vkd3d-shader/hlsl: Add support for sample index argument in Load().

Approved-by: Giovanni Mascellani (@giomasce)
Approved-by: Zebediah Figura (@zfigura)
Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Alexandre Julliard (@julliard)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/135

libs/vkd3d-shader/hlsl.c

@@ -1351,6 +1351,7 @@ struct hlsl_ir_node *hlsl_new_resource_load(struct hlsl_ctx *ctx,
     }
 
     hlsl_src_from_node(&load->coords, params->coords);
+    hlsl_src_from_node(&load->sample_index, params->sample_index);
     hlsl_src_from_node(&load->texel_offset, params->texel_offset);
     hlsl_src_from_node(&load->lod, params->lod);
     load->sampling_dim = params->sampling_dim;
@@ -1643,6 +1644,7 @@ static struct hlsl_ir_node *clone_resource_load(struct hlsl_ctx *ctx,
     }
     clone_src(map, &dst->coords, &src->coords);
     clone_src(map, &dst->lod, &src->lod);
+    clone_src(map, &dst->sample_index, &src->sample_index);
     clone_src(map, &dst->texel_offset, &src->texel_offset);
     dst->sampling_dim = src->sampling_dim;
     return &dst->node;
@@ -2439,6 +2441,11 @@ static void dump_ir_resource_load(struct vkd3d_string_buffer *buffer, const stru
     dump_deref(buffer, &load->sampler);
     vkd3d_string_buffer_printf(buffer, ", coords = ");
     dump_src(buffer, &load->coords);
+    if (load->sample_index.node)
+    {
+        vkd3d_string_buffer_printf(buffer, ", sample index = ");
+        dump_src(buffer, &load->sample_index);
+    }
     if (load->texel_offset.node)
     {
         vkd3d_string_buffer_printf(buffer, ", offset = ");
@@ -2685,6 +2692,7 @@ static void free_ir_resource_load(struct hlsl_ir_resource_load *load)
     hlsl_src_remove(&load->coords);
     hlsl_src_remove(&load->lod);
     hlsl_src_remove(&load->texel_offset);
+    hlsl_src_remove(&load->sample_index);
     vkd3d_free(load);
 }
 

libs/vkd3d-shader/hlsl.h

@@ -623,7 +623,7 @@ struct hlsl_ir_resource_load
     struct hlsl_ir_node node;
     enum hlsl_resource_load_type load_type;
     struct hlsl_deref resource, sampler;
-    struct hlsl_src coords, lod, texel_offset;
+    struct hlsl_src coords, lod, sample_index, texel_offset;
     enum hlsl_sampler_dim sampling_dim;
 };
 
@@ -825,7 +825,7 @@ struct hlsl_resource_load_params
     struct hlsl_type *format;
     enum hlsl_resource_load_type type;
     struct hlsl_ir_node *resource, *sampler;
-    struct hlsl_ir_node *coords, *lod, *texel_offset;
+    struct hlsl_ir_node *coords, *lod, *sample_index, *texel_offset;
     enum hlsl_sampler_dim sampling_dim;
 };
 

libs/vkd3d-shader/hlsl.y

@@ -3743,7 +3743,9 @@ static bool add_load_method_call(struct hlsl_ctx *ctx, struct list *instrs, stru
     }
     if (multisampled)
     {
-        hlsl_fixme(ctx, loc, "Load() sampling index parameter.");
+        if (!(load_params.sample_index = add_implicit_conversion(ctx, instrs, params->args[1],
+                hlsl_get_scalar_type(ctx, HLSL_TYPE_INT), loc)))
+            return false;
     }
 
     assert(offset_dim);
@@ -3758,9 +3760,9 @@ static bool add_load_method_call(struct hlsl_ctx *ctx, struct list *instrs, stru
         hlsl_fixme(ctx, loc, "Tiled resource status argument.");
     }
 
-    /* +1 for the mipmap level */
+    /* +1 for the mipmap level for non-multisampled textures */
     if (!(load_params.coords = add_implicit_conversion(ctx, instrs, params->args[0],
-            hlsl_get_vector_type(ctx, HLSL_TYPE_INT, sampler_dim + 1), loc)))
+            hlsl_get_vector_type(ctx, HLSL_TYPE_INT, sampler_dim + !multisampled), loc)))
         return false;
 
     load_params.format = object_type->e.resource_format;

libs/vkd3d-shader/hlsl_codegen.c

@@ -2674,6 +2674,8 @@ static void compute_liveness_recurse(struct hlsl_block *block, unsigned int loop
                 load->texel_offset.node->last_read = instr->index;
             if (load->lod.node)
                 load->lod.node->last_read = instr->index;
+            if (load->sample_index.node)
+                load->sample_index.node->last_read = instr->index;
             break;
         }
         case HLSL_IR_RESOURCE_STORE:

libs/vkd3d-shader/tpf.c

@@ -2809,6 +2809,8 @@ static D3D_SHADER_VARIABLE_TYPE sm4_base_type(const struct hlsl_type *type)
                     return D3D_SVT_TEXTURE1D;
                 case HLSL_SAMPLER_DIM_2D:
                     return D3D_SVT_TEXTURE2D;
+                case HLSL_SAMPLER_DIM_2DMS:
+                    return D3D_SVT_TEXTURE2DMS;
                 case HLSL_SAMPLER_DIM_3D:
                     return D3D_SVT_TEXTURE3D;
                 case HLSL_SAMPLER_DIM_CUBE:
@@ -3984,14 +3986,20 @@ static void write_sm4_constant(struct hlsl_ctx *ctx,
 static void write_sm4_ld(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer,
         const struct hlsl_type *resource_type, const struct hlsl_ir_node *dst,
         const struct hlsl_deref *resource, const struct hlsl_ir_node *coords,
-        const struct hlsl_ir_node *texel_offset, enum hlsl_sampler_dim dim)
+        const struct hlsl_ir_node *sample_index, const struct hlsl_ir_node *texel_offset,
+        enum hlsl_sampler_dim dim)
 {
+    bool multisampled = resource_type->base_type == HLSL_TYPE_TEXTURE
+            && (resource_type->sampler_dim == HLSL_SAMPLER_DIM_2DMS || resource_type->sampler_dim == HLSL_SAMPLER_DIM_2DMSARRAY);
     bool uav = (hlsl_type_get_regset(resource_type) == HLSL_REGSET_UAVS);
     struct sm4_instruction instr;
     unsigned int dim_count;
 
     memset(&instr, 0, sizeof(instr));
-    instr.opcode = uav ? VKD3D_SM5_OP_LD_UAV_TYPED : VKD3D_SM4_OP_LD;
+    if (uav)
+        instr.opcode = VKD3D_SM5_OP_LD_UAV_TYPED;
+    else
+        instr.opcode = multisampled ? VKD3D_SM4_OP_LD2DMS : VKD3D_SM4_OP_LD;
 
     if (texel_offset)
     {
@@ -4023,6 +4031,33 @@ static void write_sm4_ld(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buf
 
     instr.src_count = 2;
 
+    if (multisampled)
+    {
+        if (sample_index->type == HLSL_IR_CONSTANT)
+        {
+            struct sm4_register *reg = &instr.srcs[2].reg;
+            struct hlsl_ir_constant *index;
+
+            index = hlsl_ir_constant(sample_index);
+
+            memset(&instr.srcs[2], 0, sizeof(instr.srcs[2]));
+            instr.srcs[2].swizzle_type = VKD3D_SM4_SWIZZLE_NONE;
+            reg->type = VKD3D_SM4_RT_IMMCONST;
+            reg->dim = VKD3D_SM4_DIMENSION_SCALAR;
+            reg->immconst_uint[0] = index->value.u[0].u;
+        }
+        else if (ctx->profile->major_version == 4 && ctx->profile->minor_version == 0)
+        {
+            hlsl_error(ctx, &sample_index->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, "Expected literal sample index.");
+        }
+        else
+        {
+            sm4_src_from_node(&instr.srcs[2], sample_index, 0);
+        }
+
+        ++instr.src_count;
+    }
+
     write_sm4_instruction(buffer, &instr);
 }
 
@@ -4833,6 +4868,7 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx,
 {
     const struct hlsl_type *resource_type = load->resource.var->data_type;
     const struct hlsl_ir_node *texel_offset = load->texel_offset.node;
+    const struct hlsl_ir_node *sample_index = load->sample_index.node;
     const struct hlsl_ir_node *coords = load->coords.node;
 
     if (!hlsl_type_is_resource(resource_type))
@@ -4868,7 +4904,7 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx,
     {
         case HLSL_RESOURCE_LOAD:
             write_sm4_ld(ctx, buffer, resource_type, &load->node, &load->resource,
-                    coords, texel_offset, load->sampling_dim);
+                    coords, sample_index, texel_offset, load->sampling_dim);
             break;
 
         case HLSL_RESOURCE_SAMPLE:

tests/texture-load.shader_test

@@ -35,3 +35,18 @@ probe (0, 0) rgba (0.1, 0.2, 0.3, 0.4)
 probe (1, 0) rgba (0.6, 0.5, 0.2, 0.1)
 probe (0, 1) rgba (0.5, 0.7, 0.6, 0.8)
 probe (1, 1) rgba (0.8, 0.0, 0.7, 1.0)
+
+[pixel shader]
+Texture2DMS<float4, 1> t;
+
+float4 main(float4 pos : sv_position) : sv_target
+{
+    return t.Load(pos.yx, 0);
+}
+
+[test]
+draw quad
+probe (0, 0) rgba (0.1, 0.2, 0.3, 0.4)
+probe (1, 0) rgba (0.6, 0.5, 0.2, 0.1)
+probe (0, 1) rgba (0.5, 0.7, 0.6, 0.8)
+probe (1, 1) rgba (0.8, 0.0, 0.7, 1.0)