vkd3d-shader: Implement DMOV instruction.

Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>

libs/vkd3d-shader/dxbc.c

@@ -311,6 +311,7 @@ enum vkd3d_sm4_opcode
     VKD3D_SM5_OP_IMM_ATOMIC_UMIN                  = 0xbd,
     VKD3D_SM5_OP_SYNC                             = 0xbe,
     VKD3D_SM5_OP_DEQ                              = 0xc3,
+    VKD3D_SM5_OP_DMOV                             = 0xc7,
     VKD3D_SM5_OP_EVAL_SAMPLE_INDEX                = 0xcc,
     VKD3D_SM5_OP_EVAL_CENTROID                    = 0xcd,
     VKD3D_SM5_OP_DCL_GS_INSTANCES                 = 0xce,
@@ -1250,6 +1251,7 @@ static const struct vkd3d_sm4_opcode_info opcode_table[] =
     {VKD3D_SM5_OP_SYNC,                             VKD3DSIH_SYNC,                             "",     "",
             shader_sm5_read_sync},
     {VKD3D_SM5_OP_DEQ,                              VKD3DSIH_DEQ,                              "u",    "dd"},
+    {VKD3D_SM5_OP_DMOV,                             VKD3DSIH_DMOV,                             "d",    "d"},
     {VKD3D_SM5_OP_EVAL_SAMPLE_INDEX,                VKD3DSIH_EVAL_SAMPLE_INDEX,                "f",    "fi"},
     {VKD3D_SM5_OP_EVAL_CENTROID,                    VKD3DSIH_EVAL_CENTROID,                    "f",    "f"},
     {VKD3D_SM5_OP_DCL_GS_INSTANCES,                 VKD3DSIH_DCL_GS_INSTANCES,                 "",     "",
@@ -1804,6 +1806,8 @@ static bool shader_sm4_read_dst_param(struct vkd3d_sm4_data *priv, const DWORD *
     }
 
     dst_param->write_mask = (token & VKD3D_SM4_WRITEMASK_MASK) >> VKD3D_SM4_WRITEMASK_SHIFT;
+    if (data_type == VKD3D_DATA_DOUBLE)
+        dst_param->write_mask = vkd3d_write_mask_64_from_32(dst_param->write_mask);
     /* Scalar registers are declared with no write mask in shader bytecode. */
     if (!dst_param->write_mask && shader_sm4_is_scalar_register(&dst_param->reg))
         dst_param->write_mask = VKD3DSP_WRITEMASK_0;

libs/vkd3d-shader/spirv.c

@@ -3613,6 +3613,7 @@ static void vkd3d_dxbc_compiler_emit_store_reg(struct vkd3d_dxbc_compiler *compi
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     enum vkd3d_shader_component_type component_type;
     struct vkd3d_shader_register_info reg_info;
+    unsigned int src_write_mask = write_mask;
     uint32_t type_id;
 
     assert(reg->type != VKD3DSPR_IMMCONST);
@@ -3624,14 +3625,16 @@ static void vkd3d_dxbc_compiler_emit_store_reg(struct vkd3d_dxbc_compiler *compi
     component_type = vkd3d_component_type_from_data_type(reg->data_type);
     if (component_type != reg_info.component_type)
     {
-        unsigned int component_count = vkd3d_write_mask_component_count(write_mask);
+        if (reg->data_type == VKD3D_DATA_DOUBLE)
-        type_id = vkd3d_spirv_get_type_id(builder, reg_info.component_type, component_count);
+            src_write_mask = vkd3d_write_mask_32_from_64(write_mask);
+        type_id = vkd3d_spirv_get_type_id(builder, reg_info.component_type,
+                vkd3d_write_mask_component_count(src_write_mask));
         val_id = vkd3d_spirv_build_op_bitcast(builder, type_id, val_id);
         component_type = reg_info.component_type;
     }
 
     vkd3d_dxbc_compiler_emit_store(compiler,
-            reg_info.id, reg_info.write_mask, component_type, reg_info.storage_class, write_mask, val_id);
+            reg_info.id, reg_info.write_mask, component_type, reg_info.storage_class, src_write_mask, val_id);
 }
 
