OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

radv: Fiddle with opaque flag positions to reduce instructions.

Browse Source 
Totals from 7 (0.01% of 134913) affected shaders:

CodeSize: 209076 -> 208460 (-0.29%)
Instrs: 38374 -> 38276 (-0.26%)
Latency: 803899 -> 803181 (-0.09%)
InvThroughput: 165530 -> 165384 (-0.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19706>
This commit is contained in:
Bas Nieuwenhuizen
2022-11-12 02:52:25 +01:00
committed by Marge Bot
parent 3884210902
commit d09ed23b9a
4 changed files with 53 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/amd/vulkan/bvh/build_helpers.h
View File

@@ -157,6 +157,13 @@
#define VK_GEOMETRY_TYPE_TRIANGLES_KHR 0
#define VK_GEOMETRY_TYPE_AABBS_KHR 1
#define VK_GEOMETRY_OPAQUE_BIT_KHR 1
#define VK_GEOMETRY_INSTANCE_TRIANGLE_FACING_CULL_DISABLE_BIT_KHR 1
#define VK_GEOMETRY_INSTANCE_TRIANGLE_FLIP_FACING_BIT_KHR 2
#define VK_GEOMETRY_INSTANCE_FORCE_OPAQUE_BIT_KHR 4
#define VK_GEOMETRY_INSTANCE_FORCE_NO_OPAQUE_BIT_KHR 8
#define TYPE(type, align) \
layout(buffer_reference, buffer_reference_align = align, scalar) buffer type##_ref \
{ \

7
src/amd/vulkan/bvh/bvh.h
View File

@@ -35,6 +35,13 @@
#define radv_ir_node_instance 2
#define radv_ir_node_aabb 3
#define RADV_GEOMETRY_OPAQUE (1u << 31)
#define RADV_INSTANCE_FORCE_OPAQUE (1u << 31)
#define RADV_INSTANCE_NO_FORCE_NOT_OPAQUE (1u << 30)
#define RADV_INSTANCE_TRIANGLE_FACING_CULL_DISABLE (1u << 29)
#define RADV_INSTANCE_TRIANGLE_FLIP_FACING (1u << 28)
#ifdef VULKAN
#define VK_UUID_SIZE 16
#else

32
src/amd/vulkan/bvh/converter_leaf.comp
View File

@@ -43,6 +43,32 @@ layout(push_constant) uniform CONSTS {
convert_leaf_args args;
};
uint32_t
convert_geometry_id_and_flags(uint32_t src)
{
uint32_t flags = src >> 28;
uint32_t ret = src & 0xfffffffu;
if ((flags & VK_GEOMETRY_OPAQUE_BIT_KHR) != 0)
ret |= RADV_GEOMETRY_OPAQUE;
return ret;
}
uint32_t
convert_sbt_offset_and_flags(uint32_t src)
{
uint32_t flags = src >> 24;
uint32_t ret = src & 0xffffffu;
if ((flags & VK_GEOMETRY_INSTANCE_FORCE_OPAQUE_BIT_KHR) != 0)
ret |= RADV_INSTANCE_FORCE_OPAQUE;
if ((flags & VK_GEOMETRY_INSTANCE_FORCE_NO_OPAQUE_BIT_KHR) == 0)
ret |= RADV_INSTANCE_NO_FORCE_NOT_OPAQUE;
if ((flags & VK_GEOMETRY_INSTANCE_TRIANGLE_FACING_CULL_DISABLE_BIT_KHR) != 0)
ret |= RADV_INSTANCE_TRIANGLE_FACING_CULL_DISABLE;
if ((flags & VK_GEOMETRY_INSTANCE_TRIANGLE_FLIP_FACING_BIT_KHR) != 0)
ret |= RADV_INSTANCE_TRIANGLE_FLIP_FACING;
return ret;
}
void
main()
{
@@ -60,7 +86,7 @@ main()
DEREF(dst).coords = src.coords;
DEREF(dst).triangle_id = src.triangle_id;
DEREF(dst).geometry_id_and_flags = src.geometry_id_and_flags;
DEREF(dst).geometry_id_and_flags = convert_geometry_id_and_flags(src.geometry_id_and_flags);
DEREF(dst).id = src.id;
break;
}
@@ -71,7 +97,7 @@ main()
INDEX(radv_bvh_aabb_node, dst_leaves, global_id);
DEREF(dst).aabb = src.base.aabb;
DEREF(dst).primitive_id = src.primitive_id;
DEREF(dst).geometry_id_and_flags = src.geometry_id_and_flags;
DEREF(dst).geometry_id_and_flags = convert_geometry_id_and_flags(src.geometry_id_and_flags);
break;
}
default: { /* instances */
@@ -85,7 +111,7 @@ main()
DEREF(dst).bvh_ptr = src.base_ptr + bvh_offset;
DEREF(dst).custom_instance_and_mask = src.custom_instance_and_mask;
DEREF(dst).sbt_offset_and_flags = src.sbt_offset_and_flags;
DEREF(dst).sbt_offset_and_flags = convert_sbt_offset_and_flags(src.sbt_offset_and_flags);
DEREF(dst).instance_id = src.instance_id;
DEREF(dst).bvh_offset = bvh_offset;

