vkd3d-shader/hlsl: Support interlocked operations on non-indexed groupshared variables.

Approved-by: Elizabeth Figura (@zfigura)
Approved-by: Henri Verbeet (@hverbeet)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1605

libs/vkd3d-shader/hlsl.c

@@ -2383,6 +2383,19 @@ bool hlsl_index_chain_has_resource_access(struct hlsl_ir_index *index)
     return false;
 }
 
+bool hlsl_index_chain_has_tgsm_access(struct hlsl_ir_index *index)
+{
+    if (index->val.node->type == HLSL_IR_LOAD)
+    {
+        struct hlsl_ir_load *load = hlsl_ir_load(index->val.node);
+        return load->src.var->storage_modifiers & HLSL_STORAGE_GROUPSHARED;
+    }
+
+    if (index->val.node->type == HLSL_IR_INDEX)
+        return hlsl_index_chain_has_tgsm_access(hlsl_ir_index(index->val.node));
+    return false;
+}
+
 static struct hlsl_ir_node *hlsl_new_index(struct hlsl_ctx *ctx, struct hlsl_ir_node *val,
         struct hlsl_ir_node *idx, const struct vkd3d_shader_location *loc)
 {

libs/vkd3d-shader/hlsl.h

@@ -1712,6 +1712,7 @@ struct hlsl_ir_node *hlsl_new_store_index(struct hlsl_ctx *ctx, const struct hls
 bool hlsl_index_is_noncontiguous(struct hlsl_ir_index *index);
 bool hlsl_index_is_resource_access(struct hlsl_ir_index *index);
 bool hlsl_index_chain_has_resource_access(struct hlsl_ir_index *index);
+bool hlsl_index_chain_has_tgsm_access(struct hlsl_ir_index *index);
 
 struct hlsl_ir_node *hlsl_new_compile(struct hlsl_ctx *ctx, enum hlsl_compile_type compile_type,
         const char *profile_name, struct hlsl_ir_node **args, unsigned int args_count,

libs/vkd3d-shader/hlsl.y

@@ -5063,13 +5063,25 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
 
         if (hlsl_deref_get_type(ctx, &dst_deref)->class != HLSL_CLASS_UAV)
         {
-            hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, "Interlocked targets must be UAV elements.");
+            hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE,
+                    "Interlocked targets must be UAV or groupshared elements.");
             return false;
         }
     }
+    else if (lhs->type == HLSL_IR_INDEX && hlsl_index_chain_has_tgsm_access(hlsl_ir_index(lhs)))
+    {
+        hlsl_fixme(ctx, loc, "Interlocked operations on indexed groupshared elements.");
+        return false;
+    }
+    else if (lhs->type == HLSL_IR_LOAD && (hlsl_ir_load(lhs)->src.var->storage_modifiers & HLSL_STORAGE_GROUPSHARED))
+    {
+        hlsl_init_simple_deref_from_var(&dst_deref, hlsl_ir_load(lhs)->src.var);
+        coords = hlsl_block_add_uint_constant(ctx, params->instrs, 0, loc);
+    }
     else
     {
-        hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, "Interlocked targets must be UAV elements.");
+        hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE,
+                "Interlocked targets must be UAV or groupshared elements.");
         return false;
     }
 

libs/vkd3d-shader/hlsl_codegen.c

@@ -3537,10 +3537,10 @@ static bool validate_dereferences(struct hlsl_ctx *ctx, struct hlsl_ir_node *ins
         {
             struct hlsl_ir_interlocked *interlocked = hlsl_ir_interlocked(instr);
 
-            if (!interlocked->dst.var->is_uniform)
+            if (!interlocked->dst.var->is_uniform && !interlocked->dst.var->is_tgsm)
             {
                 hlsl_error(ctx, &instr->loc, VKD3D_SHADER_ERROR_HLSL_NON_STATIC_OBJECT_REF,
-                        "Accessed resource must have a single uniform source.");
+                        "Accessed resource must have a single uniform or groupshared source.");
             }
             else if (validate_component_index_range_from_deref(ctx, &interlocked->dst) == DEREF_VALIDATION_NOT_CONSTANT)
             {

tests/hlsl/tgsm.shader_test

@@ -82,7 +82,7 @@ size (raw_buffer, 4)
 
 1 0 0 0
 
-[compute shader todo]
+[compute shader]
 RWByteAddressBuffer u : register(u1);
 groupshared uint m;
 
@@ -99,7 +99,7 @@ void main(uint local_idx : SV_GroupIndex, uint group_id : SV_GroupID)
 }
 
 [test]
-todo(sm<6 | msl) dispatch 4 1 1
+todo(glsl | msl) dispatch 4 1 1
 probe uav 1 (0) u32(0)
 probe uav 1 (1) u32(33)
 probe uav 1 (2) u32(66)
@@ -112,7 +112,7 @@ size (raw_buffer, 4)
 
 1 0 0 0
 
-[compute shader todo]
+[compute shader]
 RWByteAddressBuffer u : register(u1);
 groupshared int m;
 
@@ -129,7 +129,7 @@ void main(uint local_idx : SV_GroupIndex, uint group_id : SV_GroupID)
 }
 
 [test]
-todo(sm<6 | msl) dispatch 4 1 1
+todo(glsl | msl) dispatch 4 1 1
 probe uav 1 (0) i32(0)
 probe uav 1 (1) i32(-31)
 probe uav 1 (2) i32(-62)
@@ -216,7 +216,7 @@ size (raw_buffer, 1)
 
 0
 
-[compute shader todo]
+[compute shader]
 RWByteAddressBuffer u : register(u1);
 groupshared uint m;
 
@@ -234,7 +234,7 @@ void main(uint local_idx : SV_GroupIndex)
 }
 
 [test]
-todo(sm<6 | msl) dispatch 1 1 1
+todo(glsl | msl) dispatch 1 1 1
 probe uav 1 (0) u32(39)
 
 [uav 0]