Reflect SPV_EXT_descriptor_heap heap accesses per entry point - #343
Closed
chaoticbob wants to merge 1 commit into
Closed
Reflect SPV_EXT_descriptor_heap heap accesses per entry point#343chaoticbob wants to merge 1 commit into
chaoticbob wants to merge 1 commit into
Conversation
Parses the SPV_EXT_descriptor_heap and SPV_KHR_untyped_pointers opcodes
that Slang and DXC emit for bindless descriptor-heap access, and
surfaces the result per entry point as two new arrays on
SpvReflectEntryPoint:
resource_heap_accesses[] -- one entry per distinct
(heap variable, OpTypeRuntimeArray) pair reached from the entry
point's static call graph, classified into a Vulkan descriptor
type (sampled image, storage image, uniform/storage texel buffer,
uniform/storage buffer, acceleration structure).
sampler_heap_accesses[] -- same shape for the sampler heap.
Each entry records the heap variable name, the runtime-array type id,
the array stride (UINT32_MAX when the stride is supplied at runtime via
OpConstantSizeOfEXT rather than as a literal), the descriptor type, and
a pointer to the concrete struct type description when the slot is a
StructuredBuffer/ConstantBuffer reinterpreted through OpBufferPointerEXT.
Parser additions in spirv_reflect.c:
- OpUntypedVariableKHR with the ResourceHeapEXT / SamplerHeapEXT
builtins is recognised as a heap-root variable.
- OpUntypedAccessChainKHR rooted at one of those variables is the
point of access; result ids are now captured so downstream opcodes
can match against them.
- OpTypeBufferEXT is a recognised type whose StorageClass operand
drives uniform-buffer vs storage-buffer descriptor classification.
- OpBufferPointerEXT is followed at the access site to recover the
concrete struct type used when the heap slot is the opaque buffer
descriptor placeholder.
- OpConstantSizeOfEXT no longer poisons the constant table; its
driver-supplied value is treated as unknown.
- ArrayStrideIdEXT resolves its constant operand to a literal stride
when the operand is OpConstant, falling back to UINT32_MAX when it
is OpConstantSizeOfEXT.
YAML emission in common/output_stream.cpp grows an
entry_point_heap_accesses block to expose the new arrays.
Out of scope for this change: per-resource enumeration of which heap
indices are touched, OpUntypedImageTexelPointerEXT, OpMemberDecorateIdEXT,
OffsetIdEXT, and a Pointer-style follow-up of OpUntypedAccessChainKHR
through the existing access-chain machinery.
Companion test fixtures and golden YAMLs are added in the parent PR.
chaoticbob
force-pushed
the
descriptor-heap-lib
branch
from
May 12, 2026 04:59
25f1b83 to
2c86160
Compare
| /*! @struct SpvReflectEntryPointResourceHeapAccess | ||
| @brief One distinct resource-heap access pattern (OpUntypedAccessChainKHR | ||
| element type + stride) reachable from this entry point. | ||
| Stride is UINT32_MAX when implicit (e.g. ArrayStride from |
Contributor
There was a problem hiding this comment.
nit, put the Stride comment on/next to the stride member field
| case SpvOpTypeAccelerationStructureKHR: | ||
| return SPV_REFLECT_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR; | ||
| case SpvOpTypeBufferEXT: { | ||
| // SPV_EXT_descriptor_heap opaque buffer slot. Storage Class operand |
Contributor
There was a problem hiding this comment.
nit this is a "tell me what this 4 lines of code is doing, not why" and just be removed IMO
| } break; | ||
|
|
||
| case SpvOpTypeBufferEXT: { | ||
| // SPV_EXT_descriptor_heap opaque buffer-descriptor slot. The Type |
Contributor
There was a problem hiding this comment.
what is the point of this comment?
| } break; | ||
|
|
||
| case SpvDecorationArrayStrideIdEXT: { | ||
| // SPV_EXT_descriptor_heap variant of ArrayStride. The operand is a |
Contributor
There was a problem hiding this comment.
Suggested change
| // SPV_EXT_descriptor_heap variant of ArrayStride. The operand is a | |
| // ArrayStride but with constant ID instead of literal (mainly used for spec constants) |
| if (heap_var_count >= kMaxHeapVars) { | ||
| break; | ||
| } | ||
| heap_vars[heap_var_count].result_id = p_node->result_id; |
Contributor
There was a problem hiding this comment.
why are we creating a new SpvReflectPrvHeapVar struct and not just have a
SpvReflectPrvNode heap_nodes[kMaxHeapVars];as it is just a subset of the Node data
| continue; | ||
| } | ||
| if (heap_var_count >= kMaxHeapVars) { | ||
| break; |
Contributor
There was a problem hiding this comment.
I feel we should have a SPV_REFLECT_ASSERT(false); or something here... we always do the "loop once and get size and redo it and populate"
I realize 16 is probably enough, but if someone has 17 for some reason, they will get a silent bug trying to access it
| // Upper bound on access chains; used to size scratch. | ||
| uint32_t untyped_access_chain_count = 0; | ||
| for (size_t i = 0; i < p_parser->node_count; ++i) { | ||
| if (p_parser->nodes[i].op == SpvOpUntypedAccessChainKHR) { |
Contributor
There was a problem hiding this comment.
Can't we just get this count in the loop above looking for SpvOpUntypedVariableKHR?
Contributor
Author
|
Closing in favor of a combined PR with #342. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part 2 of 2 for #339. Stacked on #342.
Parses
OpUntypedVariableKHR(ResourceHeapEXT/SamplerHeapEXT),OpUntypedAccessChainKHR,OpTypeBufferEXT,OpBufferPointerEXT,OpConstantSizeOfEXT, andArrayStrideIdEXT. Surfaces results asresource_heap_accesses[]/sampler_heap_accesses[]onSpvReflectEntryPoint— one entry per distinct (heap, runtime arraytype) pair reached from the entry point's call graph, with descriptor
type, stride, and resolved struct.
Out of scope: per-index enumeration,
OpUntypedImageTexelPointerEXT,OpMemberDecorateIdEXT,OffsetIdEXT. Fixtures and goldens in #342.