wine-staging/patches/vkd3d-latest/0001-Updated-vkd3d-to-3bafee344d8d50172d8da18512ba070c682.patch

From bff92c3733134be7c7318f1d5b0d492f36951d26 Mon Sep 17 00:00:00 2001
From: Alistair Leslie-Hughes <leslie_alistair@hotmail.com>
Date: Fri, 8 Dec 2023 13:21:19 +1100
Subject: [PATCH] Updated vkd3d to 3bafee344d8d50172d8da18512ba070c6826b18d.

---
 libs/vkd3d/include/private/vkd3d_common.h     |   4 +-
 libs/vkd3d/include/vkd3d_shader.h             |  20 +
 libs/vkd3d/libs/vkd3d-common/blob.c           |   1 +
 libs/vkd3d/libs/vkd3d-shader/d3d_asm.c        | 135 ++--
 libs/vkd3d/libs/vkd3d-shader/d3dbc.c          |  16 +-
 libs/vkd3d/libs/vkd3d-shader/dxbc.c           |   4 -
 libs/vkd3d/libs/vkd3d-shader/dxil.c           | 699 +++++++++++++++---
 libs/vkd3d/libs/vkd3d-shader/hlsl.y           |  20 +-
 libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c   |  12 +-
 libs/vkd3d/libs/vkd3d-shader/ir.c             |  29 +-
 libs/vkd3d/libs/vkd3d-shader/spirv.c          | 669 ++++++++++-------
 libs/vkd3d/libs/vkd3d-shader/tpf.c            |  82 +-
 .../libs/vkd3d-shader/vkd3d_shader_main.c     |  86 ++-
 .../libs/vkd3d-shader/vkd3d_shader_private.h  |  54 +-
 libs/vkd3d/libs/vkd3d/command.c               |   4 +
 libs/vkd3d/libs/vkd3d/device.c                | 114 ++-
 libs/vkd3d/libs/vkd3d/resource.c              |  19 +-
 libs/vkd3d/libs/vkd3d/state.c                 |  27 +-
 libs/vkd3d/libs/vkd3d/utils.c                 |   2 +
 libs/vkd3d/libs/vkd3d/vkd3d_private.h         |  23 +-
 20 files changed, 1475 insertions(+), 545 deletions(-)

diff --git a/libs/vkd3d/include/private/vkd3d_common.h b/libs/vkd3d/include/private/vkd3d_common.h
index 34fde1a2aa0..e9bff2fbba3 100644
--- a/libs/vkd3d/include/private/vkd3d_common.h
+++ b/libs/vkd3d/include/private/vkd3d_common.h
@@ -80,7 +80,7 @@ static inline size_t align(size_t addr, size_t alignment)
 # ifdef __MINGW_PRINTF_FORMAT
 #  define VKD3D_PRINTF_FUNC(fmt, args) __attribute__((format(__MINGW_PRINTF_FORMAT, fmt, args)))
 # else
-#  define VKD3D_PRINTF_FUNC(fmt, args) /* __attribute__((format(printf, fmt, args))) */
+#  define VKD3D_PRINTF_FUNC(fmt, args) __attribute__((format(printf, fmt, args)))
 # endif
 # define VKD3D_UNUSED __attribute__((unused))
 # define VKD3D_UNREACHABLE __builtin_unreachable()
@@ -107,7 +107,7 @@ static inline unsigned int vkd3d_popcount(unsigned int v)
 {
 #ifdef _MSC_VER
     return __popcnt(v);
-#elif defined(__MINGW32__)
+#elif defined(HAVE_BUILTIN_POPCOUNT)
     return __builtin_popcount(v);
 #else
     v -= (v >> 1) & 0x55555555;
diff --git a/libs/vkd3d/include/vkd3d_shader.h b/libs/vkd3d/include/vkd3d_shader.h
index 290f9085d2d..b1a1fff6451 100644
--- a/libs/vkd3d/include/vkd3d_shader.h
+++ b/libs/vkd3d/include/vkd3d_shader.h
@@ -196,6 +196,14 @@ enum vkd3d_shader_compile_option_fragment_coordinate_origin
     VKD3D_FORCE_32_BIT_ENUM(VKD3D_SHADER_COMPILE_OPTION_FRAGMENT_COORDINATE_ORIGIN),
 };
 
+/** Advertises feature availability. \since 1.11 */
+enum vkd3d_shader_compile_option_feature_flags
+{
+    VKD3D_SHADER_COMPILE_OPTION_FEATURE_INT64         = 0x00000001,
+
+    VKD3D_FORCE_32_BIT_ENUM(VKD3D_SHADER_COMPILE_OPTION_FEATURE_FLAGS),
+};
+
 enum vkd3d_shader_compile_option_name
 {
     /**
@@ -253,6 +261,16 @@ enum vkd3d_shader_compile_option_name
      * \since 1.10
      */
     VKD3D_SHADER_COMPILE_OPTION_FRAGMENT_COORDINATE_ORIGIN = 0x00000009,
+    /**
+     * This option specifies the shader features available in the target
+     * environment. These are not extensions, i.e. they are always supported
+     * by the driver, but may not be supported by the available hardware.
+     *
+     * \a value is a member of enum vkd3d_shader_compile_option_feature_flags.
+     *
+     * \since 1.11
+     */
+    VKD3D_SHADER_COMPILE_OPTION_FEATURE = 0x0000000a,
 
     VKD3D_FORCE_32_BIT_ENUM(VKD3D_SHADER_COMPILE_OPTION_NAME),
 };
@@ -1551,6 +1569,8 @@ enum vkd3d_shader_component_type
     VKD3D_SHADER_COMPONENT_BOOL     = 0x4,
     /** 64-bit IEEE floating-point. */
     VKD3D_SHADER_COMPONENT_DOUBLE   = 0x5,
+    /** 64-bit unsigned integer. \since 1.11 */
+    VKD3D_SHADER_COMPONENT_UINT64   = 0x6,
 
     VKD3D_FORCE_32_BIT_ENUM(VKD3D_SHADER_COMPONENT_TYPE),
 };
diff --git a/libs/vkd3d/libs/vkd3d-common/blob.c b/libs/vkd3d/libs/vkd3d-common/blob.c
index 71ab9a2c51e..fa2812619ac 100644
--- a/libs/vkd3d/libs/vkd3d-common/blob.c
+++ b/libs/vkd3d/libs/vkd3d-common/blob.c
@@ -17,6 +17,7 @@
  */
 
 #define COBJMACROS
+
 #define CONST_VTABLE
 #include "vkd3d.h"
 #include "vkd3d_blob.h"
diff --git a/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c
index 82d1d71d9d3..0589bc42174 100644
--- a/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c
+++ b/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c
@@ -364,6 +364,7 @@ struct vkd3d_d3d_asm_compiler
     struct vkd3d_string_buffer buffer;
     struct vkd3d_shader_version shader_version;
     struct vkd3d_d3d_asm_colours colours;
+    enum vsir_asm_dialect dialect;
 };
 
 static int VKD3D_PRINTF_FUNC(2, 3) shader_addline(struct vkd3d_string_buffer *buffer, const char *format, ...)
@@ -606,7 +607,7 @@ static void shader_dump_resource_type(struct vkd3d_d3d_asm_compiler *compiler, e
         vkd3d_string_buffer_printf(&compiler->buffer, "unknown");
 }
 
-static void shader_dump_data_type(struct vkd3d_d3d_asm_compiler *compiler, const enum vkd3d_data_type *type)
+static void shader_dump_data_type(struct vkd3d_d3d_asm_compiler *compiler, enum vkd3d_data_type type)
 {
     static const char *const data_type_names[] =
     {
@@ -623,19 +624,31 @@ static void shader_dump_data_type(struct vkd3d_d3d_asm_compiler *compiler, const
         [VKD3D_DATA_DOUBLE   ] = "double",
         [VKD3D_DATA_CONTINUED] = "<continued>",
         [VKD3D_DATA_UNUSED   ] = "<unused>",
+        [VKD3D_DATA_UINT8    ] = "uint8",
+        [VKD3D_DATA_UINT64   ] = "uint64",
+        [VKD3D_DATA_BOOL     ] = "bool",
     };
+
     const char *name;
+
+    if (type < ARRAY_SIZE(data_type_names))
+        name = data_type_names[type];
+    else
+        name = "<unknown>";
+
+    vkd3d_string_buffer_printf(&compiler->buffer, "%s", name);
+}
+
+static void shader_dump_resource_data_type(struct vkd3d_d3d_asm_compiler *compiler, const enum vkd3d_data_type *type)
+{
     int i;
 
     vkd3d_string_buffer_printf(&compiler->buffer, "(");
 
     for (i = 0; i < 4; i++)
     {
-        if (type[i] < ARRAY_SIZE(data_type_names))
-            name = data_type_names[type[i]];
-        else
-            name = "unknown";
-        vkd3d_string_buffer_printf(&compiler->buffer, "%s%s", i == 0 ? "" : ",", name);
+        vkd3d_string_buffer_printf(&compiler->buffer, "%s", i == 0 ? "" : ",");
+        shader_dump_data_type(compiler, type[i]);
     }
 
     vkd3d_string_buffer_printf(&compiler->buffer, ")");
@@ -682,7 +695,7 @@ static void shader_dump_decl_usage(struct vkd3d_d3d_asm_compiler *compiler,
         if (semantic->resource.reg.reg.type == VKD3DSPR_UAV)
             shader_dump_uav_flags(compiler, flags);
         shader_addline(buffer, " ");
-        shader_dump_data_type(compiler, semantic->resource_data_type);
+        shader_dump_resource_data_type(compiler, semantic->resource_data_type);
     }
     else
     {
@@ -1096,15 +1109,15 @@ static void shader_dump_register(struct vkd3d_d3d_asm_compiler *compiler, const
                 switch (reg->data_type)
                 {
                     case VKD3D_DATA_FLOAT:
-                        shader_print_float_literal(compiler, "", reg->u.immconst_float[0], "");
+                        shader_print_float_literal(compiler, "", reg->u.immconst_f32[0], "");
                         break;
                     case VKD3D_DATA_INT:
-                        shader_print_int_literal(compiler, "", reg->u.immconst_uint[0], "");
+                        shader_print_int_literal(compiler, "", reg->u.immconst_u32[0], "");
                         break;
                     case VKD3D_DATA_RESOURCE:
                     case VKD3D_DATA_SAMPLER:
                     case VKD3D_DATA_UINT:
-                        shader_print_uint_literal(compiler, "", reg->u.immconst_uint[0], "");
+                        shader_print_uint_literal(compiler, "", reg->u.immconst_u32[0], "");
                         break;
                     default:
                         shader_addline(buffer, "<unhandled data type %#x>", reg->data_type);
@@ -1116,24 +1129,24 @@ static void shader_dump_register(struct vkd3d_d3d_asm_compiler *compiler, const
                 switch (reg->data_type)
                 {
                     case VKD3D_DATA_FLOAT:
-                        shader_print_float_literal(compiler, "", reg->u.immconst_float[0], "");
-                        shader_print_float_literal(compiler, ", ", reg->u.immconst_float[1], "");
-                        shader_print_float_literal(compiler, ", ", reg->u.immconst_float[2], "");
-                        shader_print_float_literal(compiler, ", ", reg->u.immconst_float[3], "");
+                        shader_print_float_literal(compiler, "", reg->u.immconst_f32[0], "");
+                        shader_print_float_literal(compiler, ", ", reg->u.immconst_f32[1], "");
+                        shader_print_float_literal(compiler, ", ", reg->u.immconst_f32[2], "");
+                        shader_print_float_literal(compiler, ", ", reg->u.immconst_f32[3], "");
                         break;
                     case VKD3D_DATA_INT:
-                        shader_print_int_literal(compiler, "", reg->u.immconst_uint[0], "");
-                        shader_print_int_literal(compiler, ", ", reg->u.immconst_uint[1], "");
-                        shader_print_int_literal(compiler, ", ", reg->u.immconst_uint[2], "");
-                        shader_print_int_literal(compiler, ", ", reg->u.immconst_uint[3], "");
+                        shader_print_int_literal(compiler, "", reg->u.immconst_u32[0], "");
+                        shader_print_int_literal(compiler, ", ", reg->u.immconst_u32[1], "");
+                        shader_print_int_literal(compiler, ", ", reg->u.immconst_u32[2], "");
+                        shader_print_int_literal(compiler, ", ", reg->u.immconst_u32[3], "");
                         break;
                     case VKD3D_DATA_RESOURCE:
                     case VKD3D_DATA_SAMPLER:
                     case VKD3D_DATA_UINT:
-                        shader_print_uint_literal(compiler, "", reg->u.immconst_uint[0], "");
-                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_uint[1], "");
-                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_uint[2], "");
-                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_uint[3], "");
+                        shader_print_uint_literal(compiler, "", reg->u.immconst_u32[0], "");
+                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_u32[1], "");
+                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_u32[2], "");
+                        shader_print_uint_literal(compiler, ", ", reg->u.immconst_u32[3], "");
                         break;
                     default:
                         shader_addline(buffer, "<unhandled data type %#x>", reg->data_type);
@@ -1155,9 +1168,9 @@ static void shader_dump_register(struct vkd3d_d3d_asm_compiler *compiler, const
         {
             if (reg->data_type == VKD3D_DATA_DOUBLE)
             {
-                shader_print_double_literal(compiler, "", reg->u.immconst_double[0], "");
+                shader_print_double_literal(compiler, "", reg->u.immconst_f64[0], "");
                 if (reg->dimension == VSIR_DIMENSION_VEC4)
-                    shader_print_double_literal(compiler, ", ", reg->u.immconst_double[1], "");
+                    shader_print_double_literal(compiler, ", ", reg->u.immconst_f64[1], "");
             }
             else
             {
@@ -1265,6 +1278,32 @@ static void shader_print_non_uniform(struct vkd3d_d3d_asm_compiler *compiler, co
                 compiler->colours.modifier, compiler->colours.reset);
 }
 
+static void shader_dump_reg_type(struct vkd3d_d3d_asm_compiler *compiler,
+        const struct vkd3d_shader_register *reg)
+{
+    static const char *dimensions[] =
+    {
+        [VSIR_DIMENSION_NONE]   = "",
+        [VSIR_DIMENSION_SCALAR] = "s:",
+        [VSIR_DIMENSION_VEC4]   = "v4:",
+    };
+
+    struct vkd3d_string_buffer *buffer = &compiler->buffer;
+    const char *dimension;
+
+    if (compiler->dialect != VSIR_ASM_VSIR)
+        return;
+
+    if (reg->dimension < ARRAY_SIZE(dimensions))
+        dimension = dimensions[reg->dimension];
+    else
+        dimension = "??";
+
+    shader_addline(buffer, " <%s", dimension);
+    shader_dump_data_type(compiler, reg->data_type);
+    shader_addline(buffer, ">");
+}
+
 static void shader_dump_dst_param(struct vkd3d_d3d_asm_compiler *compiler,
         const struct vkd3d_shader_dst_param *param, bool is_declaration)
 {
@@ -1278,7 +1317,7 @@ 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);
+            write_mask = vsir_write_mask_32_from_64(write_mask);
 
         shader_addline(buffer, ".%s", compiler->colours.write_mask);
         if (write_mask & VKD3DSP_WRITEMASK_0)
@@ -1294,6 +1333,7 @@ static void shader_dump_dst_param(struct vkd3d_d3d_asm_compiler *compiler,
 
     shader_print_precision(compiler, &param->reg);
     shader_print_non_uniform(compiler, &param->reg);
+    shader_dump_reg_type(compiler, &param->reg);
 }
 
 static void shader_dump_src_param(struct vkd3d_d3d_asm_compiler *compiler,
@@ -1341,10 +1381,10 @@ static void shader_dump_src_param(struct vkd3d_d3d_asm_compiler *compiler,
     if (param->reg.type != VKD3DSPR_IMMCONST && param->reg.type != VKD3DSPR_IMMCONST64
             && param->reg.dimension == VSIR_DIMENSION_VEC4)
     {
-        unsigned int swizzle_x = vkd3d_swizzle_get_component(swizzle, 0);
-        unsigned int swizzle_y = vkd3d_swizzle_get_component(swizzle, 1);
-        unsigned int swizzle_z = vkd3d_swizzle_get_component(swizzle, 2);
-        unsigned int swizzle_w = vkd3d_swizzle_get_component(swizzle, 3);
+        unsigned int swizzle_x = vsir_swizzle_get_component(swizzle, 0);
+        unsigned int swizzle_y = vsir_swizzle_get_component(swizzle, 1);
+        unsigned int swizzle_z = vsir_swizzle_get_component(swizzle, 2);
+        unsigned int swizzle_w = vsir_swizzle_get_component(swizzle, 3);
 
         static const char swizzle_chars[] = "xyzw";
 
