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

radv/rt: Skip instances after loading the entire node

Browse Source 
This avoids waiting for instance_data which can improve performance:

vk_ray_tracing_ao_KHR_app: 0.2% (The TLAS has 2 instances)
Quake II RTX: 1%
Control: 1%

We also have to shuffle around some code to avoid increasing VGPR usage.
That leaves us with the following stats:

Quake II RTX:
Totals from 7 (14.29% of 49) affected shaders:
CodeSize: 165612 -> 165716 (+0.06%)
Instrs: 31446 -> 31460 (+0.04%)
Latency: 596709 -> 554292 (-7.11%)
InvThroughput: 121998 -> 113327 (-7.11%)
VClause: 596 -> 587 (-1.51%)
Copies: 4664 -> 4646 (-0.39%)
PreVGPRs: 620 -> 639 (+3.06%)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21421>
This commit is contained in:
Konstantin Seurer
2023-02-20 14:34:04 +01:00
committed by Marge Bot
parent ab3184c0a2
commit af19762935
1 changed files with 9 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/amd/vulkan/radv_rt_common.c
+9 -7
View File
@@ -660,8 +660,17 @@ radv_build_ray_traversal(struct radv_device *device, nir_builder *b,
/* instance */
nir_ssa_def *instance_node_addr =
build_node_to_addr(device, b, global_bvh_node, false);
nir_store_deref(b, args->vars.instance_addr, instance_node_addr, 1);
nir_ssa_def *instance_data = nir_build_load_global(
b, 4, 32, instance_node_addr, .align_mul = 64, .align_offset = 0);
nir_ssa_def *wto_matrix[3];
nir_build_wto_matrix_load(b, instance_node_addr, wto_matrix);
nir_store_deref(b, args->vars.sbt_offset_and_flags, nir_channel(b, instance_data, 3),
1);
nir_ssa_def *instance_and_mask = nir_channel(b, instance_data, 2);
nir_ssa_def *instance_mask = nir_ushr_imm(b, instance_and_mask, 24);
@@ -671,9 +680,6 @@ radv_build_ray_traversal(struct radv_device *device, nir_builder *b,
}
nir_pop_if(b, NULL);
nir_ssa_def *wto_matrix[3];
nir_build_wto_matrix_load(b, instance_node_addr, wto_matrix);
nir_store_deref(b, args->vars.top_stack, nir_load_deref(b, args->vars.stack), 1);
nir_store_deref(b, args->vars.bvh_base,
nir_pack_64_2x32(b, nir_channels(b, instance_data, 0x3)), 1);
@@ -691,10 +697,6 @@ radv_build_ray_traversal(struct radv_device *device, nir_builder *b,
nir_build_vec3_mat_mult(b, args->dir, wto_matrix, false), 7);
nir_store_deref(b, args->vars.inv_dir,
nir_fdiv(b, vec3ones, nir_load_deref(b, args->vars.dir)), 7);
nir_store_deref(b, args->vars.sbt_offset_and_flags, nir_channel(b, instance_data, 3),
1);
nir_store_deref(b, args->vars.instance_addr, instance_node_addr, 1);
}
nir_pop_if(b, NULL);
}