const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports;
struct anv_state sf_clip_state =
anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 64, 64);
- struct anv_state cc_state =
- anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32);
for (uint32_t i = 0; i < count; i++) {
const VkViewport *vp = &viewports[i];
.YMaxViewPort = vp->y + vp->height - 1,
};
+ GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64,
+ &sf_clip_viewport);
+ }
+
+ if (!cmd_buffer->device->info.has_llc)
+ anv_state_clflush(sf_clip_state);
+
+ anv_batch_emit(&cmd_buffer->batch,
+ GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) {
+ clip.SFClipViewportPointer = sf_clip_state.offset;
+ }
+}
+
+void
+gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer,
+ bool depth_clamp_enable)
+{
+ uint32_t count = cmd_buffer->state.dynamic.viewport.count;
+ const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports;
+ struct anv_state cc_state =
+ anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32);
+
+ for (uint32_t i = 0; i < count; i++) {
+ const VkViewport *vp = &viewports[i];
+
struct GENX(CC_VIEWPORT) cc_viewport = {
- .MinimumDepth = vp->minDepth,
- .MaximumDepth = vp->maxDepth
+ .MinimumDepth = depth_clamp_enable ? vp->minDepth : 0.0f,
+ .MaximumDepth = depth_clamp_enable ? vp->maxDepth : 1.0f,
};
- GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64,
- &sf_clip_viewport);
GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport);
}
- if (!cmd_buffer->device->info.has_llc) {
- anv_state_clflush(sf_clip_state);
+ if (!cmd_buffer->device->info.has_llc)
anv_state_clflush(cc_state);
- }
anv_batch_emit(&cmd_buffer->batch,
GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) {
cc.CCViewportPointer = cc_state.offset;
}
- anv_batch_emit(&cmd_buffer->batch,
- GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) {
- clip.SFClipViewportPointer = sf_clip_state.offset;
- }
}
#endif
cmd_buffer->state.pipeline->gen8.sf);
}
-#include "genxml/gen9_pack.h"
-static void
-__emit_gen9_sf_state(struct anv_cmd_buffer *cmd_buffer)
+void
+gen9_emit_sf_state(struct anv_cmd_buffer *cmd_buffer);
+
+#if GEN_GEN == 9
+
+void
+gen9_emit_sf_state(struct anv_cmd_buffer *cmd_buffer)
{
- uint32_t sf_dw[GENX(3DSTATE_SF_length)];
- struct GEN9_3DSTATE_SF sf = {
- GEN9_3DSTATE_SF_header,
- .LineWidth = cmd_buffer->state.dynamic.line_width,
- };
- GEN9_3DSTATE_SF_pack(NULL, sf_dw, &sf);
- /* FIXME: gen9.fs */
- anv_batch_emit_merge(&cmd_buffer->batch, sf_dw,
- cmd_buffer->state.pipeline->gen8.sf);
+ __emit_genx_sf_state(cmd_buffer);
}
+#endif
+
+#if GEN_GEN == 8
+
static void
__emit_sf_state(struct anv_cmd_buffer *cmd_buffer)
{
if (cmd_buffer->device->info.is_cherryview)
- __emit_gen9_sf_state(cmd_buffer);
+ gen9_emit_sf_state(cmd_buffer);
else
__emit_genx_sf_state(cmd_buffer);
}
+#else
+
+static void
+__emit_sf_state(struct anv_cmd_buffer *cmd_buffer)
+{
+ __emit_genx_sf_state(cmd_buffer);
+}
+
+#endif
+
void
genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer)
{
cmd_buffer->state.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER;
}
-static VkResult
-flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
+
+/**
+ * Emit the HZ_OP packet in the sequence specified by the BDW PRM section
+ * entitled: "Optimized Depth Buffer Clear and/or Stencil Buffer Clear."
