AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvk: Only copy 32-bits for cond render operand A

Browse Source 
Now that we're guaranteed the upper 32 bits are zero initialized,
there's no reason we need to do a 64-bit write here.

This is a 0.3% performance improvement on the Sascha Willems
conditionalrender demo with all rendering disabled (638 fps -> 640 fps)

Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37187>
This commit is contained in:
Mel Henning
2025-09-04 15:40:44 -04:00
committed by Marge Bot
parent 4d8e2f7768
commit 1c764357e8
1 changed files with 7 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/nouveau/vulkan/nvk_cmd_draw.c
+7 -17
View File
@@ -4866,7 +4866,7 @@ nvk_CmdBeginConditionalRenderingEXT(VkCommandBuffer commandBuffer,
}
/* Zero-initialize the beginning of the buffer. As an invariant, the bytes
* for operand B are always zero
* for operand B and the upper half of operand A are always zero.
*/
assert(cmd->cond_render_mem->mem->size_B > 32);
memset(cmd->cond_render_mem->mem->map, 0x00, 32);
@@ -4876,11 +4876,7 @@ nvk_CmdBeginConditionalRenderingEXT(VkCommandBuffer commandBuffer,
/* Frustratingly, the u64s are not packed together */
const uint64_t operand_a_addr = tmp_addr + 0;
struct nv_push *p = nvk_cmd_buffer_push(cmd, 20);
P_MTHD(p, NV90B5, LINE_LENGTH_IN);
P_NV90B5_LINE_LENGTH_IN(p, 1);
P_NV90B5_LINE_COUNT(p, 1);
struct nv_push *p = nvk_cmd_buffer_push(cmd, 19);
/* Copy value into operand A */
P_MTHD(p, NV90B5, OFFSET_IN_UPPER);
@@ -4888,16 +4884,10 @@ nvk_CmdBeginConditionalRenderingEXT(VkCommandBuffer commandBuffer,
P_NV90B5_OFFSET_IN_LOWER(p, addr & 0xffffffff);
P_NV90B5_OFFSET_OUT_UPPER(p, operand_a_addr >> 32);
P_NV90B5_OFFSET_OUT_LOWER(p, operand_a_addr & 0xffffffff);
P_IMMD(p, NV90B5, SET_REMAP_COMPONENTS, {
.dst_x = DST_X_SRC_X,
.dst_y = DST_Y_SRC_X,
.dst_z = DST_Z_NO_WRITE,
.dst_w = DST_W_NO_WRITE,
.component_size = COMPONENT_SIZE_FOUR,
.num_src_components = NUM_SRC_COMPONENTS_ONE,
.num_dst_components = NUM_DST_COMPONENTS_TWO,
});
P_NV90B5_PITCH_IN(p, 1);
P_NV90B5_PITCH_OUT(p, 1);
P_NV90B5_LINE_LENGTH_IN(p, 4);
P_NV90B5_LINE_COUNT(p, 1);
P_IMMD(p, NV90B5, LAUNCH_DMA, {
.data_transfer_type = DATA_TRANSFER_TYPE_PIPELINED,
@@ -4905,7 +4895,7 @@ nvk_CmdBeginConditionalRenderingEXT(VkCommandBuffer commandBuffer,
.flush_enable = FLUSH_ENABLE_TRUE,
.src_memory_layout = SRC_MEMORY_LAYOUT_PITCH,
.dst_memory_layout = DST_MEMORY_LAYOUT_PITCH,
.remap_enable = REMAP_ENABLE_TRUE,
.remap_enable = REMAP_ENABLE_FALSE,
});
/* Compare the operands */