wined3d: Handle raw thread group shared memory declarations.

dlls/wined3d/glsl_shader.c

@@ -6833,6 +6833,7 @@ static GLuint shader_glsl_generate_compute_shader(const struct wined3d_context *
     const struct wined3d_gl_info *gl_info = context->gl_info;
     struct shader_glsl_ctx_priv priv_ctx;
     GLuint shader_id;
+    unsigned int i;
 
     shader_id = GL_EXTCALL(glCreateShader(GL_COMPUTE_SHADER));
 
@@ -6846,6 +6847,12 @@ static GLuint shader_glsl_generate_compute_shader(const struct wined3d_context *
     priv_ctx.string_buffers = string_buffers;
     shader_generate_glsl_declarations(context, buffer, shader, reg_maps, &priv_ctx);
 
+    for (i = 0; i < reg_maps->tgsm_count; ++i)
+    {
+        if (reg_maps->tgsm[i].size)
+            shader_addline(buffer, "shared uint cs_g%u[%u];\n", i, reg_maps->tgsm[i].size);
+    }
+
     shader_addline(buffer, "layout(local_size_x = %u, local_size_y = %u, local_size_z = %u) in;\n",
             thread_group_size->x, thread_group_size->y, thread_group_size->z);
 
@@ -9454,7 +9461,7 @@ static const SHADER_HANDLER shader_glsl_instruction_handler_table[WINED3DSIH_TAB
     /* WINED3DSIH_DCL_TESSELLATOR_DOMAIN           */ NULL,
     /* WINED3DSIH_DCL_TESSELLATOR_OUTPUT_PRIMITIVE */ NULL,
     /* WINED3DSIH_DCL_TESSELLATOR_PARTITIONING     */ NULL,
-    /* WINED3DSIH_DCL_TGSM_RAW                     */ NULL,
+    /* WINED3DSIH_DCL_TGSM_RAW                     */ shader_glsl_nop,
     /* WINED3DSIH_DCL_TGSM_STRUCTURED              */ NULL,
     /* WINED3DSIH_DCL_THREAD_GROUP                 */ shader_glsl_nop,
     /* WINED3DSIH_DCL_UAV_RAW                      */ shader_glsl_nop,

dlls/wined3d/shader.c

@@ -1047,6 +1047,31 @@ static HRESULT shader_get_registers_used(struct wined3d_shader *shader, const st
         {
             reg_maps->temporary_count = ins.declaration.count;
         }
+        else if (ins.handler_idx == WINED3DSIH_DCL_TGSM_RAW)
+        {
+            unsigned int reg_idx = ins.declaration.tgsm_raw.reg.reg.idx[0].offset;
+            if (reg_idx >= MAX_TGSM_REGISTERS)
+            {
+                ERR("Invalid TGSM register index %u.\n", reg_idx);
+                break;
+            }
+            if (shader_version.type == WINED3D_SHADER_TYPE_COMPUTE)
+            {
+                struct wined3d_shader_tgsm *tgsm;
+
+                if (!wined3d_array_reserve((void **)&reg_maps->tgsm, &reg_maps->tgsm_capacity,
+                        reg_idx + 1, sizeof(*reg_maps->tgsm)))
+                    return E_OUTOFMEMORY;
+                reg_maps->tgsm_count = reg_idx + 1;
+                tgsm = &reg_maps->tgsm[reg_idx];
+                tgsm->size = ins.declaration.tgsm_raw.byte_count / 4;
+            }
+            else
+            {
+                FIXME("Invalid instruction %#x for shader type %#x.\n",
+                        ins.handler_idx, shader_version.type);
+            }
+        }
         else if (ins.handler_idx == WINED3DSIH_DCL_THREAD_GROUP)
         {
             if (shader_version.type == WINED3D_SHADER_TYPE_COMPUTE)
@@ -1522,6 +1547,8 @@ static void shader_cleanup_reg_maps(struct wined3d_shader_reg_maps *reg_maps)
     LIST_FOR_EACH_ENTRY_SAFE(reg, reg_next, &reg_maps->indexable_temps, struct wined3d_shader_indexable_temp, entry)
         HeapFree(GetProcessHeap(), 0, reg);
     list_init(&reg_maps->indexable_temps);
+
+    HeapFree(GetProcessHeap(), 0, reg_maps->tgsm);
 }
 
 unsigned int shader_find_free_input_register(const struct wined3d_shader_reg_maps *reg_maps, unsigned int max)

dlls/wined3d/shader_sm4.c

@@ -763,6 +763,8 @@ static void shader_sm5_read_dcl_tgsm_raw(struct wined3d_shader_instruction *ins,
 {
     shader_sm4_read_dst_param(priv, &tokens, WINED3D_DATA_FLOAT, &ins->declaration.tgsm_raw.reg);
     ins->declaration.tgsm_raw.byte_count = *tokens;
+    if (ins->declaration.tgsm_raw.byte_count % 4)
+        FIXME("Byte count %u is not multiple of 4.\n", ins->declaration.tgsm_raw.byte_count);
 }
 
 static void shader_sm5_read_dcl_tgsm_structured(struct wined3d_shader_instruction *ins,

dlls/wined3d/wined3d_private.h

@@ -185,6 +185,7 @@ static inline enum complex_fixup get_complex_fixup(struct color_fixup_desc fixup
 #define MAX_SAMPLER_OBJECTS         16
 #define MAX_SHADER_RESOURCE_VIEWS   128
 #define MAX_UNORDERED_ACCESS_VIEWS  8
+#define MAX_TGSM_REGISTERS          8192
 #define MAX_VERTEX_BLENDS           4
 #define MAX_MULTISAMPLE_TYPES       8
 
@@ -915,6 +916,10 @@ struct wined3d_shader_reg_maps
     /* Whether or not loops are used in this shader, and nesting depth */
     unsigned loop_depth;
     UINT min_rel_offset, max_rel_offset;
+
+    struct wined3d_shader_tgsm *tgsm;
+    SIZE_T tgsm_capacity;
+    unsigned int tgsm_count;
 };
 
 /* Keeps track of details for TEX_M#x# instructions which need to maintain
@@ -998,6 +1003,11 @@ struct wined3d_shader_structured_resource
     unsigned int byte_stride;
 };
 
+struct wined3d_shader_tgsm
+{
+    unsigned int size;
+};
+
 struct wined3d_shader_tgsm_raw
 {
     struct wined3d_shader_dst_param reg;