+ *
+ * \todo Enable Stencil Buffer-only clears
+ */
+void
+genX(cmd_buffer_emit_hz_op)(struct anv_cmd_buffer *cmd_buffer,
+ enum blorp_hiz_op op)
{
- struct anv_device *device = cmd_buffer->device;
- struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
- struct anv_state surfaces = { 0, }, samplers = { 0, };
- VkResult result;
-
- result = anv_cmd_buffer_emit_samplers(cmd_buffer,
- MESA_SHADER_COMPUTE, &samplers);
- if (result != VK_SUCCESS)
- return result;
- result = anv_cmd_buffer_emit_binding_table(cmd_buffer,
- MESA_SHADER_COMPUTE, &surfaces);
- if (result != VK_SUCCESS)
- return result;
-
- struct anv_state push_state = anv_cmd_buffer_cs_push_constants(cmd_buffer);
-
- const struct brw_cs_prog_data *cs_prog_data = get_cs_prog_data(pipeline);
- const struct brw_stage_prog_data *prog_data = &cs_prog_data->base;
-
- if (push_state.alloc_size) {
- anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), curbe) {
- curbe.CURBETotalDataLength = push_state.alloc_size;
- curbe.CURBEDataStartAddress = push_state.offset;
- }
- }
+ struct anv_cmd_state *cmd_state = &cmd_buffer->state;
+ const struct anv_image_view *iview =
+ anv_cmd_buffer_get_depth_stencil_view(cmd_buffer);
- const uint32_t slm_size = encode_slm_size(GEN_GEN, prog_data->total_shared);
-
- struct anv_state state =
- anv_state_pool_emit(&device->dynamic_state_pool,
- GENX(INTERFACE_DESCRIPTOR_DATA), 64,
- .KernelStartPointer = pipeline->cs_simd,
- .KernelStartPointerHigh = 0,
- .BindingTablePointer = surfaces.offset,
- .BindingTableEntryCount = 0,
- .SamplerStatePointer = samplers.offset,
- .SamplerCount = 0,
- .ConstantIndirectURBEntryReadLength =
- cs_prog_data->push.per_thread.regs,
- .ConstantURBEntryReadOffset = 0,
- .BarrierEnable = cs_prog_data->uses_barrier,
- .SharedLocalMemorySize = slm_size,
- .NumberofThreadsinGPGPUThreadGroup =
- cs_prog_data->threads,
- .CrossThreadConstantDataReadLength =
- cs_prog_data->push.cross_thread.regs);
-
- uint32_t size = GENX(INTERFACE_DESCRIPTOR_DATA_length) * sizeof(uint32_t);
- anv_batch_emit(&cmd_buffer->batch,
- GENX(MEDIA_INTERFACE_DESCRIPTOR_LOAD), mid) {
- mid.InterfaceDescriptorTotalLength = size;
- mid.InterfaceDescriptorDataStartAddress = state.offset;
- }
+ if (iview == NULL || !anv_image_has_hiz(iview->image))
+ return;
- return VK_SUCCESS;
-}
+ /* FINISHME: Implement multi-subpass HiZ */
+ if (cmd_buffer->state.pass->subpass_count > 1)
+ return;
-void
-genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
-{
- struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
- MAYBE_UNUSED VkResult result;
+ const uint32_t ds = cmd_state->subpass->depth_stencil_attachment;
- assert(pipeline->active_stages == VK_SHADER_STAGE_COMPUTE_BIT);
+ /* Section 7.4. of the Vulkan 1.0.27 spec states:
+ *
+ * "The render area must be contained within the framebuffer dimensions."
+ *
+ * Therefore, the only way the extent of the render area can match that of
+ * the image view is if the render area offset equals (0, 0).
+ */
+ const bool full_surface_op =
+ cmd_state->render_area.extent.width == iview->extent.width &&
+ cmd_state->render_area.extent.height == iview->extent.height;
+ if (full_surface_op)
+ assert(cmd_state->render_area.offset.x == 0 &&
+ cmd_state->render_area.offset.y == 0);
+
+ bool depth_clear;
+ bool stencil_clear;
+
+ /* This variable corresponds to the Pixel Dim column in the table below */
+ struct isl_extent2d px_dim;
+
+ /* Validate that we can perform the HZ operation and that it's necessary. */
+ switch (op) {
+ case BLORP_HIZ_OP_DEPTH_CLEAR:
+ stencil_clear = VK_IMAGE_ASPECT_STENCIL_BIT &
+ cmd_state->attachments[ds].pending_clear_aspects;
+ depth_clear = VK_IMAGE_ASPECT_DEPTH_BIT &
+ cmd_state->attachments[ds].pending_clear_aspects;
+
+ /* Apply alignment restrictions. Despite the BDW PRM mentioning this is
+ * only needed for a depth buffer surface type of D16_UNORM, testing
+ * showed it to be necessary for other depth formats as well
+ * (e.g., D32_FLOAT).