@@ -1367,6 +1407,7 @@ static void shader_dump_src_param(struct vkd3d_d3d_asm_compiler *compiler,
 
     shader_print_precision(compiler, &param->reg);
     shader_print_non_uniform(compiler, &param->reg);
+    shader_dump_reg_type(compiler, &param->reg);
 }
 
 static void shader_dump_ins_modifiers(struct vkd3d_d3d_asm_compiler *compiler,
@@ -1577,6 +1618,12 @@ static void shader_dump_instruction_flags(struct vkd3d_d3d_asm_compiler *compile
                 shader_addline(buffer, "p");
             break;
 
+        case VKD3DSIH_ISHL:
+        case VKD3DSIH_ISHR:
+        case VKD3DSIH_USHR:
+            if (ins->flags & VKD3DSI_SHIFT_UNMASKED)
+                shader_addline(buffer, "_unmasked");
+            /* fall through */
         default:
             shader_dump_precise_flags(compiler, ins->flags);
             break;
@@ -1835,25 +1882,25 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler,
             vkd3d_string_buffer_printf(buffer, " %sc%u%s", compiler->colours.reg,
                     shader_get_float_offset(ins->dst[0].reg.type, ins->dst[0].reg.idx[0].offset),
                     compiler->colours.reset);
-            shader_print_float_literal(compiler, " = ", ins->src[0].reg.u.immconst_float[0], "");
-            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_float[1], "");
-            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_float[2], "");
-            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_float[3], "");
+            shader_print_float_literal(compiler, " = ", ins->src[0].reg.u.immconst_f32[0], "");
+            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_f32[1], "");
+            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_f32[2], "");
+            shader_print_float_literal(compiler, ", ", ins->src[0].reg.u.immconst_f32[3], "");
             break;
 
         case VKD3DSIH_DEFI:
             vkd3d_string_buffer_printf(buffer, " %si%u%s", compiler->colours.reg,
                     ins->dst[0].reg.idx[0].offset, compiler->colours.reset);
-            shader_print_int_literal(compiler, " = ", ins->src[0].reg.u.immconst_uint[0], "");
-            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_uint[1], "");
-            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_uint[2], "");
-            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_uint[3], "");
+            shader_print_int_literal(compiler, " = ", ins->src[0].reg.u.immconst_u32[0], "");
+            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_u32[1], "");
+            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_u32[2], "");
+            shader_print_int_literal(compiler, ", ", ins->src[0].reg.u.immconst_u32[3], "");
             break;
 
         case VKD3DSIH_DEFB:
             vkd3d_string_buffer_printf(buffer, " %sb%u%s", compiler->colours.reg,
                     ins->dst[0].reg.idx[0].offset, compiler->colours.reset);
-            shader_print_bool_literal(compiler, " = ", ins->src[0].reg.u.immconst_uint[0], "");
+            shader_print_bool_literal(compiler, " = ", ins->src[0].reg.u.immconst_u32[0], "");
             break;
 
         default:
@@ -1883,7 +1930,7 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler,
                     || ins->resource_data_type[1] != VKD3D_DATA_FLOAT
                     || ins->resource_data_type[2] != VKD3D_DATA_FLOAT
                     || ins->resource_data_type[3] != VKD3D_DATA_FLOAT)
-                shader_dump_data_type(compiler, ins->resource_data_type);
+                shader_dump_resource_data_type(compiler, ins->resource_data_type);
 
             for (i = 0; i < ins->dst_count; ++i)
             {
@@ -1906,10 +1953,13 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler,
 
 enum vkd3d_result vkd3d_dxbc_binary_to_text(const struct vkd3d_shader_instruction_array *instructions,
         const struct vkd3d_shader_version *shader_version, const struct vkd3d_shader_compile_info *compile_info,
-        struct vkd3d_shader_code *out)
+        struct vkd3d_shader_code *out, enum vsir_asm_dialect dialect)
 {
     enum vkd3d_shader_compile_option_formatting_flags formatting;
-    struct vkd3d_d3d_asm_compiler compiler;
+    struct vkd3d_d3d_asm_compiler compiler =
+    {
+        .dialect = dialect,
+    };
     enum vkd3d_result result = VKD3D_OK;
     struct vkd3d_string_buffer *buffer;
     unsigned int indent, i, j;
@@ -2002,6 +2052,7 @@ enum vkd3d_result vkd3d_dxbc_binary_to_text(const struct vkd3d_shader_instructio
         {
             case VKD3DSIH_ELSE:
             case VKD3DSIH_IF:
+            case VKD3DSIH_IFC:
             case VKD3DSIH_LOOP:
             case VKD3DSIH_SWITCH:
                 ++indent;
@@ -2034,7 +2085,7 @@ void vkd3d_shader_trace(const struct vkd3d_shader_instruction_array *instruction
     const char *p, *q, *end;
     struct vkd3d_shader_code code;
 
-    if (vkd3d_dxbc_binary_to_text(instructions, shader_version, NULL, &code) != VKD3D_OK)
+    if (vkd3d_dxbc_binary_to_text(instructions, shader_version, NULL, &code, VSIR_ASM_VSIR) != VKD3D_OK)
         return;
 
     end = (const char *)code.code + code.size;
diff --git a/libs/vkd3d/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d/libs/vkd3d-shader/d3dbc.c
index 3d139416b61..ca6e3b72de9 100644
--- a/libs/vkd3d/libs/vkd3d-shader/d3dbc.c
+++ b/libs/vkd3d/libs/vkd3d-shader/d3dbc.c
@@ -987,7 +987,7 @@ static void shader_sm1_read_immconst(struct vkd3d_shader_sm1_parser *sm1, const
     src_param->reg.idx[2].rel_addr = NULL;
     src_param->reg.idx_count = 0;
     src_param->reg.dimension = dimension;
-    memcpy(src_param->reg.u.immconst_uint, *ptr, count * sizeof(uint32_t));
+    memcpy(src_param->reg.u.immconst_u32, *ptr, count * sizeof(uint32_t));
     src_param->swizzle = VKD3D_SHADER_NO_SWIZZLE;
     src_param->modifiers = 0;
 
@@ -1063,12 +1063,12 @@ static void shader_sm1_validate_instruction(struct vkd3d_shader_sm1_parser *sm1,
     }
 }
 
-static unsigned int mask_from_swizzle(unsigned int swizzle)
+static unsigned int mask_from_swizzle(uint32_t swizzle)
 {
-    return (1u << vkd3d_swizzle_get_component(swizzle, 0))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 1))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 2))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 3));
+    return (1u << vsir_swizzle_get_component(swizzle, 0))
+            | (1u << vsir_swizzle_get_component(swizzle, 1))
+            | (1u << vsir_swizzle_get_component(swizzle, 2))
+            | (1u << vsir_swizzle_get_component(swizzle, 3));
 }
 
 static void shader_sm1_read_instruction(struct vkd3d_shader_sm1_parser *sm1, struct vkd3d_shader_instruction *ins)
@@ -2253,7 +2253,7 @@ static void write_sm1_jump(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *b
         {
             struct hlsl_reg *reg = &jump->condition.node->reg;
 
-            struct sm1_instruction instr =
+            struct sm1_instruction sm1_instr =
             {
                 .opcode = D3DSIO_TEXKILL,
 
@@ -2263,7 +2263,7 @@ static void write_sm1_jump(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *b
                 .has_dst = 1,
             };
 
-            write_sm1_instruction(ctx, buffer, &instr);
+            write_sm1_instruction(ctx, buffer, &sm1_instr);
             break;
         }
 
diff --git a/libs/vkd3d/libs/vkd3d-shader/dxbc.c b/libs/vkd3d/libs/vkd3d-shader/dxbc.c
index 7834c1e1615..63deaaad29a 100644
--- a/libs/vkd3d/libs/vkd3d-shader/dxbc.c
+++ b/libs/vkd3d/libs/vkd3d-shader/dxbc.c
@@ -431,10 +431,6 @@ static int shader_parse_signature(const struct vkd3d_shader_dxbc_section_desc *s
             case TAG_OSG5:
                 if (e[i].sysval_semantic == VKD3D_SHADER_SV_NONE)
                     e[i].sysval_semantic = map_fragment_output_sysval(e[i].semantic_name);
-                /* Fall through. */
-            case TAG_PCSG:
-            case TAG_PSG1:
-                e[i].used_mask = e[i].mask & ~e[i].used_mask;
                 break;
         }
 
diff --git a/libs/vkd3d/libs/vkd3d-shader/dxil.c b/libs/vkd3d/libs/vkd3d-shader/dxil.c
index beb9ae574dc..2424b176068 100644
--- a/libs/vkd3d/libs/vkd3d-shader/dxil.c
+++ b/libs/vkd3d/libs/vkd3d-shader/dxil.c
@@ -29,6 +29,9 @@ static const uint64_t MAX_ALIGNMENT_EXPONENT = 29;
 static const uint64_t GLOBALVAR_FLAG_IS_CONSTANT = 1;
 static const uint64_t GLOBALVAR_FLAG_EXPLICIT_TYPE = 2;
 static const unsigned int GLOBALVAR_ADDRESS_SPACE_SHIFT = 2;
+static const uint64_t ALLOCA_FLAG_IN_ALLOCA = 0x20;
+static const uint64_t ALLOCA_FLAG_EXPLICIT_TYPE = 0x40;
+static const uint64_t ALLOCA_ALIGNMENT_MASK = ALLOCA_FLAG_IN_ALLOCA - 1;
 static const unsigned int SHADER_DESCRIPTOR_TYPE_COUNT = 4;
 
 static const unsigned int dx_max_thread_group_size[3] = {1024, 1024, 64};
@@ -296,8 +299,28 @@ enum dx_intrinsic_opcode
 {
     DX_LOAD_INPUT                   =   4,
     DX_STORE_OUTPUT                 =   5,
+    DX_EXP                          =  21,
+    DX_FRC                          =  22,
+    DX_LOG                          =  23,
+    DX_SQRT                         =  24,
+    DX_RSQRT                        =  25,
+    DX_ROUND_NE                     =  26,
+    DX_ROUND_NI                     =  27,
+    DX_ROUND_PI                     =  28,
+    DX_ROUND_Z                      =  29,
+    DX_BFREV                        =  30,
+    DX_COUNT_BITS                   =  31,
+    DX_FIRST_BIT_LO                 =  32,
+    DX_FIRST_BIT_HI                 =  33,
+    DX_FIRST_BIT_SHI                =  34,
     DX_CREATE_HANDLE                =  57,
     DX_CBUFFER_LOAD_LEGACY          =  59,
+    DX_DERIV_COARSEX                =  83,
+    DX_DERIV_COARSEY                =  84,
+    DX_DERIV_FINEX                  =  85,
+    DX_DERIV_FINEY                  =  86,
+    DX_LEGACY_F32TOF16              = 130,
+    DX_LEGACY_F16TOF32              = 131,
 };
 
 enum dxil_cast_code
@@ -439,6 +462,7 @@ struct dxil_record
 {
     unsigned int code;
     unsigned int operand_count;
+    const struct dxil_record *attachment;
     uint64_t operands[];
 };
 
@@ -581,6 +605,7 @@ struct sm6_parser
     size_t descriptor_count;
 
     unsigned int indexable_temp_count;
+    unsigned int icb_count;
 
     struct sm6_value *values;
     size_t value_count;
@@ -796,6 +821,7 @@ static enum vkd3d_result sm6_parser_read_unabbrev_record(struct sm6_parser *sm6)
 
     record->code = code;
     record->operand_count = count;
+    record->attachment = NULL;
 
     for (i = 0; i < count; ++i)
         record->operands[i] = sm6_parser_read_vbr(sm6, 6);
@@ -1012,6 +1038,7 @@ static enum vkd3d_result sm6_parser_read_abbrev_record(struct sm6_parser *sm6, u
         if (!abbrev->operands[i + 1].read_operand(sm6, abbrev->operands[i + 1].context, &record->operands[i]))
             goto fail;
     record->operand_count = count;
+    record->attachment = NULL;
 
     /* An array can occur only as the last operand. */
     if (abbrev->is_array)
@@ -1210,7 +1237,7 @@ static size_t dxil_block_compute_module_decl_count(const struct dxil_block *bloc
     size_t i, count;
 
     for (i = 0, count = 0; i < block->record_count; ++i)
-        count += block->records[i]->code == MODULE_CODE_FUNCTION;
+        count += block->records[i]->code == MODULE_CODE_FUNCTION || block->records[i]->code == MODULE_CODE_GLOBALVAR;
     return count;
 }
 
@@ -1572,6 +1599,11 @@ static bool sm6_type_is_bool_i16_i32_i64(const struct sm6_type *type)
     return type->class == TYPE_CLASS_INTEGER && (type->u.width == 1 || type->u.width >= 16);
 }
 
+static bool sm6_type_is_i16_i32_i64(const struct sm6_type *type)
+{
+    return type->class == TYPE_CLASS_INTEGER && type->u.width >= 16;
+}
+
 static bool sm6_type_is_bool(const struct sm6_type *type)
 {
     return type->class == TYPE_CLASS_INTEGER && type->u.width == 1;
@@ -1587,6 +1619,16 @@ static inline bool sm6_type_is_i32(const struct sm6_type *type)
     return type->class == TYPE_CLASS_INTEGER && type->u.width == 32;
 }
 
+static bool sm6_type_is_float(const struct sm6_type *type)
+{
+    return type->class == TYPE_CLASS_FLOAT && type->u.width == 32;
+}
+
+static bool sm6_type_is_f16_f32(const struct sm6_type *type)
+{
+    return type->class == TYPE_CLASS_FLOAT && (type->u.width == 16 || type->u.width == 32);
+}
+
 static inline bool sm6_type_is_floating_point(const struct sm6_type *type)
 {
     return type->class == TYPE_CLASS_FLOAT;
@@ -1844,12 +1886,12 @@ static unsigned int register_get_uint_value(const struct vkd3d_shader_register *
 
     if (reg->type == VKD3DSPR_IMMCONST64)
     {
-        if (reg->u.immconst_uint64[0] > UINT_MAX)
+        if (reg->u.immconst_u64[0] > UINT_MAX)
             FIXME("Truncating 64-bit value.\n");
-        return reg->u.immconst_uint64[0];
+        return reg->u.immconst_u64[0];
     }
 