 static uint32_t vkd3d_dxbc_compiler_emit_sat(struct vkd3d_dxbc_compiler *compiler,
@@ -9349,6 +9352,7 @@ int vkd3d_dxbc_compiler_handle_instruction(struct vkd3d_dxbc_compiler *compiler,
         case VKD3DSIH_HS_JOIN_PHASE:
             vkd3d_dxbc_compiler_enter_shader_phase(compiler, instruction);
             break;
+        case VKD3DSIH_DMOV:
         case VKD3DSIH_MOV:
             vkd3d_dxbc_compiler_emit_mov(compiler, instruction);
             break;

libs/vkd3d-shader/trace.c

@@ -105,6 +105,7 @@ static const char * const shader_opcode_names[] =
     /* VKD3DSIH_DEFI                             */ "defi",
     /* VKD3DSIH_DEQ                              */ "deq",
     /* VKD3DSIH_DIV                              */ "div",
+    /* VKD3DSIH_DMOV                             */ "dmov",
     /* VKD3DSIH_DP2                              */ "dp2",
     /* VKD3DSIH_DP2ADD                           */ "dp2add",
     /* VKD3DSIH_DP3                              */ "dp3",
@@ -1122,6 +1123,9 @@ static void shader_dump_dst_param(struct vkd3d_d3d_asm_compiler *compiler,
     {
         static const char write_mask_chars[] = "xyzw";
 
+        if (param->reg.data_type == VKD3D_DATA_DOUBLE)
+            write_mask = vkd3d_write_mask_32_from_64(write_mask);
+
         shader_addline(buffer, ".%s", compiler->colours.write_mask);
         if (write_mask & VKD3DSP_WRITEMASK_0)
             shader_addline(buffer, "%c", write_mask_chars[0]);

libs/vkd3d-shader/vkd3d_shader_private.h

@@ -193,6 +193,7 @@ enum vkd3d_shader_opcode
     VKD3DSIH_DEFI,
     VKD3DSIH_DEQ,
     VKD3DSIH_DIV,
+    VKD3DSIH_DMOV,
     VKD3DSIH_DP2,
     VKD3DSIH_DP2ADD,
     VKD3DSIH_DP3,
@@ -1064,6 +1065,12 @@ static inline unsigned int vkd3d_write_mask_from_component_count(unsigned int co
     return (VKD3DSP_WRITEMASK_0 << component_count) - 1;
 }
 
+static inline unsigned int vkd3d_write_mask_64_from_32(DWORD write_mask32)
+{
+    unsigned int write_mask64 = write_mask32 | (write_mask32 >> 1);
+    return (write_mask64 & VKD3DSP_WRITEMASK_0) | ((write_mask64 & VKD3DSP_WRITEMASK_2) >> 1);
+}
+
 static inline unsigned int vkd3d_write_mask_32_from_64(unsigned int write_mask64)
 {
     unsigned int write_mask32 = (write_mask64 | (write_mask64 << 1))

tests/d3d12.c

@@ -9878,7 +9878,7 @@ static void test_shader_instructions(void)
         {&ps_movc, {{{0, 1, 1, 0}, {1, 2, 3, 4}, {5, 6, 7, 8}}}, {{5, 2, 3, 8}}},
         {&ps_movc, {{{1, 1, 1, 1}, {1, 2, 3, 4}, {5, 6, 7, 8}}}, {{1, 2, 3, 4}}},
 
-        {&ps_dmov,      {.d = {{2.5 + 1.0e-9, -3.5 - 1.0e-9}}},  {.d = {3.5 + 1.0e-9, -2.5 - 1.0e-9}}, true, true},
+        {&ps_dmov,      {.d = {{2.5 + 1.0e-9, -3.5 - 1.0e-9}}},  {.d = {3.5 + 1.0e-9, -2.5 - 1.0e-9}}, true},
         {&ps_dadd,      {.d = {{2.5, 0.0}}},  {.d = {2.5 + 1.0000002433080226, 2.5 + 2.000000481493771}}, true, true},
         {&ps_dmin_dmax, {.d = {{-1.0, 1.0}}}, {.d = {-1.0, 1.0}}, true, true},
         {&ps_dmovc,     {.d = {{0.5, 0.0}}},  {.d = {4.5, 4.5}}, true, true},