[spv-in] sampling from depth textures returns a scalar and is splatted

CHANGELOG.md

@@ -99,6 +99,7 @@ By @bradwerth [#6216](https://github.com/gfx-rs/wgpu/pull/6216).
 
 #### Naga
 
+- SPIR-V frontend splats depth texture sample and load results. Fixes [issue #4551](https://github.com/gfx-rs/wgpu/issues/4551). By @schell in [#6384](https://github.com/gfx-rs/wgpu/pull/6384).
 - Accept only `vec3` (not `vecN`) for the `cross` built-in. By @ErichDonGubler in [#6171](https://github.com/gfx-rs/wgpu/pull/6171).
 - Configure `SourceLanguage` when enabling debug info in SPV-out. By @kvark in [#6256](https://github.com/gfx-rs/wgpu/pull/6256).
 - Per-polygon and flat inputs should not be considered subgroup uniform. By @magcius in [#6276](https://github.com/gfx-rs/wgpu/pull/6276).

naga/src/front/spv/image.rs

@@ -377,36 +377,60 @@ impl<I: Iterator<Item = u32>> super::Frontend<I> {
  let coord_handle =
  self.get_expr_handle(coordinate_id, coord_lexp, ctx, emitter, block, body_idx);
  let coord_type_handle = self.lookup_type.lookup(coord_lexp.type_id)?.handle;
- let (coordinate, array_index) = match ctx.type_arena[image_ty].inner {
+ let (coordinate, array_index, is_depth) = match ctx.type_arena[image_ty].inner {
  crate::TypeInner::Image {
  dim,
  arrayed,
- class: _,
- } => extract_image_coordinates(
- dim,
- if arrayed {
- ExtraCoordinate::ArrayLayer
- } else {
- ExtraCoordinate::Garbage
- },
- coord_handle,
- coord_type_handle,
- ctx,
- ),
+ class,
+ } => {
+ let (coord, array_index) = extract_image_coordinates(
+ dim,
+ if arrayed {
+ ExtraCoordinate::ArrayLayer
+ } else {
+ ExtraCoordinate::Garbage
+ },
+ coord_handle,
+ coord_type_handle,
+ ctx,
+ );
+ (coord, array_index, class.is_depth())
+ }
  _ => return Err(Error::InvalidImage(image_ty)),
  };
 
- let expr = crate::Expression::ImageLoad {
+ let image_load_expr = crate::Expression::ImageLoad {
  image: image_lexp.handle,
  coordinate,
  array_index,
  sample,
  level,
  };
+ let image_load_handle = ctx
+ .expressions
+ .append(image_load_expr, self.span_from_with_op(start));
+
+ let handle = if is_depth {
+ let result_ty = self.lookup_type.lookup(result_type_id)?;
+ match ctx.type_arena[result_ty.handle].inner {
+ crate::TypeInner::Vector { size, .. } => {
+ let splat_expr = crate::Expression::Splat {
+ size,
+ value: image_load_handle,
+ };
+ ctx.expressions
+ .append(splat_expr, self.span_from_with_op(start))
+ }
+ _ => image_load_handle,
+ }
+ } else {
+ image_load_handle
+ };
+
  self.lookup_expression.insert(
  result_id,
  LookupExpression {
- handle: ctx.expressions.append(expr, self.span_from_with_op(start)),
+ handle,
  type_id: result_type_id,
  block_id,
  },
@@ -593,11 +617,12 @@ impl<I: Iterator<Item = u32>> super::Frontend<I> {
  ref other => return Err(Error::InvalidGlobalVar(other.clone())),
  }
 
- let ((coordinate, array_index), depth_ref) = match ctx.type_arena[image_ty].inner {
+ let ((coordinate, array_index), depth_ref, is_depth) = match ctx.type_arena[image_ty].inner
+ {
  crate::TypeInner::Image {
  dim,
  arrayed,
- class: _,
+ class,
  } => (
  extract_image_coordinates(
  dim,
@@ -642,6 +667,7 @@ impl<I: Iterator<Item = u32>> super::Frontend<I> {
  None => None,
  }
  },
+ class.is_depth(),
  ),
  _ => return Err(Error::InvalidImage(image_ty)),
  };
@@ -656,10 +682,27 @@ impl<I: Iterator<Item = u32>> super::Frontend<I> {
  level,
  depth_ref,
  };
+ let image_sample_handle = ctx.expressions.append(expr, self.span_from_with_op(start));
+ let handle = if is_depth && depth_ref.is_none() {
+ let result_ty = self.lookup_type.lookup(result_type_id)?;
+ match ctx.type_arena[result_ty.handle].inner {
+ crate::TypeInner::Vector { size, .. } => {
+ let splat_expr = crate::Expression::Splat {
+ size,
+ value: image_sample_handle,
+ };
+ ctx.expressions
+ .append(splat_expr, self.span_from_with_op(start))
+ }
+ _ => image_sample_handle,
+ }
+ } else {
+ image_sample_handle
+ };
  self.lookup_expression.insert(
  result_id,
  LookupExpression {
- handle: ctx.expressions.append(expr, self.span_from_with_op(start)),
+ handle,
  type_id: result_type_id,
  block_id,
  },

