Fix layoutDescriptorStride bitfield truncation for large stride values

[Guang-035]

Issue: #4178

RootCause:
layoutDescriptorStride was declared as a 4-bit bitfield, which can only values 0-15. Since descriptor_stride must be a power of two, any >= 16 (e.g. 512) was silently truncated to 0 — the "not set" sentinel — causing the SPIR-V backend to fall back to instead of emitting the user-specified stride.

Fix:
Widen the field to a full unsigned int to support the complete range allowed by the GL_EXT_descriptor_heap and SPV_EXT_descriptor_heap specs.

[CLAassistant]

All committers have signed the CLA.

[Guang-035]

Hello @Tobski and @spencer-lunarg , Here is the fix for #4178, please help me review.

[Guang-035]

SPIR-V generated with this fix:

; SPIR-V
; Version: 1.0
; Generator: Khronos Glslang Reference Front End; 11
; Bound: 89
; Schema: 0
               OpCapability Shader
               OpCapability UntypedPointersKHR
               OpCapability DescriptorHeapEXT
               OpExtension "SPV_EXT_descriptor_heap"
               OpExtension "SPV_KHR_untyped_pointers"
          %1 = OpExtInstImport "GLSL.std.450"
               OpMemoryModel Logical GLSL450
               OpEntryPoint Vertex %main "main" %outNormal %inNormal %outColor %inColor %outUV %inUV %_ %gl_InstanceIndex %inPos %outInstanceIndex
               OpSource GLSL 450
               OpSourceExtension "GL_EXT_descriptor_heap"
               OpName %main "main"
               OpName %outNormal "outNormal"
               OpName %inNormal "inNormal"
               OpName %outColor "outColor"
               OpName %inColor "inColor"
               OpName %outUV "outUV"
               OpName %inUV "inUV"
               OpName %gl_PerVertex "gl_PerVertex"
               OpMemberName %gl_PerVertex 0 "gl_Position"
               OpMemberName %gl_PerVertex 1 "gl_PointSize"
               OpMemberName %gl_PerVertex 2 "gl_ClipDistance"
               OpMemberName %gl_PerVertex 3 "gl_CullDistance"
               OpName %_ ""
               OpName %resource_heap "resource_heap"
               OpName %PushConsts "PushConsts"
               OpMemberName %PushConsts 0 "samplerIndex"
               OpMemberName %PushConsts 1 "frameIndex"
               OpName %pushConsts "pushConsts"
               OpName %UBO "UBO"
               OpMemberName %UBO 0 "projection"
               OpMemberName %UBO 1 "view"
               OpMemberName %UBO 2 "model"
               OpName %UBO_0 "UBO"
               OpMemberName %UBO_0 0 "projection"
               OpMemberName %UBO_0 1 "view"
               OpMemberName %UBO_0 2 "model"
               OpName %UBO_1 "UBO"
               OpMemberName %UBO_1 0 "projection"
               OpMemberName %UBO_1 1 "view"
               OpMemberName %UBO_1 2 "model"
               OpName %gl_InstanceIndex "gl_InstanceIndex"
               OpName %inPos "inPos"
               OpName %outInstanceIndex "outInstanceIndex"
               OpDecorate %outNormal Location 0
               OpDecorate %inNormal Location 1
               OpDecorate %outColor Location 1
               OpDecorate %inColor Location 3
               OpDecorate %outUV Location 2
               OpDecorate %inUV Location 2
               OpDecorate %gl_PerVertex Block
               OpMemberDecorate %gl_PerVertex 0 BuiltIn Position
               OpMemberDecorate %gl_PerVertex 1 BuiltIn PointSize
               OpMemberDecorate %gl_PerVertex 2 BuiltIn ClipDistance
               OpMemberDecorate %gl_PerVertex 3 BuiltIn CullDistance
               OpDecorate %resource_heap BuiltIn ResourceHeapEXT
               OpDecorate %PushConsts Block
               OpMemberDecorate %PushConsts 0 Offset 0
               OpMemberDecorate %PushConsts 1 Offset 4
               OpDecorate %_arr_mat4v4float_uint_2 ArrayStride 64
               OpDecorate %UBO BufferBlock
               OpMemberDecorate %UBO 0 ColMajor
               OpMemberDecorate %UBO 0 MatrixStride 16
               OpMemberDecorate %UBO 0 Offset 0
               OpMemberDecorate %UBO 1 ColMajor
               OpMemberDecorate %UBO 1 MatrixStride 16
               OpMemberDecorate %UBO 1 Offset 64
               OpMemberDecorate %UBO 2 ColMajor
               OpMemberDecorate %UBO 2 MatrixStride 16
               OpMemberDecorate %UBO 2 Offset 128
               OpDecorateId %_runtimearr_46 ArrayStrideIdEXT %uint_512
               OpDecorate %_arr_mat4v4float_uint_2_0 ArrayStride 64
               OpDecorate %UBO_0 BufferBlock
               OpMemberDecorate %UBO_0 0 ColMajor
               OpMemberDecorate %UBO_0 0 MatrixStride 16
               OpMemberDecorate %UBO_0 0 Offset 0
               OpMemberDecorate %UBO_0 1 ColMajor
               OpMemberDecorate %UBO_0 1 MatrixStride 16
               OpMemberDecorate %UBO_0 1 Offset 64
               OpMemberDecorate %UBO_0 2 ColMajor
               OpMemberDecorate %UBO_0 2 MatrixStride 16
               OpMemberDecorate %UBO_0 2 Offset 128
               OpDecorateId %_runtimearr_46_0 ArrayStrideIdEXT %uint_512
               OpDecorate %_arr_mat4v4float_uint_2_1 ArrayStride 64
               OpDecorate %UBO_1 BufferBlock
               OpMemberDecorate %UBO_1 0 ColMajor
               OpMemberDecorate %UBO_1 0 MatrixStride 16
               OpMemberDecorate %UBO_1 0 Offset 0
