast clear until gen7. */
if (brw->gen < 7)
clear_type = REP_CLEAR;
if (irb->mt->fast_clear_state == INTEL_FAST_CLEAR_STATE_NO_MCS)
clear_type = REP_CLEAR;
/* We can't do scissored fast clears because of the restrictions on the
* fast clear rectangle size.
*/
if (partial_clear)
clear_type = REP_CLEAR;
/* Fast clear is only supported for colors where all components are
* either 0 or 1.
*/
format = _mesa_get_render_format(ctx, irb->mt->format);
if (!is_color_fast_clear_compatible(brw, format, &ctx->Color.ClearColor))
clear_type = REP_CLEAR;
/* From the SNB PRM (Vol4_Part1):
*
* "Replicated data (Message Type = 111) is only supported when
* accessing tiled memory. Using this Message Type to access
* linear (untiled) memory is UNDEFINED."
*/
if (irb->mt->tiling == I915_TILING_NONE) {
perf_debug("falling back to plain clear because buffers are untiled\n");
clear_type = PLAIN_CLEAR;
}
/* Constant color writes ignore everything in blend and color calculator
* state. This is not documented.
*/
GLubyte *color_mask = ctx->Color.ColorMask[buf];
for (int i = 0; i < 4; i++) {
if (_mesa_format_has_color_component(irb->mt->format, i) &&
!color_mask[i]) {
perf_debug("falling back to plain clear because of color mask\n");
clear_type = PLAIN_CLEAR;
}
}
/* Allocate the MCS for non MSRT surfaces now if we're doing a fast
* clear and we don't have the MCS yet. On failure, fall back to
* replicated clear.
*/
if (clear_type == FAST_CLEAR && irb->mt->mcs_mt == NULL)
if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt))
clear_type = REP_CLEAR;
switch (clear_type) {
case FAST_CLEAR:
irb->mt->fast_clear_color_value =
compute_fast_clear_color_bits(&ctx->Color.ClearColor);
irb->need_downsample = true;
/* If the buffer is already in INTEL_FAST_CLEAR_STATE_CLEAR, the
* clear is redundant and can be skipped. Only skip after we've
* updated the fast clear color above though.
*/
if (irb->mt->fast_clear_state == INTEL_FAST_CLEAR_STATE_CLEAR)
continue;
/* Set fast_clear_state to RESOLVED so we don't try resolve them when
* we draw, in case the mt is also bound as a texture.
*/
irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_RESOLVED;
irb->need_downsample = true;
fast_clear_buffers |= 1 << index;
get_fast_clear_rect(brw, fb, irb, &fast_clear_rect);
break;
case REP_CLEAR:
rep_clear_buffers |= 1 << index;
get_buffer_rect(brw, fb, irb, &clear_rect);
break;
case PLAIN_CLEAR:
plain_clear_buffers |= 1 << index;
get_buffer_rect(brw, fb, irb, &clear_rect);
continue;
}
}
if (!(fast_clear_buffers | rep_clear_buffers)) {
if (plain_clear_buffers)
/* If we only have plain clears, skip the meta save/restore. */
goto out;
else
/* Nothing left to do. This happens when we hit the redundant fast
* clear case above and nothing else.
*/
return true;
}
meta_save =
MESA_META_ALPHA_TEST |
MESA_META_BLEND |
MESA_META_DEPTH_TEST |
MESA_META_RASTERIZATION |
MESA_META_SHADER |
MESA_META_STENCIL_TEST |
MESA_META_VERTEX |
MESA_META_VIEWPORT |
MESA_META_CLIP |
MESA_META_CLAMP_FRAGMENT_COLOR |
MESA_META_MULTISAMPLE |
MESA_META_OCCLUSION_QUERY |
MESA_META_DRAW_BUFFERS;
_mesa_meta_begin(ctx, meta_save);
if (!brw_fast_clear_init(brw)) {
/* This is going to be hard to recover from, most likely out of memory.
* Bail and let meta try and (probably) fail for us.
*/
plain_clear_buffers = buffers;
goto bail_to_meta;
}
/* Clears never have the color clamped. */
if (ctx->Extensions.ARB_color_buffer_float)
_mesa_ClampColor(GL_CLAMP_FRAGMENT_COLOR, GL_FALSE);
_mesa_set_enable(ctx, GL_DEPTH_TEST, GL_FALSE);
_mesa_DepthMask(GL_FALSE);
_mesa_set_enable(ctx, GL_STENCIL_TEST, GL_FALSE);
use_rectlist(brw, true);
layers = MAX2(1, fb->MaxNumLayers);
if (fast_clear_buffers) {
_mesa_meta_drawbuffers_from_bitfield(fast_clear_buffers);
brw_bind_rep_write_shader(brw, (float *) fast_clear_color);
set_fast_clear_op(brw, GEN7_PS_RENDER_TARGET_FAST_CLEAR_ENABLE);
brw_draw_rectlist(ctx, &fast_clear_rect, layers);
set_fast_clear_op(brw, 0);
}
if (rep_clear_buffers) {
_mesa_meta_drawbuffers_from_bitfield(rep_clear_buffers);
brw_bind_rep_write_shader(brw, ctx->Color.ClearColor.f);
brw_draw_rectlist(ctx, &clear_rect, layers);
}
/* Now set the mts we cleared to INTEL_FAST_CLEAR_STATE_CLEAR so we'll
* resolve them eventually.