-    return reg->u.immconst_uint[0];
+    return reg->u.immconst_u32[0];
 }
 
 static uint64_t register_get_uint64_value(const struct vkd3d_shader_register *reg)
@@ -1860,7 +1902,7 @@ static uint64_t register_get_uint64_value(const struct vkd3d_shader_register *re
     if (reg->dimension == VSIR_DIMENSION_VEC4)
         WARN("Returning vec4.x.\n");
 
-    return (reg->type == VKD3DSPR_IMMCONST64) ? reg->u.immconst_uint64[0] : reg->u.immconst_uint[0];
+    return (reg->type == VKD3DSPR_IMMCONST64) ? reg->u.immconst_u64[0] : reg->u.immconst_u32[0];
 }
 
 static inline bool sm6_value_is_function_dcl(const struct sm6_value *value)
@@ -2412,7 +2454,7 @@ static enum vkd3d_result value_allocate_constant_array(struct sm6_value *dst, co
         return VKD3D_ERROR_INVALID_SHADER;
     }
     size = max(size, sizeof(icb->data[0]));
-    count = type->u.array.count * size / sizeof(icb->data[0]);
+    count = operands ? type->u.array.count * size / sizeof(icb->data[0]) : 0;
 
     if (!(icb = vkd3d_malloc(offsetof(struct vkd3d_shader_immediate_constant_buffer, data[count]))))
     {
@@ -2433,9 +2475,14 @@ static enum vkd3d_result value_allocate_constant_array(struct sm6_value *dst, co
     dst->value_type = VALUE_TYPE_ICB;
     dst->u.icb = icb;
 
+    icb->register_idx = sm6->icb_count++;
     icb->data_type = vkd3d_data_type_from_sm6_type(elem_type);
     icb->element_count = type->u.array.count;
     icb->component_count = 1;
+    icb->is_null = !operands;
+
+    if (!operands)
+        return VKD3D_OK;
 
     count = type->u.array.count;
     if (size > sizeof(icb->data[0]))
@@ -2510,12 +2557,10 @@ static enum vkd3d_result sm6_parser_constants_init(struct sm6_parser *sm6, const
         switch (record->code)
         {
             case CST_CODE_NULL:
-                if (sm6_type_is_array(type))
+                if (sm6_type_is_array(type)
+                        && (ret = value_allocate_constant_array(dst, type, NULL, sm6)) < 0)
                 {
-                    FIXME("Constant null arrays are not supported.\n");
-                    vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
-                            "Constant null arrays are not supported.");
-                    return VKD3D_ERROR_INVALID_SHADER;
+                    return ret;
                 }
                 /* For non-aggregates, register constant data is already zero-filled. */
                 break;
@@ -2532,9 +2577,9 @@ static enum vkd3d_result sm6_parser_constants_init(struct sm6_parser *sm6, const
 
                 value = decode_rotated_signed_value(record->operands[0]);
                 if (type->u.width <= 32)
-                    dst->u.reg.u.immconst_uint[0] = value & ((1ull << type->u.width) - 1);
+                    dst->u.reg.u.immconst_u32[0] = value & ((1ull << type->u.width) - 1);
                 else
-                    dst->u.reg.u.immconst_uint64[0] = value;
+                    dst->u.reg.u.immconst_u64[0] = value;
 
                 break;
 
@@ -2551,9 +2596,9 @@ static enum vkd3d_result sm6_parser_constants_init(struct sm6_parser *sm6, const
                 if (type->u.width == 16)
                     FIXME("Half float type is not supported yet.\n");
                 else if (type->u.width == 32)
-                    dst->u.reg.u.immconst_float[0] = bitcast_uint64_to_float(record->operands[0]);
+                    dst->u.reg.u.immconst_f32[0] = bitcast_uint64_to_float(record->operands[0]);
                 else if (type->u.width == 64)
-                    dst->u.reg.u.immconst_double[0] = bitcast_uint64_to_double(record->operands[0]);
+                    dst->u.reg.u.immconst_f64[0] = bitcast_uint64_to_double(record->operands[0]);
                 else
                     vkd3d_unreachable();
 
@@ -2588,6 +2633,13 @@ static enum vkd3d_result sm6_parser_constants_init(struct sm6_parser *sm6, const
                 break;
         }
 
+        if (record->attachment)
+        {
+            WARN("Ignoring metadata attachment.\n");
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "Ignoring a metadata attachment for a constant.");
+        }
+
         ++sm6->value_count;
     }
 
@@ -2626,13 +2678,28 @@ static struct vkd3d_shader_instruction *sm6_parser_add_instruction(struct sm6_pa
     return ins;
 }
 
-static void sm6_parser_declare_indexable_temp(struct sm6_parser *sm6, const struct sm6_type *elem_type,
-        unsigned int count, unsigned int alignment, unsigned int init, struct sm6_value *dst)
+static void sm6_parser_declare_icb(struct sm6_parser *sm6, const struct sm6_type *elem_type, unsigned int count,
+        unsigned int alignment, unsigned int init, struct sm6_value *dst)
 {
     enum vkd3d_data_type data_type = vkd3d_data_type_from_sm6_type(elem_type);
     struct vkd3d_shader_instruction *ins;
 
-    ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_INDEXABLE_TEMP);
+    ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_IMMEDIATE_CONSTANT_BUFFER);
+    /* The icb value index will be resolved later so forward references can be handled. */
+    ins->declaration.icb = (void *)(intptr_t)init;
+    register_init_with_id(&dst->u.reg, VKD3DSPR_IMMCONSTBUFFER, data_type, init);
+}
+
+static void sm6_parser_declare_indexable_temp(struct sm6_parser *sm6, const struct sm6_type *elem_type,
+        unsigned int count, unsigned int alignment, unsigned int init, struct vkd3d_shader_instruction *ins,
+        struct sm6_value *dst)
+{
+    enum vkd3d_data_type data_type = vkd3d_data_type_from_sm6_type(elem_type);
+
+    if (ins)
+        vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_DCL_INDEXABLE_TEMP);
+    else
+        ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_INDEXABLE_TEMP);
     ins->declaration.indexable_temp.register_idx = sm6->indexable_temp_count++;
     ins->declaration.indexable_temp.register_size = count;
     ins->declaration.indexable_temp.alignment = alignment;
@@ -2762,7 +2829,10 @@ static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_
 
     if (address_space == ADDRESS_SPACE_DEFAULT)
     {
-        sm6_parser_declare_indexable_temp(sm6, scalar_type, count, alignment, init, dst);
+        if (is_constant)
+            sm6_parser_declare_icb(sm6, scalar_type, count, alignment, init, dst);
+        else
+            sm6_parser_declare_indexable_temp(sm6, scalar_type, count, alignment, init, NULL, dst);
     }
     else if (address_space == ADDRESS_SPACE_GROUPSHARED)
     {
@@ -2790,31 +2860,37 @@ static const struct vkd3d_shader_immediate_constant_buffer *resolve_forward_init
 
     assert(index);
     --index;
-    if (!(value = sm6_parser_get_value_safe(sm6, index)) || !sm6_value_is_icb(value))
+    if (!(value = sm6_parser_get_value_safe(sm6, index)) || (!sm6_value_is_icb(value) && !sm6_value_is_undef(value)))
     {
         WARN("Invalid initialiser index %zu.\n", index);
         vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
                 "Global variable initialiser value index %zu is invalid.", index);
         return NULL;
     }
-    else
+    else if (sm6_value_is_icb(value))
     {
         return value->u.icb;
     }
+    /* In VSIR, initialisation with undefined values of objects is implied, not explicit. */
+    return NULL;
 }
 
 static enum vkd3d_result sm6_parser_globals_init(struct sm6_parser *sm6)
 {
+    size_t i, count, base_value_idx = sm6->value_count;
     const struct dxil_block *block = &sm6->root_block;
     struct vkd3d_shader_instruction *ins;
     const struct dxil_record *record;
     enum vkd3d_result ret;
     uint64_t version;
-    size_t i;
 
     sm6->p.location.line = block->id;
     sm6->p.location.column = 0;
 
+    for (i = 0, count = 0; i < block->record_count; ++i)
+        count += block->records[i]->code == MODULE_CODE_GLOBALVAR;
+    sm6_parser_require_space(sm6, count);
+
     for (i = 0; i < block->record_count; ++i)
     {
         sm6->p.location.column = i;
@@ -2868,6 +2944,21 @@ static enum vkd3d_result sm6_parser_globals_init(struct sm6_parser *sm6)
             ins->declaration.indexable_temp.initialiser = resolve_forward_initialiser(
                     (uintptr_t)ins->declaration.indexable_temp.initialiser, sm6);
         }
+        else if (ins->handler_idx == VKD3DSIH_DCL_IMMEDIATE_CONSTANT_BUFFER)
+        {
+            ins->declaration.icb = resolve_forward_initialiser((uintptr_t)ins->declaration.icb, sm6);
+        }
+    }
+    for (i = base_value_idx; i < sm6->value_count; ++i)
+    {
+        const struct vkd3d_shader_immediate_constant_buffer *icb;
+        struct sm6_value *value = &sm6->values[i];
+
+        if (!sm6_value_is_register(value) || value->u.reg.type != VKD3DSPR_IMMCONSTBUFFER)
+            continue;
+
+        if ((icb = resolve_forward_initialiser(value->u.reg.idx[0].offset, sm6)))
+            value->u.reg.idx[0].offset = icb->register_idx;
     }
 
     return VKD3D_OK;
@@ -2908,45 +2999,6 @@ static void sm6_parser_init_signature(struct sm6_parser *sm6, const struct shade
     }
 }
 
-static void sm6_parser_emit_signature(struct sm6_parser *sm6, const struct shader_signature *s,
-        enum vkd3d_shader_opcode handler_idx, enum vkd3d_shader_opcode siv_handler_idx,
-        struct vkd3d_shader_dst_param *params)
-{
-    struct vkd3d_shader_instruction *ins;
-    struct vkd3d_shader_dst_param *param;
-    const struct signature_element *e;
-    unsigned int i;
-
-    for (i = 0; i < s->element_count; ++i)
-    {
-        e = &s->elements[i];
-
-        /* Do not check e->used_mask because in some cases it is zero for used elements.
-         * TODO: scan ahead for used I/O elements. */
-
-        if (e->sysval_semantic != VKD3D_SHADER_SV_NONE && e->sysval_semantic != VKD3D_SHADER_SV_TARGET)
-        {
-            ins = sm6_parser_add_instruction(sm6, siv_handler_idx);
-            param = &ins->declaration.register_semantic.reg;
-            ins->declaration.register_semantic.sysval_semantic = vkd3d_siv_from_sysval(e->sysval_semantic);
-        }
-        else
-        {
-            ins = sm6_parser_add_instruction(sm6, handler_idx);
-            param = &ins->declaration.dst;
-        }
-
-        ins->flags = e->interpolation_mode;
-        *param = params[i];
-
-        if (e->register_count > 1)
-        {
-            param->reg.idx[0].rel_addr = NULL;
-            param->reg.idx[0].offset = e->register_count;
-        }
-    }
-}
-
 static void sm6_parser_init_output_signature(struct sm6_parser *sm6, const struct shader_signature *output_signature)
 {
     sm6_parser_init_signature(sm6, output_signature, VKD3DSPR_OUTPUT, sm6->output_params);
@@ -2957,19 +3009,6 @@ static void sm6_parser_init_input_signature(struct sm6_parser *sm6, const struct
     sm6_parser_init_signature(sm6, input_signature, VKD3DSPR_INPUT, sm6->input_params);
 }
 
-static void sm6_parser_emit_output_signature(struct sm6_parser *sm6, const struct shader_signature *output_signature)
-{
-    sm6_parser_emit_signature(sm6, output_signature, VKD3DSIH_DCL_OUTPUT, VKD3DSIH_DCL_OUTPUT_SIV, sm6->output_params);
-}
-
-static void sm6_parser_emit_input_signature(struct sm6_parser *sm6, const struct shader_signature *input_signature)
-{
-    sm6_parser_emit_signature(sm6, input_signature,
-            (sm6->p.shader_version.type == VKD3D_SHADER_TYPE_PIXEL) ? VKD3DSIH_DCL_INPUT_PS : VKD3DSIH_DCL_INPUT,
-            (sm6->p.shader_version.type == VKD3D_SHADER_TYPE_PIXEL) ? VKD3DSIH_DCL_INPUT_PS_SIV : VKD3DSIH_DCL_INPUT_SIV,
-            sm6->input_params);
-}
-
 static const struct sm6_value *sm6_parser_next_function_definition(struct sm6_parser *sm6)
 {
     size_t i, count = sm6->function_count;
@@ -2992,6 +3031,81 @@ static struct sm6_block *sm6_block_create()
     return block;
 }
 
+static void sm6_parser_emit_alloca(struct sm6_parser *sm6, const struct dxil_record *record,
+        struct vkd3d_shader_instruction *ins, struct sm6_value *dst)
+{
+    const struct sm6_type *type[2], *elem_type;
+    const struct sm6_value *size;
+    unsigned int i, alignment;
+    uint64_t packed_operands;
+
+    if (!dxil_record_validate_operand_count(record, 4, 4, sm6))
+        return;
+
+    for (i = 0; i < 2; ++i)
+    {
+        if (!(type[i] = sm6_parser_get_type(sm6, record->operands[i])))
+            return;
+    }
+
+    packed_operands = record->operands[3];
+    if (packed_operands & ALLOCA_FLAG_IN_ALLOCA)
+        WARN("Ignoring in_alloca flag.\n");
+    if (!(packed_operands & ALLOCA_FLAG_EXPLICIT_TYPE))
+    {
+        FIXME("Unhandled implicit type.\n");
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
+                "Implicit result type for ALLOCA instructions is not supported.");
+        return;
+    }
+    packed_operands &= ~(ALLOCA_FLAG_IN_ALLOCA | ALLOCA_FLAG_EXPLICIT_TYPE);
+
+    if (!sm6_type_is_array(type[0]) || !sm6_type_is_numeric(elem_type = type[0]->u.array.elem_type))
+    {
+        WARN("Type is not a numeric array.\n");
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
+                "Result type of an ALLOCA instruction is not a numeric array.");
+        return;
+    }
+    /* The second type operand is the type of the allocation size operand, in case it is a
+     * forward reference. We only support a constant size, so no forward ref support is needed. */
+    if (!sm6_type_is_integer(type[1]))
+    {
+        WARN("Size operand type is not scalar integer.\n");
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
+                "The type of the allocation size operand of an ALLOCA instruction is not scalar integer.");
+        return;
+    }
+
+    if (!(dst->type = sm6_type_get_pointer_to_type(type[0], ADDRESS_SPACE_DEFAULT, sm6)))
+    {
+        WARN("Failed to get pointer type for type class %u.\n", type[0]->class);
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE,
+                "Module does not define a pointer type for an ALLOCA instruction.");
+        return;
+    }
+
+    if (!(size = sm6_parser_get_value_safe(sm6, record->operands[2])))
+        return;
+    /* A size of 1 means one instance of type[0], i.e. one array. */
+    if (sm6_value_get_constant_uint(size) != 1)
+    {
+        FIXME("Allocation size is not 1.\n");
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
+                "ALLOCA instruction allocation sizes other than 1 are not supported.");
+        return;
+    }
+
+    if (!bitcode_parse_alignment(packed_operands & ALLOCA_ALIGNMENT_MASK, &alignment))
+        WARN("Invalid alignment %"PRIu64".\n", packed_operands);
+    packed_operands &= ~ALLOCA_ALIGNMENT_MASK;
+
+    if (packed_operands)
+        WARN("Ignoring flags %#"PRIx64".\n", packed_operands);
+
+    sm6_parser_declare_indexable_temp(sm6, elem_type, type[0]->u.array.count, alignment, 0, ins, dst);
+}
+
 static enum vkd3d_shader_opcode map_binary_op(uint64_t code, const struct sm6_type *type_a,
         const struct sm6_type *type_b, struct sm6_parser *sm6)
 {
@@ -3092,8 +3206,10 @@ static void sm6_parser_emit_binop(struct sm6_parser *sm6, const struct dxil_reco
     unsigned int i = 0;
 
     a = sm6_parser_get_value_by_ref(sm6, record, NULL, &i);
+    if (!a)
+        return;
     b = sm6_parser_get_value_by_ref(sm6, record, a->type, &i);
-    if (!a || !b)
+    if (!b)
         return;
 
     if (!dxil_record_validate_operand_count(record, i + 1, i + 2, sm6))
@@ -3167,10 +3283,77 @@ static void sm6_parser_emit_binop(struct sm6_parser *sm6, const struct dxil_reco
     }
     else
     {
+        if (handler_idx == VKD3DSIH_ISHL || handler_idx == VKD3DSIH_ISHR || handler_idx == VKD3DSIH_USHR)
+        {
+            /* DXC emits AND instructions where necessary to mask shift counts. Shift binops
+             * do not imply masking the shift as the TPF equivalents do. */
+            ins->flags |= VKD3DSI_SHIFT_UNMASKED;
+        }
         instruction_dst_param_init_ssa_scalar(ins, sm6);
     }
 }
 
