[PATCH vkd3d 7/7] vkd3d-shader/hlsl: Write SM4 image load instructions.

[Zebediah Figura]

Signed-off-by: Zebediah Figura 
---
 Makefile.am                  |  2 -
 libs/vkd3d-shader/hlsl_sm4.c | 95 +++++++++++++++++++++++++++++++++---
 2 files changed, 88 insertions(+), 9 deletions(-)

diff --git a/Makefile.am b/Makefile.am
index eaa4f8168..5a6dd6318 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -302,8 +302,6 @@ XFAIL_TESTS = \
    tests/hlsl-vector-indexing-uniform.shader_test \
    tests/math.shader_test \
    tests/max.shader_test \
-	tests/texture-load.shader_test \
-	tests/texture-load-typed.shader_test \
    tests/trigonometry.shader_test \
    tests/writemask-assignop-1.shader_test
 endif
diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c
index 9d45e1633..246aeebfe 100644
--- a/libs/vkd3d-shader/hlsl_sm4.c
+++ b/libs/vkd3d-shader/hlsl_sm4.c
@@ -763,6 +763,7 @@ static unsigned int sm4_swizzle_type(enum vkd3d_sm4_register_type type)

case VKD3D_SM4_RT_CONSTBUFFER:
         case VKD3D_SM4_RT_INPUT:
+        case VKD3D_SM4_RT_RESOURCE:
         case VKD3D_SM4_RT_TEMP:
             return VKD3D_SM4_SWIZZLE_VEC4;

@@ -779,14 +780,26 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r

if (var->is_uniform)
     {
-        unsigned int offset = hlsl_offset_from_deref(deref) + var->buffer_offset;
+        if (data_type->type == HLSL_CLASS_OBJECT && data_type->base_type == HLSL_TYPE_TEXTURE)
+        {
+            reg->type = VKD3D_SM4_RT_RESOURCE;
+            reg->dim = VKD3D_SM4_DIMENSION_VEC4;
+            reg->idx[0] = var->reg.id;
+            reg->idx_count = 1;
+            *writemask = VKD3DSP_WRITEMASK_ALL;
+        }
+        else
+        {
+            unsigned int offset = hlsl_offset_from_deref(deref) + var->buffer_offset;

-        reg->type = VKD3D_SM4_RT_CONSTBUFFER;
-        reg->dim = VKD3D_SM4_DIMENSION_VEC4;
-        reg->idx[0] = var->buffer->reg.id;
-        reg->idx[1] = offset / 4;
-        reg->idx_count = 2;
-        *writemask = ((1u << data_type->dimx) - 1) << (offset & 3);
+            assert(data_type->type <= HLSL_CLASS_VECTOR);
+            reg->type = VKD3D_SM4_RT_CONSTBUFFER;
+            reg->dim = VKD3D_SM4_DIMENSION_VEC4;
+            reg->idx[0] = var->buffer->reg.id;
+            reg->idx[1] = offset / 4;
+            reg->idx_count = 2;
+            *writemask = ((1u << data_type->dimx) - 1) << (offset & 3);
+        }
     }
     else if (var->is_input_semantic)
     {
@@ -1154,6 +1167,50 @@ static void write_sm4_constant(struct hlsl_ctx *ctx,
     write_sm4_instruction(buffer, &instr);
 }

+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)
+{
+    struct sm4_instruction instr;
+    unsigned int writemask;
+
+    memset(&instr, 0, sizeof(instr));
+    instr.opcode = VKD3D_SM4_OP_LD;
+
+    sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, dst);
+    instr.dst_count = 1;
+
+    sm4_register_from_node(&instr.srcs[0].reg, &writemask, coords);
+    instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask);
+
+    /* Mipmap level is in the last component in the IR, but needs to be in the W
+     * component in the instruction. */
+    switch (resource_type->sampler_dim)
+    {
+        case HLSL_SAMPLER_DIM_1D:
+            instr.srcs[0].swizzle = hlsl_combine_swizzles(instr.srcs[0].swizzle, HLSL_SWIZZLE(X, X, X, Y), 4);
+            break;
+
+        case HLSL_SAMPLER_DIM_2D:
+            instr.srcs[0].swizzle = hlsl_combine_swizzles(instr.srcs[0].swizzle, HLSL_SWIZZLE(X, Y, X, Z), 4);
+            break;
+
+        case HLSL_SAMPLER_DIM_3D:
+        case HLSL_SAMPLER_DIM_CUBE:
+            break;
+
+        case HLSL_SAMPLER_DIM_GENERIC:
+            assert(0);
+    }
+
+    sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, resource, resource_type);
+    instr.srcs[1].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask);
+
+    instr.src_count = 2;
+
+    write_sm4_instruction(buffer, &instr);
+}
+
 static void write_sm4_expr(struct hlsl_ctx *ctx,
         struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_expr *expr)
 {
@@ -1375,6 +1432,26 @@ static void write_sm4_load(struct hlsl_ctx *ctx,
     write_sm4_instruction(buffer, &instr);
 }

+static void write_sm4_resource_load(struct hlsl_ctx *ctx,
+        struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_resource_load *load)
+{
+    const struct hlsl_type *resource_type = load->resource.var->data_type;
+    const struct hlsl_ir_node *coords = load->coords.node;
+
+    if (!load->resource.var->is_uniform)
+    {
+        hlsl_fixme(ctx, load->node.loc, "Load from non-uniform resource variable.");
+        return;
+    }
+
+    switch (load->load_type)
+    {
+        case HLSL_RESOURCE_LOAD:
+            write_sm4_ld(ctx, buffer, resource_type, &load->node, &load->resource, coords);
+            break;
+    }
+}
+
 static void write_sm4_store(struct hlsl_ctx *ctx,
         struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_store *store)
 {
@@ -1504,6 +1581,10 @@ static void write_sm4_shdr(struct hlsl_ctx *ctx,
                 write_sm4_load(ctx, &buffer, hlsl_ir_load(instr));
                 break;

+            case HLSL_IR_RESOURCE_LOAD:
+                write_sm4_resource_load(ctx, &buffer, hlsl_ir_resource_load(instr));
+                break;
+
             case HLSL_IR_STORE:
                 write_sm4_store(ctx, &buffer, hlsl_ir_store(instr));
                 break;