void
blorp_ccs_resolve(struct blorp_batch *batch,
struct blorp_surf *surf, enum isl_format format)
{
struct blorp_params params;
blorp_params_init(¶ms);
brw_blorp_surface_info_init(batch->blorp, ¶ms.dst, surf,
0 /* level */, 0 /* layer */, format, true);
/* From the Ivy Bridge PRM, Vol2 Part1 11.9 "Render Target Resolve":
*
* A rectangle primitive must be scaled down by the following factors
* with respect to render target being resolved.
*
* The scaledown factors in the table that follows are related to the block
* size of the CCS format. For IVB and HSW, we divide by two, for BDW we
* multiply by 8 and 16. On Sky Lake, we multiply by 8.
*/
const struct isl_format_layout *aux_fmtl =
isl_format_get_layout(params.dst.aux_surf.format);
assert(aux_fmtl->txc == ISL_TXC_CCS);
unsigned x_scaledown, y_scaledown;
if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 9) {
x_scaledown = aux_fmtl->bw * 8;
y_scaledown = aux_fmtl->bh * 8;
} else if (ISL_DEV_GEN(batch->blorp->isl_dev) >= 8) {
x_scaledown = aux_fmtl->bw * 8;
y_scaledown = aux_fmtl->bh * 16;
} else {
x_scaledown = aux_fmtl->bw / 2;
y_scaledown = aux_fmtl->bh / 2;
}
params.x0 = params.y0 = 0;
params.x1 = params.dst.aux_surf.logical_level0_px.width;
params.y1 = params.dst.aux_surf.logical_level0_px.height;
params.x1 = ALIGN(params.x1, x_scaledown) / x_scaledown;
params.y1 = ALIGN(params.y1, y_scaledown) / y_scaledown;
if (batch->blorp->isl_dev->info->gen >= 9) {
if (params.dst.aux_usage == ISL_AUX_USAGE_CCS_E)
params.fast_clear_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL;
else
params.fast_clear_op = BLORP_FAST_CLEAR_OP_RESOLVE_PARTIAL;
} else {
/* Broadwell and earlier do not have a partial resolve */
params.fast_clear_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL;
}
/* Note: there is no need to initialize push constants because it doesn't
* matter what data gets dispatched to the render target. However, we must
* ensure that the fragment shader delivers the data using the "replicated
* color" message.
*/
blorp_params_get_clear_kernel(batch->blorp, ¶ms, true);
batch->blorp->exec(batch, ¶ms);
}
void
blorp_fast_clear(struct blorp_batch *batch,
const struct blorp_surf *surf, enum isl_format format,
uint32_t level, uint32_t start_layer, uint32_t num_layers,
uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1)
{
struct blorp_params params;
blorp_params_init(¶ms);
params.num_layers = num_layers;
params.x0 = x0;
params.y0 = y0;
params.x1 = x1;
params.y1 = y1;
memset(¶ms.wm_inputs, 0xff, 4*sizeof(float));
params.fast_clear_op = BLORP_FAST_CLEAR_OP_CLEAR;
get_fast_clear_rect(batch->blorp->isl_dev, surf->aux_surf,
¶ms.x0, ¶ms.y0, ¶ms.x1, ¶ms.y1);
blorp_params_get_clear_kernel(batch->blorp, ¶ms, true);
brw_blorp_surface_info_init(batch->blorp, ¶ms.dst, surf, level,
start_layer, format, true);
batch->blorp->exec(batch, ¶ms);
}
void
brw_blorp_resolve_color(struct brw_context *brw, struct intel_mipmap_tree *mt)
{
DBG("%s to mt %p\n", __FUNCTION__, mt);
const mesa_format format = _mesa_get_srgb_format_linear(mt->format);
struct brw_blorp_params params;
brw_blorp_params_init(¶ms);
brw_blorp_surface_info_init(brw, ¶ms.dst, mt,
0 /* level */, 0 /* layer */, format, true);
brw_get_resolve_rect(brw, mt, ¶ms.x0, ¶ms.y0,
¶ms.x1, ¶ms.y1);
if (intel_miptree_is_lossless_compressed(brw, mt))
params.resolve_type = GEN9_PS_RENDER_TARGET_RESOLVE_FULL;
else
params.resolve_type = GEN7_PS_RENDER_TARGET_RESOLVE_ENABLE;
/* Note: there is no need to initialize push constants because it doesn't
* matter what data gets dispatched to the render target. However, we must
* ensure that the fragment shader delivers the data using the "replicated
* color" message.