+static enum vkd3d_shader_opcode map_dx_unary_op(enum dx_intrinsic_opcode op)
+{
+    switch (op)
+    {
+        case DX_EXP:
+            return VKD3DSIH_EXP;
+        case DX_FRC:
+            return VKD3DSIH_FRC;
+        case DX_LOG:
+            return VKD3DSIH_LOG;
+        case DX_SQRT:
+            return VKD3DSIH_SQRT;
+        case DX_RSQRT:
+            return VKD3DSIH_RSQ;
+        case DX_ROUND_NE:
+            return VKD3DSIH_ROUND_NE;
+        case DX_ROUND_NI:
+            return VKD3DSIH_ROUND_NI;
+        case DX_ROUND_PI:
+            return VKD3DSIH_ROUND_PI;
+        case DX_ROUND_Z:
+            return VKD3DSIH_ROUND_Z;
+        case DX_BFREV:
+            return VKD3DSIH_BFREV;
+        case DX_COUNT_BITS:
+            return VKD3DSIH_COUNTBITS;
+        case DX_FIRST_BIT_LO:
+            return VKD3DSIH_FIRSTBIT_LO;
+        case DX_FIRST_BIT_HI:
+            return VKD3DSIH_FIRSTBIT_HI;
+        case DX_FIRST_BIT_SHI:
+            return VKD3DSIH_FIRSTBIT_SHI;
+        case DX_DERIV_COARSEX:
+            return VKD3DSIH_DSX_COARSE;
+        case DX_DERIV_COARSEY:
+            return VKD3DSIH_DSY_COARSE;
+        case DX_DERIV_FINEX:
+            return VKD3DSIH_DSX_FINE;
+        case DX_DERIV_FINEY:
+            return VKD3DSIH_DSY_FINE;
+        case DX_LEGACY_F32TOF16:
+            return VKD3DSIH_F32TOF16;
+        case DX_LEGACY_F16TOF32:
+            return VKD3DSIH_F16TOF32;
+        default:
+            vkd3d_unreachable();
+    }
+}
+
+static void sm6_parser_emit_dx_unary(struct sm6_parser *sm6, struct sm6_block *code_block,
+        enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct vkd3d_shader_instruction *ins)
+{
+    struct vkd3d_shader_src_param *src_param;
+
+    vsir_instruction_init(ins, &sm6->p.location, map_dx_unary_op(op));
+    src_param = instruction_src_params_alloc(ins, 1, sm6);
+    src_param_init_from_value(src_param, operands[0]);
+
+    instruction_dst_param_init_ssa_scalar(ins, sm6);
+}
+
 static void sm6_parser_emit_dx_cbuffer_load(struct sm6_parser *sm6, struct sm6_block *code_block,
         enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct vkd3d_shader_instruction *ins)
 {
@@ -3344,7 +3527,7 @@ static void sm6_parser_emit_dx_store_output(struct sm6_parser *sm6, struct sm6_b
 
 struct sm6_dx_opcode_info
 {
-    const char ret_type;
+    const char *ret_type;
     const char *operand_info;
     void (*handler)(struct sm6_parser *, struct sm6_block *, enum dx_intrinsic_opcode,
             const struct sm6_value **, struct vkd3d_shader_instruction *);
@@ -3355,16 +3538,40 @@ struct sm6_dx_opcode_info
     b -> constant int1
     c -> constant int8/16/32
     i -> int32
+    m -> int16/32/64
+    f -> float
+    e -> half/float
+    g -> half/float/double
     H -> handle
     v -> void
     o -> overloaded
  */
 static const struct sm6_dx_opcode_info sm6_dx_op_table[] =
 {
-    [DX_CBUFFER_LOAD_LEGACY           ] = {'o', "Hi",   sm6_parser_emit_dx_cbuffer_load},
-    [DX_CREATE_HANDLE                 ] = {'H', "ccib", sm6_parser_emit_dx_create_handle},
-    [DX_LOAD_INPUT                    ] = {'o', "ii8i", sm6_parser_emit_dx_load_input},
-    [DX_STORE_OUTPUT                  ] = {'v', "ii8o", sm6_parser_emit_dx_store_output},
+    [DX_BFREV                         ] = {"m0", "m",   sm6_parser_emit_dx_unary},
+    [DX_CBUFFER_LOAD_LEGACY           ] = {"o", "Hi",   sm6_parser_emit_dx_cbuffer_load},
+    [DX_COUNT_BITS                    ] = {"i", "m",    sm6_parser_emit_dx_unary},
+    [DX_CREATE_HANDLE                 ] = {"H", "ccib", sm6_parser_emit_dx_create_handle},
+    [DX_DERIV_COARSEX                 ] = {"e0", "e",   sm6_parser_emit_dx_unary},
+    [DX_DERIV_COARSEY                 ] = {"e0", "e",   sm6_parser_emit_dx_unary},
+    [DX_DERIV_FINEX                   ] = {"e0", "e",   sm6_parser_emit_dx_unary},
+    [DX_DERIV_FINEY                   ] = {"e0", "e",   sm6_parser_emit_dx_unary},
+    [DX_EXP                           ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_FIRST_BIT_HI                  ] = {"i", "m",    sm6_parser_emit_dx_unary},
+    [DX_FIRST_BIT_LO                  ] = {"i", "m",    sm6_parser_emit_dx_unary},
+    [DX_FIRST_BIT_SHI                 ] = {"i", "m",    sm6_parser_emit_dx_unary},
+    [DX_FRC                           ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_LEGACY_F16TOF32               ] = {"f", "i",    sm6_parser_emit_dx_unary},
+    [DX_LEGACY_F32TOF16               ] = {"i", "f",    sm6_parser_emit_dx_unary},
+    [DX_LOAD_INPUT                    ] = {"o", "ii8i", sm6_parser_emit_dx_load_input},
+    [DX_LOG                           ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_ROUND_NE                      ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_ROUND_NI                      ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_ROUND_PI                      ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_ROUND_Z                       ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_RSQRT                         ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_SQRT                          ] = {"g0", "g",   sm6_parser_emit_dx_unary},
+    [DX_STORE_OUTPUT                  ] = {"v", "ii8o", sm6_parser_emit_dx_store_output},
 };
 
 static bool sm6_parser_validate_operand_type(struct sm6_parser *sm6, const struct sm6_value *value, char info_type,
@@ -3389,6 +3596,14 @@ static bool sm6_parser_validate_operand_type(struct sm6_parser *sm6, const struc
                     && type->u.width <= 32;
         case 'i':
             return sm6_type_is_i32(type);
+        case 'm':
+            return sm6_type_is_i16_i32_i64(type);
+        case 'f':
+            return sm6_type_is_float(type);
+        case 'e':
+            return sm6_type_is_f16_f32(type);
+        case 'g':
+            return sm6_type_is_floating_point(type);
         case 'H':
             return (is_return || sm6_value_is_handle(value)) && type == sm6->handle_type;
         case 'v':
@@ -3402,6 +3617,17 @@ static bool sm6_parser_validate_operand_type(struct sm6_parser *sm6, const struc
     }
 }
 
+static bool operand_types_match(struct sm6_value *dst, const struct sm6_value **operands, unsigned int operand_count,
+        char index_char)
+{
+    unsigned int i = index_char - '0';
+
+    assert(i < 10);
+    if (i >= operand_count)
+        return false;
+    return dst->type == operands[i]->type;
+}
+
 static bool sm6_parser_validate_dx_op(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, const char *name,
         const struct sm6_value **operands, unsigned int operand_count, struct sm6_value *dst)
 {
@@ -3410,7 +3636,9 @@ static bool sm6_parser_validate_dx_op(struct sm6_parser *sm6, enum dx_intrinsic_
 
     info = &sm6_dx_op_table[op];
 
-    if (!sm6_parser_validate_operand_type(sm6, dst, info->ret_type, true))
+    assert(info->ret_type[0]);
+    if (!sm6_parser_validate_operand_type(sm6, dst, info->ret_type[0], true)
+            || (info->ret_type[1] && !operand_types_match(dst, operands, operand_count, info->ret_type[1])))
     {
         WARN("Failed to validate return type for dx intrinsic id %u, '%s'.\n", op, name);
         /* Return type validation failure is not so critical. We only need to set
@@ -3700,6 +3928,9 @@ static void sm6_parser_emit_cast(struct sm6_parser *sm6, const struct dxil_recor
     if (handler_idx == VKD3DSIH_NOP)
     {
         dst->u.reg = value->u.reg;
+        /* Set the result type for casts from 16-bit min precision. */
+        if (type->u.width != 16)
+            dst->u.reg.data_type = vkd3d_data_type_from_sm6_type(type);
         return;
     }
 
@@ -3775,8 +4006,10 @@ static void sm6_parser_emit_cmp2(struct sm6_parser *sm6, const struct dxil_recor
     }
 
     a = sm6_parser_get_value_by_ref(sm6, record, NULL, &i);
+    if (!a)
+        return;
     b = sm6_parser_get_value_by_ref(sm6, record, a->type, &i);
-    if (!a || !b)
+    if (!b)
         return;
 
     if (!dxil_record_validate_operand_count(record, i + 1, i + 2, sm6))
@@ -4065,6 +4298,57 @@ static void sm6_parser_emit_ret(struct sm6_parser *sm6, const struct dxil_record
     ins->handler_idx = VKD3DSIH_NOP;
 }
 
+static void sm6_parser_emit_store(struct sm6_parser *sm6, const struct dxil_record *record,
+        struct vkd3d_shader_instruction *ins, struct sm6_value *dst)
+{
+    struct vkd3d_shader_src_param *src_param;
+    struct vkd3d_shader_dst_param *dst_param;
+    const struct sm6_type *pointee_type;
+    const struct sm6_value *ptr, *src;
+    unsigned int i = 0, alignment;
+    uint64_t alignment_code;
+
+    if (!(ptr = sm6_parser_get_value_by_ref(sm6, record, NULL, &i))
+            || !sm6_value_validate_is_register(ptr, sm6)
+            || !sm6_value_validate_is_pointer(ptr, sm6))
+    {
+        return;
+    }
+
+    pointee_type = ptr->type->u.pointer.type;
+    if (!(src = sm6_parser_get_value_by_ref(sm6, record, pointee_type, &i)))
+        return;
+    if (!sm6_value_validate_is_numeric(src, sm6))
+        return;
+
+    if (pointee_type != src->type)
+    {
+        WARN("Type mismatch.\n");
+        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_TYPE_MISMATCH,
+                "Type mismatch in pointer store arguments.");
+    }
+
+    if (!dxil_record_validate_operand_count(record, i + 2, i + 2, sm6))
+        return;
+
+    alignment_code = record->operands[i++];
+    if (!bitcode_parse_alignment(alignment_code, &alignment))
+        WARN("Invalid alignment %"PRIu64".\n", alignment_code);
+
+    if (record->operands[i])
+        WARN("Ignoring volatile modifier.\n");
+
+    vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_MOV);
+
+    src_param = instruction_src_params_alloc(ins, 1, sm6);
+    src_param_init_from_value(&src_param[0], src);
+
+    dst_param = instruction_dst_params_alloc(ins, 1, sm6);
+    dst_param_init(dst_param);
+    dst_param->reg = ptr->u.reg;
+    dst_param->reg.alignment = alignment;
+}
+
 static void sm6_parser_emit_vselect(struct sm6_parser *sm6, const struct dxil_record *record,
         struct vkd3d_shader_instruction *ins, struct sm6_value *dst)
 {
@@ -4153,6 +4437,208 @@ static bool sm6_metadata_get_uint64_value(const struct sm6_parser *sm6,
     return true;
 }
 