+ */
+#if GEN_GEN == 8
+ /* Pre-SKL, HiZ has an 8x4 sample block. As the number of samples
+ * increases, the number of pixels representable by this block
+ * decreases by a factor of the sample dimensions. Sample dimensions
+ * scale following the MSAA interleaved pattern.
+ *
+ * Sample|Sample|Pixel
+ * Count |Dim |Dim
+ * ===================
+ * 1 | 1x1 | 8x4
+ * 2 | 2x1 | 4x4
+ * 4 | 2x2 | 4x2
+ * 8 | 4x2 | 2x2
+ * 16 | 4x4 | 2x1
+ *
+ * Table: Pixel Dimensions in a HiZ Sample Block Pre-SKL
+ */
+ /* This variable corresponds to the Sample Dim column in the table
+ * above.
+ */
+ const struct isl_extent2d sa_dim =
+ isl_get_interleaved_msaa_px_size_sa(iview->image->samples);
+ px_dim.w = 8 / sa_dim.w;
+ px_dim.h = 4 / sa_dim.h;
+#elif GEN_GEN >= 9
+ /* SKL+, the sample block becomes a "pixel block" so the expected
+ * pixel dimension is a constant 8x4 px for all sample counts.
+ */
+ px_dim = (struct isl_extent2d) { .w = 8, .h = 4};
+#endif
- genX(cmd_buffer_config_l3)(cmd_buffer, pipeline);
+ if (depth_clear && !full_surface_op) {
+ /* Fast depth clears clear an entire sample block at a time. As a
+ * result, the rectangle must be aligned to the pixel dimensions of
+ * a sample block for a successful operation.
+ *
+ * Fast clears can still work if the offset is aligned and the render
+ * area offset + extent touches the edge of a depth buffer whose extent
+ * is unaligned. This is because each physical HiZ miplevel is padded
+ * by the px_dim. In this case, the size of the clear rectangle will be
+ * padded later on in this function.
+ */
+ if (cmd_state->render_area.offset.x % px_dim.w ||
+ cmd_state->render_area.offset.y % px_dim.h)
+ depth_clear = false;
+ if (cmd_state->render_area.offset.x +
+ cmd_state->render_area.extent.width != iview->extent.width &&
+ cmd_state->render_area.extent.width % px_dim.w)
+ depth_clear = false;
+ if (cmd_state->render_area.offset.y +
+ cmd_state->render_area.extent.height != iview->extent.height &&
+ cmd_state->render_area.extent.height % px_dim.h)
+ depth_clear = false;
+ }
- genX(flush_pipeline_select_gpgpu)(cmd_buffer);
+ if (!depth_clear) {
+ if (stencil_clear) {
+ /* Stencil has no alignment requirements */
+ px_dim = (struct isl_extent2d) { .w = 1, .h = 1};
+ } else {
+ /* Nothing to clear */
+ return;
+ }
+ }
+ break;
+ case BLORP_HIZ_OP_DEPTH_RESOLVE:
+ if (cmd_buffer->state.pass->attachments[ds].store_op !=
+ VK_ATTACHMENT_STORE_OP_STORE)
+ return;
+ break;
+ case BLORP_HIZ_OP_HIZ_RESOLVE:
+ /* If the render area covers the entire surface *and* load_op is either
+ * CLEAR or DONT_CARE then the previous contents of the depth buffer
+ * will be entirely discarded. In this case, we can skip the HiZ
+ * resolve.
+ *
+ * If the render area is not the full surface, we need to do
+ * the resolve because otherwise data outside the render area may get
+ * garbled by the resolve at the end of the render pass.
+ */
+ if (full_surface_op &&
+ cmd_buffer->state.pass->attachments[ds].load_op !=
+ VK_ATTACHMENT_LOAD_OP_LOAD)
+ return;
+ break;
+ case BLORP_HIZ_OP_NONE:
+ unreachable("Invalid HiZ OP");
+ break;
+ }
- if (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE)
- anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
+ anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_WM_HZ_OP), hzp) {
+ switch (op) {
+ case BLORP_HIZ_OP_DEPTH_CLEAR:
+ hzp.StencilBufferClearEnable = stencil_clear;
+ hzp.DepthBufferClearEnable = depth_clear;
+ hzp.FullSurfaceDepthandStencilClear = full_surface_op;
+ hzp.StencilClearValue =
+ cmd_state->attachments[ds].clear_value.depthStencil.stencil & 0xff;
+ break;
+ case BLORP_HIZ_OP_DEPTH_RESOLVE:
+ hzp.DepthBufferResolveEnable = true;
+ break;
+ case BLORP_HIZ_OP_HIZ_RESOLVE:
+ hzp.HierarchicalDepthBufferResolveEnable = true;
+ break;
+ case BLORP_HIZ_OP_NONE:
+ unreachable("Invalid HiZ OP");
+ break;
+ }
- if ((cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) ||
- (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE)) {
- result = flush_compute_descriptor_set(cmd_buffer);
- assert(result == VK_SUCCESS);
- cmd_buffer->state.descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT;
+ if (op != BLORP_HIZ_OP_DEPTH_CLEAR) {
+ /* The Optimized HiZ resolve rectangle must be the size of the full RT
+ * and aligned to 8x4. The non-optimized Depth resolve rectangle must
+ * be the size of the full RT. The same alignment is assumed to be
+ * required.
