radv: clear CMASK layers instead of the whole buffer on GFX8

This reduces the size of fill operations needed to clear CMASK
for layered color textures.

GFX9 unsupported for now.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

diff --git a/src/amd/common/ac_surface.c b/src/amd/common/ac_surface.c
index 05ad181a87a6bc643ae3ae96daf1c5ea0f97e2ee..f8b9d2b70f8aa7a22767d4018e1f99128b08ca81 100644 (file)
--- a/src/amd/common/ac_surface.c
+++ b/src/amd/common/ac_surface.c
@@ -504,7 +504,8 @@ static void ac_compute_cmask(const struct radeon_info *info,
                num_layers = config->info.array_size;
 
        surf->cmask_alignment = MAX2(256, base_align);
-       surf->cmask_size = align(slice_bytes, base_align) * num_layers;
+       surf->cmask_slice_size = align(slice_bytes, base_align);
+       surf->cmask_size = surf->cmask_slice_size * num_layers;
 }
 
 /**

diff --git a/src/amd/common/ac_surface.h b/src/amd/common/ac_surface.h
index aa93e917270c81fed1870dc28cbd0c3378e2447f..31623634936332562b654bf50a02498edf59a8d5 100644 (file)
--- a/src/amd/common/ac_surface.h
+++ b/src/amd/common/ac_surface.h
@@ -219,6 +219,7 @@ struct radeon_surf {
     uint32_t                    htile_alignment;
 
     uint32_t                    cmask_size;
+    uint32_t                    cmask_slice_size;
     uint32_t                    cmask_alignment;
 
     union {

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index ef659a6c48ce5cf259f292c76a1f6d05278674f0..e35ccf809565e9d7a819d36f5c9e9fe395d0959a 100644 (file)
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -4892,14 +4892,16 @@ static void radv_handle_depth_image_transition(struct radv_cmd_buffer *cmd_buffe
 }
 
 static void radv_initialise_cmask(struct radv_cmd_buffer *cmd_buffer,
-                                 struct radv_image *image, uint32_t value)
+                                 struct radv_image *image,
+                                 const VkImageSubresourceRange *range,
+                                 uint32_t value)
 {
        struct radv_cmd_state *state = &cmd_buffer->state;
 
        state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
                            RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
 
-       state->flush_bits |= radv_clear_cmask(cmd_buffer, image, value);
+       state->flush_bits |= radv_clear_cmask(cmd_buffer, image, range, value);
 
        state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
 }
@@ -5005,7 +5007,7 @@ static void radv_init_color_image_metadata(struct radv_cmd_buffer *cmd_buffer,
                        value = 0xccccccccu;
                }
 
-               radv_initialise_cmask(cmd_buffer, image, value);
+               radv_initialise_cmask(cmd_buffer, image, range, value);
        }
 
        if (radv_image_has_fmask(image)) {

diff --git a/src/amd/vulkan/radv_image.c b/src/amd/vulkan/radv_image.c
index ca007d1dd65e8e7014c0fecded201ecf3d160e21..4099c57aa8528f1230809d32cdaa14f33eabfaf5 100644 (file)
--- a/src/amd/vulkan/radv_image.c
+++ b/src/amd/vulkan/radv_image.c
@@ -942,6 +942,7 @@ radv_image_get_cmask_info(struct radv_device *device,
 
        out->slice_tile_max = image->planes[0].surface.u.legacy.cmask_slice_tile_max;
        out->alignment = image->planes[0].surface.cmask_alignment;
+       out->slice_size = image->planes[0].surface.cmask_slice_size;
        out->size = image->planes[0].surface.cmask_size;
 }
 

diff --git a/src/amd/vulkan/radv_meta.h b/src/amd/vulkan/radv_meta.h
index 30981f00790ea214fd974dc5e5417d3485b7bf63..c3d37bb07d24145239efcafb2b970e6c492bc031 100644 (file)
--- a/src/amd/vulkan/radv_meta.h
+++ b/src/amd/vulkan/radv_meta.h
@@ -212,7 +212,8 @@ void radv_decompress_resolve_src(struct radv_cmd_buffer *cmd_buffer,
                                 const VkImageResolve *regions);
 