               OpMemberDecorate %UBO_1 1 ColMajor
               OpMemberDecorate %UBO_1 1 MatrixStride 16
               OpMemberDecorate %UBO_1 1 Offset 64
               OpMemberDecorate %UBO_1 2 ColMajor
               OpMemberDecorate %UBO_1 2 MatrixStride 16
               OpMemberDecorate %UBO_1 2 Offset 128
               OpDecorate %gl_InstanceIndex BuiltIn InstanceIndex
               OpDecorateId %_runtimearr_46_1 ArrayStrideIdEXT %uint_512
               OpDecorate %inPos Location 0
               OpDecorate %outInstanceIndex Flat
               OpDecorate %outInstanceIndex Location 3
       %void = OpTypeVoid
          %3 = OpTypeFunction %void
      %float = OpTypeFloat 32
    %v3float = OpTypeVector %float 3
%_ptr_Output_v3float = OpTypePointer Output %v3float
  %outNormal = OpVariable %_ptr_Output_v3float Output
%_ptr_Input_v3float = OpTypePointer Input %v3float
   %inNormal = OpVariable %_ptr_Input_v3float Input
   %outColor = OpVariable %_ptr_Output_v3float Output
    %inColor = OpVariable %_ptr_Input_v3float Input
    %v2float = OpTypeVector %float 2
%_ptr_Output_v2float = OpTypePointer Output %v2float
      %outUV = OpVariable %_ptr_Output_v2float Output
%_ptr_Input_v2float = OpTypePointer Input %v2float
       %inUV = OpVariable %_ptr_Input_v2float Input
    %v4float = OpTypeVector %float 4
       %uint = OpTypeInt 32 0
     %uint_1 = OpConstant %uint 1
%_arr_float_uint_1 = OpTypeArray %float %uint_1
%gl_PerVertex = OpTypeStruct %v4float %float %_arr_float_uint_1 %_arr_float_uint_1
%_ptr_Output_gl_PerVertex = OpTypePointer Output %gl_PerVertex
          %_ = OpVariable %_ptr_Output_gl_PerVertex Output
        %int = OpTypeInt 32 1
      %int_0 = OpConstant %int 0
%_ptr_UniformConstant = OpTypeUntypedPointerKHR UniformConstant
%resource_heap = OpUntypedVariableKHR %_ptr_UniformConstant UniformConstant
 %PushConsts = OpTypeStruct %int %int
%_ptr_PushConstant_PushConsts = OpTypePointer PushConstant %PushConsts
 %pushConsts = OpVariable %_ptr_PushConstant_PushConsts PushConstant
      %int_1 = OpConstant %int 1
%_ptr_PushConstant_int = OpTypePointer PushConstant %int
%mat4v4float = OpTypeMatrix %v4float 4
     %uint_2 = OpConstant %uint 2
%_arr_mat4v4float_uint_2 = OpTypeArray %mat4v4float %uint_2
        %UBO = OpTypeStruct %mat4v4float %mat4v4float %_arr_mat4v4float_uint_2
%_ptr_Uniform = OpTypeUntypedPointerKHR Uniform
         %46 = OpTypeBufferEXT Uniform
   %uint_512 = OpConstant %uint 512
%_runtimearr_46 = OpTypeRuntimeArray %46
%_arr_mat4v4float_uint_2_0 = OpTypeArray %mat4v4float %uint_2
      %UBO_0 = OpTypeStruct %mat4v4float %mat4v4float %_arr_mat4v4float_uint_2_0
%_runtimearr_46_0 = OpTypeRuntimeArray %46
%_arr_mat4v4float_uint_2_1 = OpTypeArray %mat4v4float %uint_2
      %UBO_1 = OpTypeStruct %mat4v4float %mat4v4float %_arr_mat4v4float_uint_2_1
      %int_2 = OpConstant %int 2
%_ptr_Input_int = OpTypePointer Input %int
%gl_InstanceIndex = OpVariable %_ptr_Input_int Input
%_runtimearr_46_1 = OpTypeRuntimeArray %46
      %inPos = OpVariable %_ptr_Input_v3float Input
    %float_1 = OpConstant %float 1
%_ptr_Output_v4float = OpTypePointer Output %v4float
%_ptr_Output_int = OpTypePointer Output %int
%outInstanceIndex = OpVariable %_ptr_Output_int Output
       %main = OpFunction %void None %3
          %5 = OpLabel
         %12 = OpLoad %v3float %inNormal
               OpStore %outNormal %12
         %15 = OpLoad %v3float %inColor
               OpStore %outColor %15
         %21 = OpLoad %v2float %inUV
               OpStore %outUV %21
         %38 = OpAccessChain %_ptr_PushConstant_int %pushConsts %int_1
         %39 = OpLoad %int %38
         %45 = OpUntypedAccessChainKHR %_ptr_UniformConstant %_runtimearr_46 %resource_heap %39
         %49 = OpBufferPointerEXT %_ptr_Uniform %45
         %50 = OpUntypedAccessChainKHR %_ptr_Uniform %UBO %49 %int_0
         %51 = OpLoad %mat4v4float %50
         %52 = OpAccessChain %_ptr_PushConstant_int %pushConsts %int_1
         %53 = OpLoad %int %52
         %56 = OpUntypedAccessChainKHR %_ptr_UniformConstant %_runtimearr_46_0 %resource_heap %53
         %58 = OpBufferPointerEXT %_ptr_Uniform %56
         %59 = OpUntypedAccessChainKHR %_ptr_Uniform %UBO_0 %58 %int_1
         %60 = OpLoad %mat4v4float %59
         %61 = OpMatrixTimesMatrix %mat4v4float %51 %60
         %62 = OpAccessChain %_ptr_PushConstant_int %pushConsts %int_1
         %63 = OpLoad %int %62
         %69 = OpLoad %int %gl_InstanceIndex
         %70 = OpUntypedAccessChainKHR %_ptr_UniformConstant %_runtimearr_46_1 %resource_heap %63
         %72 = OpBufferPointerEXT %_ptr_Uniform %70
         %73 = OpUntypedAccessChainKHR %_ptr_Uniform %UBO_1 %72 %int_2 %69
         %74 = OpLoad %mat4v4float %73
         %75 = OpMatrixTimesMatrix %mat4v4float %61 %74
         %77 = OpLoad %v3float %inPos
         %79 = OpCompositeExtract %float %77 0
         %80 = OpCompositeExtract %float %77 1
         %81 = OpCompositeExtract %float %77 2
         %82 = OpCompositeConstruct %v4float %79 %80 %81 %float_1
         %83 = OpMatrixTimesVector %v4float %75 %82
         %85 = OpAccessChain %_ptr_Output_v4float %_ %int_0
               OpStore %85 %83
         %88 = OpLoad %int %gl_InstanceIndex
               OpStore %outInstanceIndex %88
               OpReturn
               OpFunctionEnd