+ */
+ hzp.ClearRectangleXMin = 0;
+ hzp.ClearRectangleYMin = 0;
+ hzp.ClearRectangleXMax = align_u32(iview->extent.width, 8);
+ hzp.ClearRectangleYMax = align_u32(iview->extent.height, 4);
+ } else {
+ /* Contrary to the HW docs both fields are inclusive */
+ hzp.ClearRectangleXMin = cmd_state->render_area.offset.x;
+ hzp.ClearRectangleYMin = cmd_state->render_area.offset.y;
+ /* Contrary to the HW docs both fields are exclusive */
+ hzp.ClearRectangleXMax = cmd_state->render_area.offset.x +
+ align_u32(cmd_state->render_area.extent.width, px_dim.width);
+ hzp.ClearRectangleYMax = cmd_state->render_area.offset.y +
+ align_u32(cmd_state->render_area.extent.height, px_dim.height);
+ }
+
+
+ /* Due to a hardware issue, this bit MBZ */
+ hzp.ScissorRectangleEnable = false;
+ hzp.NumberofMultisamples = ffs(iview->image->samples) - 1;
+ hzp.SampleMask = 0xFFFF;
+ }
+
+ anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ pc.PostSyncOperation = WriteImmediateData;
+ pc.Address =
+ (struct anv_address){ &cmd_buffer->device->workaround_bo, 0 };
}
- cmd_buffer->state.compute_dirty = 0;
+ anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_WM_HZ_OP), hzp);
- genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
+ /* Perform clear specific flushing and state updates */
+ if (op == BLORP_HIZ_OP_DEPTH_CLEAR) {
+ if (depth_clear && !full_surface_op) {
+ anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ pc.DepthStallEnable = true;
+ pc.DepthCacheFlushEnable = true;
+ }
+ }
+
+ /* Remove cleared aspects from the pending mask */
+ if (stencil_clear) {
+ cmd_state->attachments[ds].pending_clear_aspects &=
+ ~VK_IMAGE_ASPECT_STENCIL_BIT;
+ }
+ if (depth_clear) {
+ cmd_state->attachments[ds].pending_clear_aspects &=
+ ~VK_IMAGE_ASPECT_DEPTH_BIT;
+ }
+ }
}
+/* Set of stage bits for which are pipelined, i.e. they get queued by the
+ * command streamer for later execution.
+ */
+#define ANV_PIPELINE_STAGE_PIPELINED_BITS \
+ (VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | \
+ VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | \
+ VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | \
+ VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | \
+ VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | \
+ VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | \
+ VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | \
+ VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | \
+ VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | \
+ VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | \
+ VK_PIPELINE_STAGE_TRANSFER_BIT | \
+ VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT | \
+ VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | \
+ VK_PIPELINE_STAGE_ALL_COMMANDS_BIT)
+
void genX(CmdSetEvent)(
VkCommandBuffer commandBuffer,
VkEvent _event,
ANV_FROM_HANDLE(anv_event, event, _event);
anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
+ pc.StallAtPixelScoreboard = true;
+ pc.CommandStreamerStallEnable = true;
+ }
+
pc.DestinationAddressType = DAT_PPGTT,
pc.PostSyncOperation = WriteImmediateData,
pc.Address = (struct anv_address) {
ANV_FROM_HANDLE(anv_event, event, _event);
anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+ if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) {
+ pc.StallAtPixelScoreboard = true;
+ pc.CommandStreamerStallEnable = true;
+ }
+
pc.DestinationAddressType = DAT_PPGTT;
pc.PostSyncOperation = WriteImmediateData;
pc.Address = (struct anv_address) {