+static void sm6_parser_metadata_attachment_block_init(struct sm6_parser *sm6, const struct dxil_block *target_block,
+        const struct dxil_block *block)
+{
+    struct dxil_record *target_record;
+    const struct dxil_record *record;
+    unsigned int i;
+    uint64_t index;
+
+    for (i = 0; i < block->record_count; ++i)
+    {
+        record = block->records[i];
+        if (record->code != METADATA_ATTACHMENT)
+        {
+            WARN("Ignoring record with code %u.\n", record->code);
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "Ignoring a metadata attachment record with code %u.", record->code);
+            continue;
+        }
+        if (!(record->operand_count & 1))
+        {
+            WARN("Ignoring function attachment.\n");
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "Ignoring a metadata function attachment.");
+            continue;
+        }
+
+        index = record->operands[0];
+        if (!target_block->record_count || index >= target_block->record_count - 1)
+        {
+            WARN("Invalid record index %"PRIu64".\n", index);
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND,
+                    "Invalid record index %"PRIu64" for a metadata attachment.", index);
+            continue;
+        }
+        /* 'index' is an instruction index, but records[0] is DECLAREBLOCKS, not an instruction. */
+        target_record = target_block->records[index + 1];
+        if (target_record->attachment)
+        {
+            WARN("Overwriting record attachment.\n");
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "The target record for a metadata attachment already has an attachment.");
+        }
+        target_record->attachment = record;
+    }
+}
+
+static void sm6_parser_metadata_attachments_init(struct sm6_parser *sm6, const struct dxil_block *block)
+{
+    unsigned int i;
+
+    for (i = 0; i < block->child_block_count; ++i)
+    {
+        if (block->child_blocks[i]->id == METADATA_ATTACHMENT_BLOCK)
+            sm6_parser_metadata_attachment_block_init(sm6, block, block->child_blocks[i]);
+    }
+}
+
+static const struct sm6_metadata_value *sm6_parser_find_metadata_kind(const struct sm6_parser *sm6, uint64_t id)
+{
+    unsigned int i, j;
+
+    for (i = 0; i < ARRAY_SIZE(sm6->metadata_tables); ++i)
+    {
+        for (j = 0; j < sm6->metadata_tables[i].count; ++j)
+        {
+            if (sm6->metadata_tables[i].values[j].type == VKD3D_METADATA_KIND
+                    && sm6->metadata_tables[i].values[j].u.kind.id == id)
+                return &sm6->metadata_tables[i].values[j];
+        }
+    }
+
+    return NULL;
+}
+
+static const struct sm6_metadata_value *sm6_parser_metadata_get_value(const struct sm6_parser *sm6, uint64_t index)
+{
+    unsigned int i;
+
+    for (i = 0; i < ARRAY_SIZE(sm6->metadata_tables); ++i)
+    {
+        if (sm6->metadata_tables[i].count > index)
+            break;
+        index -= sm6->metadata_tables[i].count;
+    }
+
+    return (index < sm6->metadata_tables[i].count) ? &sm6->metadata_tables[i].values[index] : NULL;
+}
+
+static bool metadata_node_get_unary_uint(const struct sm6_metadata_node *node, unsigned int *operand,
+        struct sm6_parser *sm6)
+{
+    if (node->operand_count != 1)
+    {
+        FIXME("Ignoring node with %u operands.\n", node->operand_count);
+        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                "Ignoring metadata attachment node with %u operands; expected unary.", node->operand_count);
+        return false;
+    }
+    if (!sm6_metadata_value_is_value(node->operands[0])
+            || !sm6_metadata_get_uint_value(sm6, node->operands[0], operand))
+    {
+        WARN("Failed to get operand value.\n");
+        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                "Failed to get a metadata attachment operand value; ignoring the attachment.");
+        return false;
+    }
+
+    return true;
+}
+
+static void metadata_attachment_record_apply(const struct dxil_record *record, enum bitcode_function_code func_code,
+        struct vkd3d_shader_instruction *ins, struct sm6_value *dst, struct sm6_parser *sm6)
+{
+    static const char *ignored_names[] =
+    {
+        "alias.scope",
+        "dx.controlflow.hints",
+        "llvm.loop",
+        "noalias",
+        "tbaa",
+        "range",
+    };
+    const struct sm6_metadata_node *node;
+    const struct sm6_metadata_value *m;
+    unsigned int i, j, operand;
+    bool ignored = false;
+    const char *name;
+
+    if (record->attachment)
+    {
+        WARN("Ignoring nested metadata attachment.\n");
+        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                "Ignoring a nested metadata attachment.");
+    }
+
+    assert(record->operand_count & 1);
+    for (i = 1; i < record->operand_count; i += 2)
+    {
+        if (!(m = sm6_parser_find_metadata_kind(sm6, record->operands[i])))
+        {
+            WARN("Failed to find metadata kind %"PRIx64".\n", record->operands[i]);
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "Failed to find metadata kind %"PRIx64" for an attachment.", record->operands[i]);
+            continue;
+        }
+        name = m->u.kind.name;
+
+        m = sm6_parser_metadata_get_value(sm6, record->operands[i + 1]);
+        if (!m || !sm6_metadata_value_is_node(m))
+        {
+            WARN("Failed to retrieve metadata attachment node.\n");
+            vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                    "Failed to retrieve a metadata attachment node.");
+            continue;
+        }
+        node = m->u.node;
+
+        if (!strcmp(name, "dx.precise"))
+        {
+            if (!sm6_value_is_register(dst))
+            {
+                WARN("Precise value is not a register.\n");
+                vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                        "A value marked as precise is not a register.");
+            }
+            else if (metadata_node_get_unary_uint(node, &operand, sm6) && operand)
+            {
+                ins->flags |= sm6_type_is_scalar(dst->type) ? VKD3DSI_PRECISE_X : VKD3DSI_PRECISE_XYZW;
+            }
+        }
+        else if (!strcmp(name, "dx.nonuniform"))
+        {
+            if (!sm6_value_is_register(dst))
+            {
+                WARN("Non-uniform value is not a register.\n");
+                vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                        "A value marked as non-uniform is not a register.");
+            }
+            else if (metadata_node_get_unary_uint(node, &operand, sm6))
+            {
+                dst->u.reg.non_uniform = !!operand;
+            }
+        }
+        else
+        {
+            for (j = 0; j < ARRAY_SIZE(ignored_names); ++j)
+                if (!strcmp(name, ignored_names[j]))
+                    break;
+            if (j == ARRAY_SIZE(ignored_names))
+            {
+                WARN("Ignoring metadata attachment '%s'.\n", name);
+                vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT,
+                        "Ignoring a metadata attachment named '%s'.", name);
+            }
+            ignored = true;
+        }
+
+        if (func_code != FUNC_CODE_INST_CALL && !ignored)
+            WARN("Metadata attachment target is not a function call.\n");
+    }
+}
+
 static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const struct dxil_block *block,
         struct sm6_function *function)
 {
@@ -4240,6 +4726,9 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const
         record = block->records[i];
         switch (record->code)
         {
+            case FUNC_CODE_INST_ALLOCA:
+                sm6_parser_emit_alloca(sm6, record, ins, dst);
+                break;
             case FUNC_CODE_INST_BINOP:
                 sm6_parser_emit_binop(sm6, record, ins, dst);
                 break;
@@ -4266,6 +4755,9 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const
                 is_terminator = true;
                 ret_found = true;
                 break;
+            case FUNC_CODE_INST_STORE:
+                sm6_parser_emit_store(sm6, record, ins, dst);
+                break;
             case FUNC_CODE_INST_VSELECT:
                 sm6_parser_emit_vselect(sm6, record, ins, dst);
                 break;
@@ -4278,6 +4770,9 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const
             return VKD3D_ERROR;
         assert(ins->handler_idx != VKD3DSIH_INVALID);
 
+        if (record->attachment)
+            metadata_attachment_record_apply(record->attachment, record->code, ins, dst, sm6);
+
         if (is_terminator)
         {
             ++block_idx;
@@ -4331,6 +4826,8 @@ static enum vkd3d_result sm6_parser_module_init(struct sm6_parser *sm6, const st
     sm6->p.location.line = block->id;
     sm6->p.location.column = 0;
 
+    sm6_parser_metadata_attachments_init(sm6, block);
+
     switch (block->id)
     {
         case CONSTANTS_BLOCK:
@@ -4645,7 +5142,7 @@ static const enum vkd3d_shader_sysval_semantic sysval_semantic_table[] =
 {
     [SEMANTIC_KIND_ARBITRARY]            = VKD3D_SHADER_SV_NONE,
     [SEMANTIC_KIND_POSITION]             = VKD3D_SHADER_SV_POSITION,
-    [SEMANTIC_KIND_TARGET]               = VKD3D_SHADER_SV_NONE,
+    [SEMANTIC_KIND_TARGET]               = VKD3D_SHADER_SV_TARGET,
 };
 
 static enum vkd3d_shader_sysval_semantic sysval_semantic_from_dxil_semantic_kind(enum dxil_semantic_kind kind)
@@ -5538,7 +6035,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((magic = sm6->start[0]) != BITCODE_MAGIC)
     {
         WARN("Unknown magic number 0x%08x.\n", magic);
-        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_UNKNOWN_MAGIC_NUMBER,
+        vkd3d_shader_warning(message_context, &location, VKD3D_SHADER_WARNING_DXIL_UNKNOWN_MAGIC_NUMBER,
                 "DXIL bitcode chunk magic number 0x%08x is not the expected 0x%08x.", magic, BITCODE_MAGIC);
     }
 
@@ -5547,7 +6044,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((version.type = version_token >> 16) >= VKD3D_SHADER_TYPE_COUNT)
     {
         FIXME("Unknown shader type %#x.\n", version.type);
-        vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_UNKNOWN_SHADER_TYPE,
+        vkd3d_shader_warning(message_context, &location, VKD3D_SHADER_WARNING_DXIL_UNKNOWN_SHADER_TYPE,
                 "Unknown shader type %#x.", version.type);
     }
 
@@ -5573,10 +6070,10 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((ret = dxil_block_init(block, NULL, sm6)) < 0)
     {
         if (ret == VKD3D_ERROR_OUT_OF_MEMORY)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                     "Out of memory parsing DXIL bitcode chunk.");
         else if (ret == VKD3D_ERROR_INVALID_SHADER)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_BITCODE,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_BITCODE,
                     "DXIL bitcode chunk has invalid bitcode.");
         else
             vkd3d_unreachable();
@@ -5606,10 +6103,10 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((ret = sm6_parser_type_table_init(sm6)) < 0)
     {
         if (ret == VKD3D_ERROR_OUT_OF_MEMORY)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                     "Out of memory parsing DXIL type table.");
         else if (ret == VKD3D_ERROR_INVALID_SHADER)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_TYPE_TABLE,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_TYPE_TABLE,
                     "DXIL type table is invalid.");
         else
             vkd3d_unreachable();
@@ -5619,10 +6116,10 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((ret = sm6_parser_symtab_init(sm6)) < 0)
     {
         if (ret == VKD3D_ERROR_OUT_OF_MEMORY)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                     "Out of memory parsing DXIL value symbol table.");
         else if (ret == VKD3D_ERROR_INVALID_SHADER)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_VALUE_SYMTAB,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_VALUE_SYMTAB,
                     "DXIL value symbol table is invalid.");
         else
             vkd3d_unreachable();
@@ -5633,7 +6130,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
             || !(sm6->input_params = shader_parser_get_dst_params(&sm6->p, input_signature->element_count)))
     {
         ERR("Failed to allocate input/output parameters.\n");
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                 "Out of memory allocating input/output parameters.");
         return VKD3D_ERROR_OUT_OF_MEMORY;
     }
@@ -5642,7 +6139,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if (!(sm6->functions = vkd3d_calloc(function_count, sizeof(*sm6->functions))))
     {
         ERR("Failed to allocate function array.\n");
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                 "Out of memory allocating DXIL function array.");
         return VKD3D_ERROR_OUT_OF_MEMORY;
     }
@@ -5650,14 +6147,14 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if (sm6_parser_compute_max_value_count(sm6, &sm6->root_block, 0) == SIZE_MAX)
     {
         WARN("Value array count overflowed.\n");
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE,
                 "Overflow occurred in the DXIL module value count.");
         return VKD3D_ERROR_INVALID_SHADER;
     }
     if (!(sm6->values = vkd3d_calloc(sm6->value_capacity, sizeof(*sm6->values))))
     {
         ERR("Failed to allocate value array.\n");
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                 "Out of memory allocating DXIL value array.");
         return VKD3D_ERROR_OUT_OF_MEMORY;
     }
@@ -5684,7 +6181,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
         if (j == ARRAY_SIZE(sm6->metadata_tables))
         {
             FIXME("Too many metadata tables.\n");
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_METADATA,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_METADATA,
                     "A metadata table count greater than %zu is unsupported.", ARRAY_SIZE(sm6->metadata_tables));
             return VKD3D_ERROR_INVALID_SHADER;
         }
@@ -5702,22 +6199,20 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     if ((ret = sm6_parser_module_init(sm6, &sm6->root_block, 0)) < 0)
     {
         if (ret == VKD3D_ERROR_OUT_OF_MEMORY)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                     "Out of memory parsing DXIL module.");
         else if (ret == VKD3D_ERROR_INVALID_SHADER)
-            vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE,
+            vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE,
                     "DXIL module is invalid.");
         return ret;
     }
 
     if (!sm6_parser_require_space(sm6, output_signature->element_count + input_signature->element_count))
     {
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                 "Out of memory emitting shader signature declarations.");
         return VKD3D_ERROR_OUT_OF_MEMORY;
     }
-    sm6_parser_emit_output_signature(sm6, output_signature);
-    sm6_parser_emit_input_signature(sm6, input_signature);
 
     sm6->p.shader_desc.ssa_count = sm6->ssa_next_id;
 
@@ -5732,7 +6227,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t
     assert(sm6->function_count == 1);
     if (!sm6_block_emit_instructions(fn->blocks[0], sm6))
     {
-        vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
+        vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY,
                 "Out of memory emitting shader instructions.");
         return VKD3D_ERROR_OUT_OF_MEMORY;
     }
diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl.y b/libs/vkd3d/libs/vkd3d-shader/hlsl.y
index 0dde4c18587..ed053f16312 100644
--- a/libs/vkd3d/libs/vkd3d-shader/hlsl.y
+++ b/libs/vkd3d/libs/vkd3d-shader/hlsl.y
@@ -487,27 +487,27 @@ static void resolve_loop_continue(struct hlsl_ctx *ctx, struct hlsl_block *block
         else if (instr->type == HLSL_IR_JUMP)
         {
             struct hlsl_ir_jump *jump = hlsl_ir_jump(instr);
-            struct hlsl_block block;
+            struct hlsl_block cond_block;
 
             if (jump->type != HLSL_IR_JUMP_UNRESOLVED_CONTINUE)
                 continue;
 
             if (type == LOOP_DO_WHILE)
             {
-                if (!hlsl_clone_block(ctx, &block, cond))
+                if (!hlsl_clone_block(ctx, &cond_block, cond))
                     return;
-                if (!append_conditional_break(ctx, &block))
+                if (!append_conditional_break(ctx, &cond_block))
                 {
-                    hlsl_block_cleanup(&block);
+                    hlsl_block_cleanup(&cond_block);
                     return;
                 }
-                list_move_before(&instr->entry, &block.instrs);
+                list_move_before(&instr->entry, &cond_block.instrs);
             }
             else if (type == LOOP_FOR)
             {
-                if (!hlsl_clone_block(ctx, &block, iter))
+                if (!hlsl_clone_block(ctx, &cond_block, iter))
                     return;
-                list_move_before(&instr->entry, &block.instrs);
+                list_move_before(&instr->entry, &cond_block.instrs);
             }
             jump->type = HLSL_IR_JUMP_CONTINUE;
         }
@@ -3553,7 +3553,7 @@ static bool intrinsic_tex(struct hlsl_ctx *ctx, const struct parse_initializer *
 {
     struct hlsl_resource_load_params load_params = { 0 };
     const struct hlsl_type *sampler_type;
-    struct hlsl_ir_node *coords, *load;
+    struct hlsl_ir_node *coords, *sample;
 
     if (params->args_count != 2 && params->args_count != 4)
     {
@@ -3688,9 +3688,9 @@ static bool intrinsic_tex(struct hlsl_ctx *ctx, const struct parse_initializer *
     load_params.format = hlsl_get_vector_type(ctx, HLSL_TYPE_FLOAT, 4);
     load_params.sampling_dim = dim;
 
-    if (!(load = hlsl_new_resource_load(ctx, &load_params, loc)))
+    if (!(sample = hlsl_new_resource_load(ctx, &load_params, loc)))
         return false;
-    hlsl_block_add_instr(params->instrs, load);
+    hlsl_block_add_instr(params->instrs, sample);
     return true;
 }
 
diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c
index 4a622185741..88cbef61d5f 100644
--- a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c
@@ -1090,7 +1090,7 @@ static bool lower_index_loads(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr,
         unsigned int dim_count = hlsl_sampler_dim_count(val->data_type->sampler_dim);
         struct hlsl_ir_node *coords = index->idx.node;
         struct hlsl_resource_load_params params = {0};
-        struct hlsl_ir_node *load;
+        struct hlsl_ir_node *resource_load;
 
         assert(coords->data_type->class == HLSL_CLASS_VECTOR);
         assert(coords->data_type->base_type == HLSL_TYPE_UINT);
@@ -1104,9 +1104,9 @@ static bool lower_index_loads(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr,
         params.coords = coords;
         params.format = val->data_type->e.resource_format;
 