*/
for (unsigned buf = 0; buf < fb->_NumColorDrawBuffers; buf++) {
struct gl_renderbuffer *rb = fb->_ColorDrawBuffers[buf];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
int index = fb->_ColorDrawBufferIndexes[buf];
if ((1 << index) & fast_clear_buffers)
irb->mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_CLEAR;
}
bail_to_meta:
/* Dirty _NEW_BUFFERS so we reemit SURFACE_STATE which sets the fast clear
* color before resolve and sets irb->mt->fast_clear_state to UNRESOLVED if
* we render to it.
*/
brw->NewGLState |= _NEW_BUFFERS;
/* Set the custom state back to normal and dirty the same bits as above */
use_rectlist(brw, false);
_mesa_meta_end(ctx);
/* From BSpec: Render Target Fast Clear:
*
* After Render target fast clear, pipe-control with color cache
* write-flush must be issued before sending any DRAW commands on that
* render target.
*/
intel_batchbuffer_emit_mi_flush(brw);
/* If we had to fall back to plain clear for any buffers, clear those now
* by calling into meta.
*/
out:
if (plain_clear_buffers)
_mesa_meta_glsl_Clear(&brw->ctx, plain_clear_buffers);
return true;
}
/**
* 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;
struct intel_renderbuffer *irb;
/* color draw buffers */
for (i = 0; i < ctx->Const.MaxDrawBuffers; i++) {
for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers[i]; j++) {
struct gl_renderbuffer *rb =
ctx->DrawBuffer->_ColorDrawBuffers[i][j];
irb = intel_renderbuffer(rb);
if (irb) {
/* this is a user-created intel_renderbuffer */
if (irb->region) {
if (map)
intel_region_map(intel->intelScreen, irb->region);
else
intel_region_unmap(intel->intelScreen, irb->region);
}
irb->pfMap = irb->region->map;
irb->pfPitch = irb->region->pitch;
}
}
}
/* 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 */
irb = intel_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
if (irb && irb->region) {
if (map)
intel_region_map(intel->intelScreen, irb->region);
else
intel_region_unmap(intel->intelScreen, irb->region);
irb->pfMap = irb->region->map;
irb->pfPitch = irb->region->pitch;
}
/* Account for front/back color page flipping.
* The span routines use the pfMap and pfPitch fields which will
* swap the front/back region map/pitch if we're page flipped.
* Do this after mapping, above, so the map field is valid.
*/
#if 0
if (map && ctx->DrawBuffer->Name == 0) {
struct intel_renderbuffer *irbFront
= intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_FRONT_LEFT);
struct intel_renderbuffer *irbBack
= intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_BACK_LEFT);
if (irbBack) {
/* double buffered */
if (intel->sarea->pf_current_page == 0) {
irbFront->pfMap = irbFront->region->map;
irbFront->pfPitch = irbFront->region->pitch;
irbBack->pfMap = irbBack->region->map;
irbBack->pfPitch = irbBack->region->pitch;
}
else {
irbFront->pfMap = irbBack->region->map;
irbFront->pfPitch = irbBack->region->pitch;
irbBack->pfMap = irbFront->region->map;
irbBack->pfPitch = irbFront->region->pitch;
}
}
}
#endif
/* depth buffer (Note wrapper!) */
if (ctx->DrawBuffer->_DepthBuffer) {
irb = intel_renderbuffer(ctx->DrawBuffer->_DepthBuffer->Wrapped);
if (irb && irb->region && irb->Base.Name != 0) {
if (map) {
intel_region_map(intel->intelScreen, irb->region);
irb->pfMap = irb->region->map;
irb->pfPitch = irb->region->pitch;
}
else {
intel_region_unmap(intel->intelScreen, irb->region);
irb->pfMap = NULL;
irb->pfPitch = 0;
}
}
}
/* stencil buffer (Note wrapper!) */
if (ctx->DrawBuffer->_StencilBuffer) {
irb = intel_renderbuffer(ctx->DrawBuffer->_StencilBuffer->Wrapped);
if (irb && irb->region && irb->Base.Name != 0) {
if (map) {
intel_region_map(intel->intelScreen, irb->region);
irb->pfMap = irb->region->map;
irb->pfPitch = irb->region->pitch;
}
else {
intel_region_unmap(intel->intelScreen, irb->region);
irb->pfMap = NULL;
irb->pfPitch = 0;
}
}
}
}