25
src/amd/vulkan/radv_rt_common.c
View File

@@ -395,17 +395,9 @@ static nir_ssa_def *
hit_is_opaque(nir_builder *b, nir_ssa_def *sbt_offset_and_flags,
const struct radv_ray_flags *ray_flags, nir_ssa_def *geometry_id_and_flags)
{
nir_ssa_def *geom_force_opaque =
nir_test_mask(b, geometry_id_and_flags, VK_GEOMETRY_OPAQUE_BIT_KHR << 28);
nir_ssa_def *instance_force_opaque =
nir_test_mask(b, sbt_offset_and_flags, VK_GEOMETRY_INSTANCE_FORCE_OPAQUE_BIT_KHR << 24);
nir_ssa_def *instance_force_non_opaque =
nir_test_mask(b, sbt_offset_and_flags, VK_GEOMETRY_INSTANCE_FORCE_NO_OPAQUE_BIT_KHR << 24);
nir_ssa_def *opaque = geom_force_opaque;
opaque = nir_bcsel(b, instance_force_opaque, nir_imm_bool(b, true), opaque);
opaque = nir_bcsel(b, instance_force_non_opaque, nir_imm_bool(b, false), opaque);
nir_ssa_def *opaque =
nir_uge(b, nir_ior(b, geometry_id_and_flags, sbt_offset_and_flags),
nir_imm_int(b, RADV_INSTANCE_FORCE_OPAQUE | RADV_INSTANCE_NO_FORCE_NOT_OPAQUE));
opaque = nir_bcsel(b, ray_flags->force_opaque, nir_imm_bool(b, true), opaque);
opaque = nir_bcsel(b, ray_flags->force_not_opaque, nir_imm_bool(b, false), opaque);
return opaque;
@@ -440,20 +432,19 @@ insert_traversal_triangle_case(struct radv_device *device, nir_builder *b,
nir_push_if(b, nir_flt(b, intersection.t, nir_load_deref(b, args->vars.tmax)));
{
intersection.frontface = nir_flt(b, nir_imm_float(b, 0), div);
nir_ssa_def *switch_ccw =
nir_test_mask(b, nir_load_deref(b, args->vars.sbt_offset_and_flags),
VK_GEOMETRY_INSTANCE_TRIANGLE_FLIP_FACING_BIT_KHR << 24);
nir_ssa_def *switch_ccw = nir_test_mask(b, nir_load_deref(b, args->vars.sbt_offset_and_flags),
RADV_INSTANCE_TRIANGLE_FLIP_FACING);
intersection.frontface = nir_ixor(b, intersection.frontface, switch_ccw);
nir_ssa_def *not_cull = ray_flags->no_skip_triangles;
nir_ssa_def *not_facing_cull =
nir_bcsel(b, intersection.frontface, ray_flags->no_cull_front, ray_flags->no_cull_back);
not_cull = nir_iand(
b, not_cull,
not_cull =
nir_iand(b, not_cull,
nir_ior(b, not_facing_cull,
nir_test_mask(b, nir_load_deref(b, args->vars.sbt_offset_and_flags),
VK_GEOMETRY_INSTANCE_TRIANGLE_FACING_CULL_DISABLE_BIT_KHR << 24)));
RADV_INSTANCE_TRIANGLE_FACING_CULL_DISABLE)));
nir_push_if(b, nir_iand(b,