-        if (!(load = hlsl_new_resource_load(ctx, &params, &instr->loc)))
+        if (!(resource_load = hlsl_new_resource_load(ctx, &params, &instr->loc)))
             return false;
-        hlsl_block_add_instr(block, load);
+        hlsl_block_add_instr(block, resource_load);
         return true;
     }
 
@@ -2301,7 +2301,6 @@ static bool normalize_switch_cases(struct hlsl_ctx *ctx, struct hlsl_ir_node *in
     struct hlsl_ir_switch_case *c, *def = NULL;
     bool missing_terminal_break = false;
     struct hlsl_ir_node *node;
-    struct hlsl_ir_jump *jump;
     struct hlsl_ir_switch *s;
 
     if (instr->type != HLSL_IR_SWITCH)
@@ -2320,10 +2319,7 @@ static bool normalize_switch_cases(struct hlsl_ctx *ctx, struct hlsl_ir_node *in
         {
             node = LIST_ENTRY(list_tail(&c->body.instrs), struct hlsl_ir_node, entry);
             if (node->type == HLSL_IR_JUMP)
-            {
-                jump = hlsl_ir_jump(node);
-                terminal_break = jump->type == HLSL_IR_JUMP_BREAK;
-            }
+                terminal_break = (hlsl_ir_jump(node)->type == HLSL_IR_JUMP_BREAK);
         }
 
         missing_terminal_break |= !terminal_break;
diff --git a/libs/vkd3d/libs/vkd3d-shader/ir.c b/libs/vkd3d/libs/vkd3d-shader/ir.c
index 710811c6761..d38b3c39712 100644
--- a/libs/vkd3d/libs/vkd3d-shader/ir.c
+++ b/libs/vkd3d/libs/vkd3d-shader/ir.c
@@ -63,7 +63,7 @@ static void shader_instruction_eliminate_phase_instance_id(struct vkd3d_shader_i
         if (shader_register_is_phase_instance_id(reg))
         {
             vsir_register_init(reg, VKD3DSPR_IMMCONST, reg->data_type, 0);
-            reg->u.immconst_uint[0] = instance_id;
+            reg->u.immconst_u32[0] = instance_id;
             continue;
         }
         shader_register_eliminate_phase_addressing(reg, instance_id);
@@ -463,6 +463,7 @@ static enum vkd3d_result instruction_array_normalise_hull_shader_control_point_i
     struct vkd3d_shader_instruction_array *instructions;
     struct control_point_normaliser normaliser;
     unsigned int input_control_point_count;
+    struct vkd3d_shader_location location;
     struct vkd3d_shader_instruction *ins;
     enum vkd3d_result ret;
     unsigned int i, j;
@@ -513,8 +514,10 @@ static enum vkd3d_result instruction_array_normalise_hull_shader_control_point_i
                 return VKD3D_OK;
             case VKD3DSIH_HS_FORK_PHASE:
             case VKD3DSIH_HS_JOIN_PHASE:
+                /* ins may be relocated if the instruction array expands. */
+                location = ins->location;
                 ret = control_point_normaliser_emit_hs_input(&normaliser, input_signature,
-                        input_control_point_count, i, &ins->location);
+                        input_control_point_count, i, &location);
                 *src_instructions = normaliser.instructions;
                 return ret;
             default:
@@ -590,19 +593,19 @@ struct signature_element *vsir_signature_find_element_for_reg(const struct shade
 }
 
 static unsigned int range_map_get_register_count(uint8_t range_map[][VKD3D_VEC4_SIZE],
-        unsigned int register_idx, unsigned int write_mask)
+        unsigned int register_idx, uint32_t write_mask)
 {
-    return range_map[register_idx][vkd3d_write_mask_get_component_idx(write_mask)];
+    return range_map[register_idx][vsir_write_mask_get_component_idx(write_mask)];
 }
 
 static void range_map_set_register_range(uint8_t range_map[][VKD3D_VEC4_SIZE], unsigned int register_idx,
-        unsigned int register_count, unsigned int write_mask, bool is_dcl_indexrange)
+        unsigned int register_count, uint32_t write_mask, bool is_dcl_indexrange)
 {
     unsigned int i, j, r, c, component_idx, component_count;
 
     assert(write_mask <= VKD3DSP_WRITEMASK_ALL);
-    component_idx = vkd3d_write_mask_get_component_idx(write_mask);
-    component_count = vkd3d_write_mask_component_count(write_mask);
+    component_idx = vsir_write_mask_get_component_idx(write_mask);
+    component_count = vsir_write_mask_component_count(write_mask);
 
     assert(register_idx < MAX_REG_OUTPUT && MAX_REG_OUTPUT - register_idx >= register_count);
 
@@ -896,7 +899,7 @@ static bool shader_dst_param_io_normalise(struct vkd3d_shader_dst_param *dst_par
     element_idx = shader_signature_find_element_for_reg(signature, reg_idx, write_mask);
     e = &signature->elements[element_idx];
 
-    dst_param->write_mask >>= vkd3d_write_mask_get_component_idx(e->mask);
+    dst_param->write_mask >>= vsir_write_mask_get_component_idx(e->mask);
     if (is_io_dcl)
     {
         /* Validated in the TPF reader. */
@@ -999,7 +1002,7 @@ static void shader_src_param_io_normalise(struct vkd3d_shader_src_param *src_par
 
     id_idx = reg->idx_count - 1;
     reg_idx = reg->idx[id_idx].offset;
-    write_mask = VKD3DSP_WRITEMASK_0 << vkd3d_swizzle_get_component(src_param->swizzle, 0);
+    write_mask = VKD3DSP_WRITEMASK_0 << vsir_swizzle_get_component(src_param->swizzle, 0);
     element_idx = shader_signature_find_element_for_reg(signature, reg_idx, write_mask);
 
     e = &signature->elements[element_idx];
@@ -1008,10 +1011,10 @@ static void shader_src_param_io_normalise(struct vkd3d_shader_src_param *src_par
     reg->idx[id_idx].offset = element_idx;
     reg->idx_count = id_idx + 1;
 
-    if ((component_idx = vkd3d_write_mask_get_component_idx(e->mask)))
+    if ((component_idx = vsir_write_mask_get_component_idx(e->mask)))
     {
         for (i = 0; i < VKD3D_VEC4_SIZE; ++i)
-            if (vkd3d_swizzle_get_component(src_param->swizzle, i))
+            if (vsir_swizzle_get_component(src_param->swizzle, i))
                 src_param->swizzle -= component_idx << VKD3D_SHADER_SWIZZLE_SHIFT(i);
     }
 }
@@ -1215,7 +1218,7 @@ static void shader_register_normalise_flat_constants(struct vkd3d_shader_src_par
             param->reg.idx_count = 0;
             param->reg.dimension = VSIR_DIMENSION_VEC4;
             for (j = 0; j < 4; ++j)
-                param->reg.u.immconst_uint[j] = normaliser->defs[i].value[j];
+                param->reg.u.immconst_u32[j] = normaliser->defs[i].value[j];
             return;
         }
     }
@@ -1251,7 +1254,7 @@ static enum vkd3d_result instruction_array_normalise_flat_constants(struct vkd3d
 
             get_flat_constant_register_type((struct vkd3d_shader_register *)&ins->dst[0].reg, &def->set, &def->index);
             for (j = 0; j < 4; ++j)
-                def->value[j] = ins->src[0].reg.u.immconst_uint[j];
+                def->value[j] = ins->src[0].reg.u.immconst_u32[j];
 
             vkd3d_shader_instruction_make_nop(ins);
         }
diff --git a/libs/vkd3d/libs/vkd3d-shader/spirv.c b/libs/vkd3d/libs/vkd3d-shader/spirv.c
index c8a43adbe03..387784f2358 100644
--- a/libs/vkd3d/libs/vkd3d-shader/spirv.c
+++ b/libs/vkd3d/libs/vkd3d-shader/spirv.c
@@ -133,7 +133,7 @@ static void vkd3d_spirv_dump(const struct vkd3d_shader_code *spirv,
     }
 }
 
-static void vkd3d_spirv_validate(const struct vkd3d_shader_code *spirv,
+static bool vkd3d_spirv_validate(struct vkd3d_string_buffer *buffer, const struct vkd3d_shader_code *spirv,
         enum vkd3d_shader_spirv_environment environment)
 {
     spv_diagnostic diagnostic = NULL;
@@ -145,12 +145,13 @@ static void vkd3d_spirv_validate(const struct vkd3d_shader_code *spirv,
     if ((ret = spvValidateBinary(context, spirv->code, spirv->size / sizeof(uint32_t),
             &diagnostic)))
     {
-        FIXME("Failed to validate SPIR-V binary, ret %d.\n", ret);
-        FIXME("Diagnostic message: %s.\n", debugstr_a(diagnostic->error));
+        vkd3d_string_buffer_printf(buffer, "%s", diagnostic->error);
     }
 
     spvDiagnosticDestroy(diagnostic);
     spvContextDestroy(context);
+
+    return !ret;
 }
 
 #else
@@ -163,8 +164,11 @@ static enum vkd3d_result vkd3d_spirv_binary_to_text(const struct vkd3d_shader_co
 }
 static void vkd3d_spirv_dump(const struct vkd3d_shader_code *spirv,
         enum vkd3d_shader_spirv_environment environment) {}
-static void vkd3d_spirv_validate(const struct vkd3d_shader_code *spirv,
-        enum vkd3d_shader_spirv_environment environment) {}
+static bool vkd3d_spirv_validate(struct vkd3d_string_buffer *buffer, const struct vkd3d_shader_code *spirv,
+        enum vkd3d_shader_spirv_environment environment)
+{
+    return true;
+}
 
 #endif  /* HAVE_SPIRV_TOOLS */
 
@@ -1194,6 +1198,16 @@ static uint32_t vkd3d_spirv_get_op_constant64(struct vkd3d_spirv_builder *builde
             (const uint32_t *)&value, 2, vkd3d_spirv_build_op_constant64);
 }
 
+static uint32_t vkd3d_spirv_build_op_constant_null(struct vkd3d_spirv_builder *builder, uint32_t result_type)
+{
+    return vkd3d_spirv_build_op_tr(builder, &builder->global_stream, SpvOpConstantNull, result_type);
+}
+
+static uint32_t vkd3d_spirv_get_op_constant_null(struct vkd3d_spirv_builder *builder, uint32_t result_type)
+{
+    return vkd3d_spirv_build_once1(builder, SpvOpConstantNull, result_type, vkd3d_spirv_build_op_constant_null);
+}
+
 static uint32_t vkd3d_spirv_build_op_constant_composite(struct vkd3d_spirv_builder *builder,
         uint32_t result_type, const uint32_t *constituents, unsigned int constituent_count)
 {
@@ -1783,6 +1797,8 @@ static uint32_t vkd3d_spirv_get_type_id(struct vkd3d_spirv_builder *builder,
                 break;
             case VKD3D_SHADER_COMPONENT_DOUBLE:
                 return vkd3d_spirv_get_op_type_float(builder, 64);
+            case VKD3D_SHADER_COMPONENT_UINT64:
+                return vkd3d_spirv_get_op_type_int(builder, 64, 0);
             default:
                 FIXME("Unhandled component type %#x.\n", component_type);
                 return 0;
@@ -1816,6 +1832,8 @@ static uint32_t vkd3d_spirv_get_type_id_for_data_type(struct vkd3d_spirv_builder
                 break;
             case VKD3D_DATA_DOUBLE:
                 return vkd3d_spirv_get_op_type_float(builder, 64);
+            case VKD3D_DATA_UINT64:
+                return vkd3d_spirv_get_op_type_int(builder, 64, 0);
             case VKD3D_DATA_BOOL:
                 return vkd3d_spirv_get_op_type_bool(builder);
             default:
@@ -2152,6 +2170,7 @@ static void vkd3d_symbol_make_register(struct vkd3d_symbol *symbol,
             break;
 
         case VKD3DSPR_IMMCONSTBUFFER:
+            symbol->key.reg.idx = reg->idx_count > 1 ? reg->idx[0].offset : 0;
             break;
 
         default:
@@ -2159,6 +2178,15 @@ static void vkd3d_symbol_make_register(struct vkd3d_symbol *symbol,
     }
 }
 
+static void vkd3d_symbol_make_io(struct vkd3d_symbol *symbol,
+        enum vkd3d_shader_register_type type, unsigned int index)
+{
+    symbol->type = VKD3D_SYMBOL_REGISTER;
+    memset(&symbol->key, 0, sizeof(symbol->key));
+    symbol->key.reg.type = type;
+    symbol->key.reg.idx = index;
+}
+
 static void vkd3d_symbol_set_register_info(struct vkd3d_symbol *symbol,
         uint32_t val_id, SpvStorageClass storage_class,
         enum vkd3d_shader_component_type component_type, DWORD write_mask)
@@ -2376,12 +2404,15 @@ struct spirv_compiler
     struct vkd3d_shader_spec_constant *spec_constants;
     size_t spec_constants_size;
     enum vkd3d_shader_compile_option_formatting_flags formatting;
+    enum vkd3d_shader_compile_option_feature_flags features;
     bool write_tess_geom_point_size;
 
     struct vkd3d_string_buffer_cache string_buffers;
 
     struct ssa_register_info *ssa_register_info;
     unsigned int ssa_register_count;
+
+    uint64_t config_flags;
 };
 
 static bool is_in_default_phase(const struct spirv_compiler *compiler)
@@ -2437,7 +2468,8 @@ static void spirv_compiler_destroy(struct spirv_compiler *compiler)
 static struct spirv_compiler *spirv_compiler_create(const struct vkd3d_shader_version *shader_version,
         struct vkd3d_shader_desc *shader_desc, const struct vkd3d_shader_compile_info *compile_info,
         const struct vkd3d_shader_scan_descriptor_info1 *scan_descriptor_info,
-        struct vkd3d_shader_message_context *message_context, const struct vkd3d_shader_location *location)
+        struct vkd3d_shader_message_context *message_context, const struct vkd3d_shader_location *location,
+        uint64_t config_flags)
 {
     const struct shader_signature *patch_constant_signature = &shader_desc->patch_constant_signature;
     const struct shader_signature *output_signature = &shader_desc->output_signature;
@@ -2454,6 +2486,7 @@ static struct spirv_compiler *spirv_compiler_create(const struct vkd3d_shader_ve
     memset(compiler, 0, sizeof(*compiler));
     compiler->message_context = message_context;
     compiler->location = *location;
+    compiler->config_flags = config_flags;
 
     if ((target_info = vkd3d_find_struct(compile_info->next, SPIRV_TARGET_INFO)))
     {
@@ -2533,6 +2566,10 @@ static struct spirv_compiler *spirv_compiler_create(const struct vkd3d_shader_ve
                     WARN("Ignoring unrecognised value %#x for option %#x.\n", option->value, option->name);
                 break;
 