[spencer-lunarg]

@Guang-035 can we add this as a test (or preferably a test of various structs using descriptor_stride)

[Guang-035]

@Guang-035 can we add this as a test (or preferably a test of various structs using descriptor_stride)

Hello @spencer-lunarg , I've added a test with descriptor_stride=512 using a buffer block with mat4 members. Could you clarify what you mean by "various structs"? A GLSL example would be helpful.

[spencer-lunarg]

various structs

Basically test that setting descriptor_stride= is doing what is should be, glad this is working for this specific case, but clearly we were lacking testing for it from the beginning

[Tobski]

Yea I don't think we need to test different structs; the struct definition for the UBO has no interaction with the stride. What might be worth adding a test for is the same but for the stride of an image descriptor. E.g.

layout(descriptor_heap, descriptor_stride = 1024) image2D image;

[Tobski]

(This is more for regression testing - I don't think there's any reason it's necessary to test it for this particular change)

[spencer-lunarg]

@Guang-035 we can't merge this until you sign the CLA

[arcady-lunarg]

@Guang-035 You will need to sign the CLA, then I will be able to merge this.

[Guang-035]

Add an image2D declaration with descriptor_stride=1024 alongside the buffer test with descriptor_stride=512, covering both buffer image descriptor types as suggested by @Tobski .

[SaschaWillems]

Can you provide a pre-compiled binary so I can test if this fixes the issues I've been having?

[spencer-lunarg]

@SaschaWillems https://github.com/KhronosGroup/glslang/releases/tag/main-tot