*/
brw_blorp_params_get_clear_kernel(brw, ¶ms, true);
brw_blorp_exec(brw, ¶ms);
mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_RESOLVED;
}
void
blorp_clear_depth_stencil(struct blorp_batch *batch,
const struct blorp_surf *depth,
const struct blorp_surf *stencil,
uint32_t level, uint32_t start_layer,
uint32_t num_layers,
uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
bool clear_depth, float depth_value,
uint8_t stencil_mask, uint8_t stencil_value)
{
struct blorp_params params;
blorp_params_init(¶ms);
params.x0 = x0;
params.y0 = y0;
params.x1 = x1;
params.y1 = y1;
while (num_layers > 0) {
params.num_layers = num_layers;
if (stencil_mask) {
brw_blorp_surface_info_init(batch->blorp, ¶ms.stencil, stencil,
level, start_layer,
ISL_FORMAT_UNSUPPORTED, true);
params.stencil_mask = stencil_mask;
params.stencil_ref = stencil_value;
params.dst.surf.samples = params.stencil.surf.samples;
params.dst.surf.logical_level0_px =
params.stencil.surf.logical_level0_px;
params.dst.view = params.depth.view;
/* We may be restricted on the number of layers we can bind at any
* one time. In particular, Sandy Bridge has a maximum number of
* layers of 512 but a maximum 3D texture size is much larger.
*/
if (params.stencil.view.array_len < params.num_layers)
params.num_layers = params.stencil.view.array_len;
}
if (clear_depth) {
brw_blorp_surface_info_init(batch->blorp, ¶ms.depth, depth,
level, start_layer,
ISL_FORMAT_UNSUPPORTED, true);
params.z = depth_value;
params.depth_format =
isl_format_get_depth_format(depth->surf->format, false);
params.dst.surf.samples = params.depth.surf.samples;
params.dst.surf.logical_level0_px =
params.depth.surf.logical_level0_px;
params.dst.view = params.depth.view;
/* We may be restricted on the number of layers we can bind at any
* one time. In particular, Sandy Bridge has a maximum number of
* layers of 512 but a maximum 3D texture size is much larger.
*/
if (params.depth.view.array_len < params.num_layers)
params.num_layers = params.depth.view.array_len;
}
batch->blorp->exec(batch, ¶ms);
start_layer += params.num_layers;
num_layers -= params.num_layers;
}
}
void
blorp_clear(struct blorp_batch *batch,
const struct blorp_surf *surf,
enum isl_format format, struct isl_swizzle swizzle,
uint32_t level, uint32_t start_layer, uint32_t num_layers,
uint32_t x0, uint32_t y0, uint32_t x1, uint32_t y1,
union isl_color_value clear_color,
const bool color_write_disable[4])
{
struct blorp_params params;
blorp_params_init(¶ms);
params.x0 = x0;
params.y0 = y0;
params.x1 = x1;
params.y1 = y1;
if (format == ISL_FORMAT_R9G9B9E5_SHAREDEXP) {
clear_color.u32[0] = float3_to_rgb9e5(clear_color.f32);
format = ISL_FORMAT_R32_UINT;
}
memcpy(¶ms.wm_inputs, clear_color.f32, 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 (surf->surf->tiling == ISL_TILING_LINEAR)
use_simd16_replicated_data = false;
/* Constant color writes ignore everyting in blend and color calculator
* state. This is not documented.
*/
if (color_write_disable) {
for (unsigned i = 0; i < 4; i++) {
params.color_write_disable[i] = color_write_disable[i];
if (color_write_disable[i])
use_simd16_replicated_data = false;
}
}
blorp_params_get_clear_kernel(batch->blorp, ¶ms,
use_simd16_replicated_data);
while (num_layers > 0) {
brw_blorp_surface_info_init(batch->blorp, ¶ms.dst, surf, level,
start_layer, format, true);
params.dst.view.swizzle = swizzle;
/* We may be restricted on the number of layers we can bind at any one
* time. In particular, Sandy Bridge has a maximum number of layers of
* 512 but a maximum 3D texture size is much larger.
*/
params.num_layers = MIN2(params.dst.view.array_len, num_layers);
batch->blorp->exec(batch, ¶ms);
start_layer += params.num_layers;
num_layers -= params.num_layers;
}
}
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;
}