+            case VKD3D_SHADER_COMPILE_OPTION_FEATURE:
+                compiler->features = option->value;
+                break;
+
             default:
                 WARN("Ignoring unrecognised option %#x with value %#x.\n", option->name, option->value);
                 break;
@@ -2955,7 +2992,7 @@ static uint32_t spirv_compiler_get_constant64(struct spirv_compiler *compiler,
     assert(0 < component_count && component_count <= VKD3D_DVEC2_SIZE);
     type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count);
 
-    if (component_type != VKD3D_SHADER_COMPONENT_DOUBLE)
+    if (component_type != VKD3D_SHADER_COMPONENT_DOUBLE && component_type != VKD3D_SHADER_COMPONENT_UINT64)
     {
         FIXME("Unhandled component_type %#x.\n", component_type);
         return vkd3d_spirv_get_op_undef(builder, type_id);
@@ -3015,14 +3052,21 @@ static uint32_t spirv_compiler_get_constant_double_vector(struct spirv_compiler
             component_count, (const uint64_t *)values);
 }
 
+static uint32_t spirv_compiler_get_constant_uint64_vector(struct spirv_compiler *compiler,
+        uint64_t value, unsigned int component_count)
+{
+    const uint64_t values[] = {value, value};
+    return spirv_compiler_get_constant64(compiler, VKD3D_SHADER_COMPONENT_UINT64, component_count, values);
+}
+
 static uint32_t spirv_compiler_get_type_id_for_reg(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD write_mask)
+        const struct vkd3d_shader_register *reg, uint32_t write_mask)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
 
     return vkd3d_spirv_get_type_id(builder,
             vkd3d_component_type_from_data_type(reg->data_type),
-            vkd3d_write_mask_component_count(write_mask));
+            vsir_write_mask_component_count(write_mask));
 }
 
 static uint32_t spirv_compiler_get_type_id_for_dst(struct spirv_compiler *compiler,
@@ -3498,7 +3542,7 @@ static void spirv_compiler_emit_dereference_register(struct spirv_compiler *comp
     }
     else if (reg->type == VKD3DSPR_IMMCONSTBUFFER)
     {
-        indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, &reg->idx[0]);
+        indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, &reg->idx[reg->idx_count - 1]);
     }
     else if (reg->type == VKD3DSPR_IDXTEMP)
     {
@@ -3530,7 +3574,7 @@ static void spirv_compiler_emit_dereference_register(struct spirv_compiler *comp
 
     if (index_count)
     {
-        component_count = vkd3d_write_mask_component_count(register_info->write_mask);
+        component_count = vsir_write_mask_component_count(register_info->write_mask);
         type_id = vkd3d_spirv_get_type_id(builder, register_info->component_type, component_count);
         ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, register_info->storage_class, type_id);
         register_info->id = vkd3d_spirv_build_op_access_chain(builder, ptr_type_id,
@@ -3562,24 +3606,24 @@ static bool vkd3d_swizzle_is_equal(unsigned int dst_write_mask,
     return vkd3d_compact_swizzle(VKD3D_SHADER_NO_SWIZZLE, dst_write_mask) == vkd3d_compact_swizzle(swizzle, write_mask);
 }
 
-static bool vkd3d_swizzle_is_scalar(unsigned int swizzle)
+static bool vkd3d_swizzle_is_scalar(uint32_t swizzle)
 {
-    unsigned int component_idx = vkd3d_swizzle_get_component(swizzle, 0);
-    return vkd3d_swizzle_get_component(swizzle, 1) == component_idx
-            && vkd3d_swizzle_get_component(swizzle, 2) == component_idx
-            && vkd3d_swizzle_get_component(swizzle, 3) == component_idx;
+    unsigned int component_idx = vsir_swizzle_get_component(swizzle, 0);
+    return vsir_swizzle_get_component(swizzle, 1) == component_idx
+            && vsir_swizzle_get_component(swizzle, 2) == component_idx
+            && vsir_swizzle_get_component(swizzle, 3) == component_idx;
 }
 
 static uint32_t spirv_compiler_emit_swizzle(struct spirv_compiler *compiler,
-        uint32_t val_id, unsigned int val_write_mask, enum vkd3d_shader_component_type component_type,
-        unsigned int swizzle, unsigned int write_mask)
+        uint32_t val_id, uint32_t val_write_mask, enum vkd3d_shader_component_type component_type,
+        uint32_t swizzle, uint32_t write_mask)
 {
     unsigned int i, component_idx, component_count, val_component_count;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, components[VKD3D_VEC4_SIZE];
 
-    component_count = vkd3d_write_mask_component_count(write_mask);
-    val_component_count = vkd3d_write_mask_component_count(val_write_mask);
+    component_count = vsir_write_mask_component_count(write_mask);
+    val_component_count = vsir_write_mask_component_count(val_write_mask);
 
     if (component_count == val_component_count
             && (component_count == 1 || vkd3d_swizzle_is_equal(val_write_mask, swizzle, write_mask)))
@@ -3589,9 +3633,9 @@ static uint32_t spirv_compiler_emit_swizzle(struct spirv_compiler *compiler,
 
     if (component_count == 1)
     {
-        component_idx = vkd3d_write_mask_get_component_idx(write_mask);
-        component_idx = vkd3d_swizzle_get_component(swizzle, component_idx);
-        component_idx -= vkd3d_write_mask_get_component_idx(val_write_mask);
+        component_idx = vsir_write_mask_get_component_idx(write_mask);
+        component_idx = vsir_swizzle_get_component(swizzle, component_idx);
+        component_idx -= vsir_write_mask_get_component_idx(val_write_mask);
         return vkd3d_spirv_build_op_composite_extract1(builder, type_id, val_id, component_idx);
     }
 
@@ -3601,7 +3645,7 @@ static uint32_t spirv_compiler_emit_swizzle(struct spirv_compiler *compiler,
         {
             if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
             {
-                assert(VKD3DSP_WRITEMASK_0 << vkd3d_swizzle_get_component(swizzle, i) == val_write_mask);
+                assert(VKD3DSP_WRITEMASK_0 << vsir_swizzle_get_component(swizzle, i) == val_write_mask);
                 components[component_idx++] = val_id;
             }
         }
@@ -3611,14 +3655,14 @@ static uint32_t spirv_compiler_emit_swizzle(struct spirv_compiler *compiler,
     for (i = 0, component_idx = 0; i < VKD3D_VEC4_SIZE; ++i)
     {
         if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
-            components[component_idx++] = vkd3d_swizzle_get_component(swizzle, i);
+            components[component_idx++] = vsir_swizzle_get_component(swizzle, i);
     }
     return vkd3d_spirv_build_op_vector_shuffle(builder,
             type_id, val_id, val_id, components, component_count);
 }
 
 static uint32_t spirv_compiler_emit_vector_shuffle(struct spirv_compiler *compiler,
-        uint32_t vector1_id, uint32_t vector2_id, unsigned int swizzle, unsigned int write_mask,
+        uint32_t vector1_id, uint32_t vector2_id, uint32_t swizzle, uint32_t write_mask,
         enum vkd3d_shader_component_type component_type, unsigned int component_count)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
@@ -3631,9 +3675,9 @@ static uint32_t spirv_compiler_emit_vector_shuffle(struct spirv_compiler *compil
     for (i = 0; i < component_count; ++i)
     {
         if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
-            components[i] = vkd3d_swizzle_get_component(swizzle, i);
+            components[i] = vsir_swizzle_get_component(swizzle, i);
         else
-            components[i] = VKD3D_VEC4_SIZE + vkd3d_swizzle_get_component(swizzle, i);
+            components[i] = VKD3D_VEC4_SIZE + vsir_swizzle_get_component(swizzle, i);
     }
 
     type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count);
@@ -3642,9 +3686,9 @@ static uint32_t spirv_compiler_emit_vector_shuffle(struct spirv_compiler *compil
 }
 
 static uint32_t spirv_compiler_emit_load_constant(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD swizzle, DWORD write_mask)
+        const struct vkd3d_shader_register *reg, uint32_t swizzle, uint32_t write_mask)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(write_mask);
     uint32_t values[VKD3D_VEC4_SIZE] = {0};
     unsigned int i, j;
 
@@ -3653,14 +3697,14 @@ static uint32_t spirv_compiler_emit_load_constant(struct spirv_compiler *compile
     if (reg->dimension == VSIR_DIMENSION_SCALAR)
     {
         for (i = 0; i < component_count; ++i)
-            values[i] = *reg->u.immconst_uint;
+            values[i] = *reg->u.immconst_u32;
     }
     else
     {
         for (i = 0, j = 0; i < VKD3D_VEC4_SIZE; ++i)
         {
             if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
-                values[j++] = reg->u.immconst_uint[vkd3d_swizzle_get_component(swizzle, i)];
+                values[j++] = reg->u.immconst_u32[vsir_swizzle_get_component(swizzle, i)];
         }
     }
 
@@ -3669,9 +3713,9 @@ static uint32_t spirv_compiler_emit_load_constant(struct spirv_compiler *compile
 }
 
 static uint32_t spirv_compiler_emit_load_constant64(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD swizzle, DWORD write_mask)
+        const struct vkd3d_shader_register *reg, uint32_t swizzle, uint32_t write_mask)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(write_mask);
     uint64_t values[VKD3D_DVEC2_SIZE] = {0};
     unsigned int i, j;
 
@@ -3680,14 +3724,14 @@ static uint32_t spirv_compiler_emit_load_constant64(struct spirv_compiler *compi
     if (reg->dimension == VSIR_DIMENSION_SCALAR)
     {
         for (i = 0; i < component_count; ++i)
-            values[i] = *reg->u.immconst_uint64;
+            values[i] = *reg->u.immconst_u64;
     }
     else
     {
         for (i = 0, j = 0; i < VKD3D_DVEC2_SIZE; ++i)
         {
             if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
-                values[j++] = reg->u.immconst_uint64[vkd3d_swizzle_get_component64(swizzle, i)];
+                values[j++] = reg->u.immconst_u64[vsir_swizzle_get_component64(swizzle, i)];
         }
     }
 
@@ -3696,9 +3740,9 @@ static uint32_t spirv_compiler_emit_load_constant64(struct spirv_compiler *compi
 }
 
 static uint32_t spirv_compiler_emit_load_undef(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD write_mask)
+        const struct vkd3d_shader_register *reg, uint32_t write_mask)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(write_mask);
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id;
 
@@ -3709,28 +3753,28 @@ static uint32_t spirv_compiler_emit_load_undef(struct spirv_compiler *compiler,
 }
 
 static uint32_t spirv_compiler_emit_load_scalar(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD swizzle, DWORD write_mask,
+        const struct vkd3d_shader_register *reg, uint32_t swizzle, uint32_t write_mask,
         const struct vkd3d_shader_register_info *reg_info)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, ptr_type_id, index, reg_id, val_id;
     unsigned int component_idx, reg_component_count;
     enum vkd3d_shader_component_type component_type;
-    unsigned int skipped_component_mask;
+    uint32_t skipped_component_mask;
 
     assert(!register_is_constant_or_undef(reg));
-    assert(vkd3d_write_mask_component_count(write_mask) == 1);
+    assert(vsir_write_mask_component_count(write_mask) == 1);
 
-    component_idx = vkd3d_write_mask_get_component_idx(write_mask);
-    component_idx = vkd3d_swizzle_get_component(swizzle, component_idx);
+    component_idx = vsir_write_mask_get_component_idx(write_mask);
+    component_idx = vsir_swizzle_get_component(swizzle, component_idx);
     skipped_component_mask = ~reg_info->write_mask & ((VKD3DSP_WRITEMASK_0 << component_idx) - 1);
     if (skipped_component_mask)
-        component_idx -= vkd3d_write_mask_component_count(skipped_component_mask);
+        component_idx -= vsir_write_mask_component_count(skipped_component_mask);
     component_type = vkd3d_component_type_from_data_type(reg->data_type);
 
-    reg_component_count = vkd3d_write_mask_component_count(reg_info->write_mask);
+    reg_component_count = vsir_write_mask_component_count(reg_info->write_mask);
 
-    if (component_idx >= vkd3d_write_mask_component_count(reg_info->write_mask))
+    if (component_idx >= vsir_write_mask_component_count(reg_info->write_mask))
     {
         ERR("Invalid component_idx %u for register %#x, %u (write_mask %#x).\n",
                 component_idx, reg->type, reg->idx[0].offset, reg_info->write_mask);
@@ -3756,6 +3800,69 @@ static uint32_t spirv_compiler_emit_load_scalar(struct spirv_compiler *compiler,
     return val_id;
 }
 
+static uint32_t spirv_compiler_emit_constant_array(struct spirv_compiler *compiler,
+        const struct vkd3d_shader_immediate_constant_buffer *icb, uint32_t *type_id_out)
+{
+    uint32_t *elements, elem_type_id, length_id, type_id, const_id;
+    struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    enum vkd3d_shader_component_type component_type;
+    unsigned int i, element_count, component_count;
+
+    element_count = icb->element_count;
+
+    component_type = vkd3d_component_type_from_data_type(icb->data_type);
+    component_count = icb->component_count;
+    elem_type_id = vkd3d_spirv_get_type_id_for_data_type(builder, icb->data_type, component_count);
+    length_id = spirv_compiler_get_constant_uint(compiler, element_count);
+    type_id = vkd3d_spirv_get_op_type_array(builder, elem_type_id, length_id);
+
+    if (type_id_out)
+        *type_id_out = type_id;
+
+    if (icb->is_null)
+    {
+        /* All values are null. Workgroup memory initialisers require OpConstantNull. */
+        return vkd3d_spirv_get_op_constant_null(builder, type_id);
+    }
+
+    if (!(elements = vkd3d_calloc(element_count, sizeof(*elements))))
+    {
+        ERR("Failed to allocate %u elements.", element_count);
+        spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_OUT_OF_MEMORY,
+                "Failed to allocate %u constant array elements.", element_count);
+        return 0;
+    }
+
+    switch (icb->data_type)
+    {
+        case VKD3D_DATA_FLOAT:
+        case VKD3D_DATA_INT:
+        case VKD3D_DATA_UINT:
+            for (i = 0; i < element_count; ++i)
+                elements[i] = spirv_compiler_get_constant(compiler, component_type, component_count,
+                        &icb->data[component_count * i]);
+            break;
+        case VKD3D_DATA_DOUBLE:
+        case VKD3D_DATA_UINT64:
+        {
+            uint64_t *data = (uint64_t *)icb->data;
+            for (i = 0; i < element_count; ++i)
+                elements[i] = spirv_compiler_get_constant64(compiler, component_type, component_count,
+                        &data[component_count * i]);
+            break;
+        }
+        default:
+            FIXME("Unhandled data type %u.\n", icb->data_type);
+            spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_INVALID_TYPE,
+                    "Immediate constant buffer data type %u is unhandled.", icb->data_type);
+            break;
+    }
+
+    const_id = vkd3d_spirv_build_op_constant_composite(builder, type_id, elements, element_count);
+    vkd3d_free(elements);
+    return const_id;
+}
+
 static const struct ssa_register_info *spirv_compiler_get_ssa_register_info(const struct spirv_compiler *compiler,
         const struct vkd3d_shader_register *reg)
 {
@@ -3775,7 +3882,7 @@ static void spirv_compiler_set_ssa_register_info(const struct spirv_compiler *co
 
 static uint32_t spirv_compiler_emit_load_ssa_reg(struct spirv_compiler *compiler,
         const struct vkd3d_shader_register *reg, enum vkd3d_shader_component_type component_type,
-        unsigned int swizzle)
+        uint32_t swizzle)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     enum vkd3d_shader_component_type reg_component_type;
@@ -3801,19 +3908,19 @@ static uint32_t spirv_compiler_emit_load_ssa_reg(struct spirv_compiler *compiler
     }
 
     type_id = vkd3d_spirv_get_type_id(builder, component_type, 1);
-    component_idx = vkd3d_swizzle_get_component(swizzle, 0);
+    component_idx = vsir_swizzle_get_component(swizzle, 0);
     return vkd3d_spirv_build_op_composite_extract1(builder, type_id, val_id, component_idx);
 }
 
 static uint32_t spirv_compiler_emit_load_reg(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD swizzle, DWORD write_mask)
+        const struct vkd3d_shader_register *reg, DWORD swizzle, uint32_t write_mask)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     enum vkd3d_shader_component_type component_type;
     struct vkd3d_shader_register_info reg_info;
     unsigned int component_count;
-    unsigned int write_mask32;
     uint32_t type_id, val_id;
+    uint32_t write_mask32;
 
     if (reg->type == VKD3DSPR_IMMCONST)
         return spirv_compiler_emit_load_constant(compiler, reg, swizzle, write_mask);
@@ -3822,7 +3929,7 @@ static uint32_t spirv_compiler_emit_load_reg(struct spirv_compiler *compiler,
     else if (reg->type == VKD3DSPR_UNDEF)
         return spirv_compiler_emit_load_undef(compiler, reg, write_mask);
 