naga/src/proc/mod.rs

@@ -622,6 +622,10 @@ impl super::ImageClass {
  crate::ImageClass::Storage { .. } => false,
  }
  }
+
+ pub const fn is_depth(self) -> bool {
+ matches!(self, crate::ImageClass::Depth { .. })
+ }
 }
 
 impl crate::Module {

naga/tests/in/spv/fetch_depth.spvasm

@@ -0,0 +1,52 @@
+; SPIR-V
+; Version: 1.5
+; Generator: Google rspirv; 0
+; Bound: 56
+; Schema: 0
+ OpCapability Shader
+ OpCapability VulkanMemoryModel
+ OpMemoryModel Logical Vulkan
+ OpEntryPoint GLCompute %1 "cull::fetch_depth" %2 %3 %4
+ OpExecutionMode %1 LocalSize 32 1 1
+ OpDecorate %_struct_10 Block
+ OpMemberDecorate %_struct_10 0 Offset 0
+ OpDecorate %_struct_11 Block
+ OpMemberDecorate %_struct_11 0 Offset 0
+ OpDecorate %2 Binding 0
+ OpDecorate %2 DescriptorSet 0
+ OpDecorate %3 NonWritable
+ OpDecorate %3 Binding 1
+ OpDecorate %3 DescriptorSet 0
+ OpDecorate %4 Binding 2
+ OpDecorate %4 DescriptorSet 0
+ %uint = OpTypeInt 32 0
+ %float = OpTypeFloat 32
+ %v4float = OpTypeVector %float 4
+ %void = OpTypeVoid
+ %16 = OpTypeFunction %void
+ %uint_0 = OpConstant %uint 0
+ %int = OpTypeInt 32 1
+ %int_0 = OpConstant %int 0
+ %_struct_10 = OpTypeStruct %float
+%_ptr_StorageBuffer__struct_10 = OpTypePointer StorageBuffer %_struct_10
+ %v2uint = OpTypeVector %uint 2
+ %_struct_11 = OpTypeStruct %v2uint
+%_ptr_StorageBuffer__struct_11 = OpTypePointer StorageBuffer %_struct_11
+ %24 = OpTypeImage %float 2D 1 0 0 1 Unknown
+%_ptr_UniformConstant_24 = OpTypePointer UniformConstant %24
+%_ptr_StorageBuffer_float = OpTypePointer StorageBuffer %float
+ %2 = OpVariable %_ptr_StorageBuffer__struct_10 StorageBuffer
+%_ptr_StorageBuffer_v2uint = OpTypePointer StorageBuffer %v2uint
+ %3 = OpVariable %_ptr_StorageBuffer__struct_11 StorageBuffer
+ %4 = OpVariable %_ptr_UniformConstant_24 UniformConstant
+ %1 = OpFunction %void None %16
+ %32 = OpLabel
+ %33 = OpInBoundsAccessChain %_ptr_StorageBuffer_float %2 %uint_0
+ %34 = OpInBoundsAccessChain %_ptr_StorageBuffer_v2uint %3 %uint_0
+ %35 = OpLoad %v2uint %34
+ %54 = OpLoad %24 %4
+ %55 = OpImageFetch %v4float %54 %35 Lod %int_0
+ %38 = OpCompositeExtract %float %55 0
+ OpStore %33 %38
+ OpReturn
+ OpFunctionEnd

naga/tests/out/hlsl/fetch_depth.hlsl

@@ -0,0 +1,25 @@
+struct type_2 {
+ float member;
+};
+
+struct type_4 {
+ uint2 member;
+};
+
+RWByteAddressBuffer global : register(u0);
+ByteAddressBuffer global_1 : register(t1);
+Texture2D<float> global_2 : register(t2);
+
+void function()
+{
+ uint2 _e6 = asuint(global_1.Load2(0));
+ float _e7 = global_2.Load(int3(_e6, 0)).x;
+ global.Store(0, asuint((_e7).xxxx.x));
+ return;
+}
+
+[numthreads(32, 1, 1)]
+void cullfetch_depth()
+{
+ function();
+}

naga/tests/out/hlsl/fetch_depth.ron

@@ -0,0 +1,12 @@
+(
+ vertex:[
+ ],
+ fragment:[
+ ],
+ compute:[
+ (
+ entry_point:"cullfetch_depth",
+ target_profile:"cs_5_1",
+ ),
+ ],
+)