 uint32_t radv_clear_cmask(struct radv_cmd_buffer *cmd_buffer,
-                         struct radv_image *image, uint32_t value);
+                         struct radv_image *image,
+                         const VkImageSubresourceRange *range, uint32_t value);
 uint32_t radv_clear_fmask(struct radv_cmd_buffer *cmd_buffer,
                          struct radv_image *image,
                          const VkImageSubresourceRange *range, uint32_t value);

diff --git a/src/amd/vulkan/radv_meta_clear.c b/src/amd/vulkan/radv_meta_clear.c
index 0a9d9e76ca424422ce9a9a647c0b230ab4da1582..4d569729dda1ed27e0c8d18f21e8884793fc9fcc 100644 (file)
--- a/src/amd/vulkan/radv_meta_clear.c
+++ b/src/amd/vulkan/radv_meta_clear.c
@@ -1326,11 +1326,21 @@ radv_get_cmask_fast_clear_value(const struct radv_image *image)
 
 uint32_t
 radv_clear_cmask(struct radv_cmd_buffer *cmd_buffer,
-                struct radv_image *image, uint32_t value)
+                struct radv_image *image,
+                const VkImageSubresourceRange *range, uint32_t value)
 {
-       return radv_fill_buffer(cmd_buffer, image->bo,
-                               image->offset + image->cmask.offset,
-                               image->cmask.size, value);
+       uint64_t offset = image->offset + image->cmask.offset;
+       uint64_t size;
+
+       if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
+               /* TODO: clear layers. */
+               size = image->cmask.size;
+       } else {
+               offset += image->cmask.slice_size * range->baseArrayLayer;
+               size = image->cmask.slice_size * radv_get_layerCount(image, range);
+       }
+
+       return radv_fill_buffer(cmd_buffer, image->bo, offset, size, value);
 }
 
 
@@ -1578,6 +1588,13 @@ radv_fast_clear_color(struct radv_cmd_buffer *cmd_buffer,
        VkClearColorValue clear_value = clear_att->clearValue.color;
        uint32_t clear_color[2], flush_bits = 0;
        uint32_t cmask_clear_value;
+       VkImageSubresourceRange range = {
+               .aspectMask = iview->aspect_mask,
+               .baseMipLevel = iview->base_mip,
+               .levelCount = iview->level_count,
+               .baseArrayLayer = iview->base_layer,
+               .layerCount = iview->layer_count,
+       };
 
        if (pre_flush) {
                cmd_buffer->state.flush_bits |= (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
@@ -1595,13 +1612,6 @@ radv_fast_clear_color(struct radv_cmd_buffer *cmd_buffer,
                uint32_t reset_value;
                bool can_avoid_fast_clear_elim;
                bool need_decompress_pass = false;
-               VkImageSubresourceRange range = {
-                       .aspectMask = iview->aspect_mask,
-                       .baseMipLevel = iview->base_mip,
-                       .levelCount = iview->level_count,
-                       .baseArrayLayer = iview->base_layer,
-                       .layerCount = iview->layer_count,
-               };
 
                vi_get_fast_clear_parameters(iview->vk_format,
                                             &clear_value, &reset_value,
@@ -1609,7 +1619,7 @@ radv_fast_clear_color(struct radv_cmd_buffer *cmd_buffer,
 
                if (radv_image_has_cmask(iview->image)) {
                        flush_bits = radv_clear_cmask(cmd_buffer, iview->image,
-                                                     cmask_clear_value);
+                                                     &range, cmask_clear_value);
 
                        need_decompress_pass = true;
                }
@@ -1624,7 +1634,7 @@ radv_fast_clear_color(struct radv_cmd_buffer *cmd_buffer,
                                         need_decompress_pass);
        } else {
                flush_bits = radv_clear_cmask(cmd_buffer, iview->image,
-                                             cmask_clear_value);
+                                             &range, cmask_clear_value);
        }
 
        if (post_flush) {

diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index fd7baa5f5b573906130b0f0a89a8fc8bb36d9289..284d212d027826ac4f387cb62e1e73d275dca969 100644 (file)
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1563,6 +1563,7 @@ struct radv_cmask_info {
        uint64_t size;
        unsigned alignment;
        unsigned slice_tile_max;
+       unsigned slice_size;
 };