-    component_count = vkd3d_write_mask_component_count(write_mask);
+    component_count = vsir_write_mask_component_count(write_mask);
     component_type = vkd3d_component_type_from_data_type(reg->data_type);
 
     if (reg->type == VKD3DSPR_SSA)
@@ -3836,21 +3943,21 @@ static uint32_t spirv_compiler_emit_load_reg(struct spirv_compiler *compiler,
     assert(reg_info.component_type != VKD3D_SHADER_COMPONENT_DOUBLE);
     spirv_compiler_emit_dereference_register(compiler, reg, &reg_info);
 
-    write_mask32 = (reg->data_type == VKD3D_DATA_DOUBLE) ? vkd3d_write_mask_32_from_64(write_mask) : write_mask;
+    write_mask32 = data_type_is_64_bit(reg->data_type) ? vsir_write_mask_32_from_64(write_mask) : write_mask;
 
     /* Intermediate value (no storage class). */
     if (reg_info.storage_class == SpvStorageClassMax)
     {
         val_id = reg_info.id;
     }
-    else if (vkd3d_write_mask_component_count(write_mask32) == 1)
+    else if (vsir_write_mask_component_count(write_mask32) == 1)
     {
         return spirv_compiler_emit_load_scalar(compiler, reg, swizzle, write_mask, &reg_info);
     }
     else
     {
         type_id = vkd3d_spirv_get_type_id(builder,
-                reg_info.component_type, vkd3d_write_mask_component_count(reg_info.write_mask));
+                reg_info.component_type, vsir_write_mask_component_count(reg_info.write_mask));
         val_id = vkd3d_spirv_build_op_load(builder, type_id, reg_info.id, SpvMemoryAccessMaskNone);
     }
 
@@ -3904,7 +4011,7 @@ static uint32_t spirv_compiler_emit_neg(struct spirv_compiler *compiler,
     type_id = spirv_compiler_get_type_id_for_reg(compiler, reg, write_mask);
     if (reg->data_type == VKD3D_DATA_FLOAT || reg->data_type == VKD3D_DATA_DOUBLE)
         return vkd3d_spirv_build_op_fnegate(builder, type_id, val_id);
-    else if (reg->data_type == VKD3D_DATA_INT || reg->data_type == VKD3D_DATA_UINT)
+    else if (data_type_is_integer(reg->data_type))
         return vkd3d_spirv_build_op_snegate(builder, type_id, val_id);
 
     FIXME("Unhandled data type %#x.\n", reg->data_type);
@@ -3953,19 +4060,19 @@ static uint32_t spirv_compiler_emit_load_src_with_type(struct spirv_compiler *co
 }
 
 static void spirv_compiler_emit_store_scalar(struct spirv_compiler *compiler,
-        uint32_t dst_id, unsigned int dst_write_mask, enum vkd3d_shader_component_type component_type,
-        SpvStorageClass storage_class, unsigned int write_mask, uint32_t val_id)
+        uint32_t dst_id, uint32_t dst_write_mask, enum vkd3d_shader_component_type component_type,
+        SpvStorageClass storage_class, uint32_t write_mask, uint32_t val_id)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, ptr_type_id, index;
     unsigned int component_idx;
 
-    if (vkd3d_write_mask_component_count(dst_write_mask) > 1)
+    if (vsir_write_mask_component_count(dst_write_mask) > 1)
     {
         type_id = vkd3d_spirv_get_type_id(builder, component_type, 1);
         ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, storage_class, type_id);
-        component_idx = vkd3d_write_mask_get_component_idx(write_mask);
-        component_idx -= vkd3d_write_mask_get_component_idx(dst_write_mask);
+        component_idx = vsir_write_mask_get_component_idx(write_mask);
+        component_idx -= vsir_write_mask_get_component_idx(dst_write_mask);
         index = spirv_compiler_get_constant_uint(compiler, component_idx);
         dst_id = vkd3d_spirv_build_op_in_bounds_access_chain1(builder, ptr_type_id, dst_id, index);
     }
@@ -3974,8 +4081,8 @@ static void spirv_compiler_emit_store_scalar(struct spirv_compiler *compiler,
 }
 
 static void spirv_compiler_emit_store(struct spirv_compiler *compiler,
-        uint32_t dst_id, unsigned int dst_write_mask, enum vkd3d_shader_component_type component_type,
-        SpvStorageClass storage_class, unsigned int write_mask, uint32_t val_id)
+        uint32_t dst_id, uint32_t dst_write_mask, enum vkd3d_shader_component_type component_type,
+        SpvStorageClass storage_class, uint32_t write_mask, uint32_t val_id)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     unsigned int component_count, dst_component_count;
@@ -3985,14 +4092,14 @@ static void spirv_compiler_emit_store(struct spirv_compiler *compiler,
 
     assert(write_mask);
 
-    component_count = vkd3d_write_mask_component_count(write_mask);
-    dst_component_count = vkd3d_write_mask_component_count(dst_write_mask);
+    component_count = vsir_write_mask_component_count(write_mask);
+    dst_component_count = vsir_write_mask_component_count(dst_write_mask);
 
     if (dst_component_count == 1 && component_count != 1)
     {
         type_id = vkd3d_spirv_get_type_id(builder, component_type, 1);
         val_id = vkd3d_spirv_build_op_composite_extract1(builder, type_id, val_id,
-                vkd3d_write_mask_get_component_idx(dst_write_mask));
+                vsir_write_mask_get_component_idx(dst_write_mask));
         write_mask &= dst_write_mask;
         component_count = 1;
     }
@@ -4029,12 +4136,12 @@ static void spirv_compiler_emit_store(struct spirv_compiler *compiler,
 }
 
 static void spirv_compiler_emit_store_reg(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, unsigned int write_mask, uint32_t val_id)
+        const struct vkd3d_shader_register *reg, uint32_t write_mask, uint32_t val_id)
 {
     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 src_write_mask = write_mask;
     uint32_t type_id;
 
     assert(!register_is_constant_or_undef(reg));
@@ -4052,10 +4159,10 @@ static void spirv_compiler_emit_store_reg(struct spirv_compiler *compiler,
     component_type = vkd3d_component_type_from_data_type(reg->data_type);
     if (component_type != reg_info.component_type)
     {
-        if (reg->data_type == VKD3D_DATA_DOUBLE)
-            src_write_mask = vkd3d_write_mask_32_from_64(write_mask);
+        if (data_type_is_64_bit(reg->data_type))
+            src_write_mask = vsir_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));
+                vsir_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;
     }
@@ -4065,9 +4172,9 @@ static void spirv_compiler_emit_store_reg(struct spirv_compiler *compiler,
 }
 
 static uint32_t spirv_compiler_emit_sat(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg, DWORD write_mask, uint32_t val_id)
+        const struct vkd3d_shader_register *reg, uint32_t write_mask, uint32_t val_id)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(write_mask);
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, zero_id, one_id;
 
@@ -4118,7 +4225,7 @@ static void spirv_compiler_emit_store_dst_components(struct spirv_compiler *comp
         const struct vkd3d_shader_dst_param *dst, enum vkd3d_shader_component_type component_type,
         uint32_t *component_ids)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(dst->write_mask);
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, val_id;
 
@@ -4137,16 +4244,16 @@ static void spirv_compiler_emit_store_dst_components(struct spirv_compiler *comp
 
 static void spirv_compiler_emit_store_dst_scalar(struct spirv_compiler *compiler,
         const struct vkd3d_shader_dst_param *dst, uint32_t val_id,
-        enum vkd3d_shader_component_type component_type, DWORD swizzle)
+        enum vkd3d_shader_component_type component_type, uint32_t swizzle)
 {
-    unsigned int component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    unsigned int component_count = vsir_write_mask_component_count(dst->write_mask);
     uint32_t component_ids[VKD3D_VEC4_SIZE];
     unsigned int component_idx, i;
 
-    component_idx = vkd3d_write_mask_get_component_idx(dst->write_mask);
+    component_idx = vsir_write_mask_get_component_idx(dst->write_mask);
     for (i = 0; i < component_count; ++i)
     {
-        if (vkd3d_swizzle_get_component(swizzle, component_idx + i))
+        if (vsir_swizzle_get_component(swizzle, component_idx + i))
             ERR("Invalid swizzle %#x for scalar value, write mask %#x.\n", swizzle, dst->write_mask);
 
         component_ids[i] = val_id;
@@ -4191,14 +4298,18 @@ static void spirv_compiler_decorate_builtin(struct spirv_compiler *compiler,
 }
 
 static void spirv_compiler_emit_interpolation_decorations(struct spirv_compiler *compiler,
-        uint32_t id, enum vkd3d_shader_interpolation_mode mode)
+        enum vkd3d_shader_component_type component_type, uint32_t id, enum vkd3d_shader_interpolation_mode mode)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
 
     switch (mode)
     {
         case VKD3DSIM_NONE:
-            break;
+            /* VUID-StandaloneSpirv-Flat-04744: integer or double types must be
+             * decorated 'Flat' for fragment shaders. */
+            if (compiler->shader_type != VKD3D_SHADER_TYPE_PIXEL || component_type == VKD3D_SHADER_COMPONENT_FLOAT)
+                break;
+            /* fall through */
         case VKD3DSIM_CONSTANT:
             vkd3d_spirv_build_op_decorate(builder, id, SpvDecorationFlat, NULL, 0);
             break;
@@ -4226,7 +4337,8 @@ static void spirv_compiler_emit_interpolation_decorations(struct spirv_compiler
 }
 
 static uint32_t spirv_compiler_emit_int_to_bool(struct spirv_compiler *compiler,
-        enum vkd3d_shader_conditional_op condition, unsigned int component_count, uint32_t val_id)
+        enum vkd3d_shader_conditional_op condition, enum vkd3d_data_type data_type,
+        unsigned int component_count, uint32_t val_id)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id;
@@ -4237,7 +4349,9 @@ static uint32_t spirv_compiler_emit_int_to_bool(struct spirv_compiler *compiler,
     type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_BOOL, component_count);
     op = condition & VKD3D_SHADER_CONDITIONAL_OP_Z ? SpvOpIEqual : SpvOpINotEqual;
     return vkd3d_spirv_build_op_tr2(builder, &builder->function_stream, op, type_id, val_id,
-            spirv_compiler_get_constant_uint_vector(compiler, 0, component_count));
+            data_type == VKD3D_DATA_UINT64
+            ? spirv_compiler_get_constant_uint64_vector(compiler, 0, component_count)
+            : spirv_compiler_get_constant_uint_vector(compiler, 0, component_count));
 }
 
 static uint32_t spirv_compiler_emit_bool_to_int(struct spirv_compiler *compiler,
@@ -4252,6 +4366,19 @@ static uint32_t spirv_compiler_emit_bool_to_int(struct spirv_compiler *compiler,
     return vkd3d_spirv_build_op_select(builder, type_id, val_id, true_id, false_id);
 }
 
+static uint32_t spirv_compiler_emit_bool_to_int64(struct spirv_compiler *compiler,
+        unsigned int component_count, uint32_t val_id, bool signedness)
+{
+    struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    uint32_t type_id, true_id, false_id;
+
+    true_id = spirv_compiler_get_constant_uint64_vector(compiler, signedness ? UINT64_MAX : 1,
+            component_count);
+    false_id = spirv_compiler_get_constant_uint64_vector(compiler, 0, component_count);
+    type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_UINT64, component_count);
+    return vkd3d_spirv_build_op_select(builder, type_id, val_id, true_id, false_id);
+}
+
 static uint32_t spirv_compiler_emit_bool_to_float(struct spirv_compiler *compiler,
         unsigned int component_count, uint32_t val_id, bool signedness)
 {
@@ -4422,9 +4549,9 @@ vkd3d_register_builtins[] =
 };
 
 static void spirv_compiler_emit_register_execution_mode(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_register *reg)
+        enum vkd3d_shader_register_type type)
 {
-    switch (reg->type)
+    switch (type)
     {
         case VKD3DSPR_DEPTHOUTGE:
             spirv_compiler_emit_execution_mode(compiler, SpvExecutionModeDepthGreater, NULL, 0);
@@ -4437,9 +4564,9 @@ static void spirv_compiler_emit_register_execution_mode(struct spirv_compiler *c
                     VKD3D_SHADER_SPIRV_EXTENSION_EXT_STENCIL_EXPORT))
             {
                 FIXME("The target environment does not support stencil export.\n");
-                spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_STENCIL_EXPORT_UNSUPPORTED,
-                        "Cannot export stencil reference value for register id %u. "
-                        "The target environment does not support stencil export.", reg->idx[0].offset);
+                spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_UNSUPPORTED_FEATURE,
+                        "Cannot export stencil reference value. "
+                        "The target environment does not support stencil export.");
             }
             vkd3d_spirv_enable_capability(&compiler->spirv_builder, SpvCapabilityStencilExportEXT);
             spirv_compiler_emit_execution_mode(compiler, SpvExecutionModeStencilRefReplacingEXT, NULL, 0);
@@ -4674,17 +4801,16 @@ static unsigned int shader_signature_next_location(const struct shader_signature
 }
 
 static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_dst_param *dst)
+        enum vkd3d_shader_register_type reg_type, unsigned int element_idx)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
-    const struct vkd3d_shader_register *reg = &dst->reg;
     unsigned int component_idx, input_component_count;
     const struct signature_element *signature_element;
     const struct shader_signature *shader_signature;
     enum vkd3d_shader_component_type component_type;
     const struct vkd3d_spirv_builtin *builtin;
     enum vkd3d_shader_sysval_semantic sysval;
-    unsigned int write_mask, reg_write_mask;
+    uint32_t write_mask, reg_write_mask;
     struct vkd3d_symbol *symbol = NULL;
     uint32_t val_id, input_id, var_id;
     uint32_t type_id, float_type_id;
@@ -4693,26 +4819,21 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
     struct rb_entry *entry = NULL;
     bool use_private_var = false;
     unsigned int array_sizes[2];
-    unsigned int element_idx;
-
-    assert(!reg->idx_count || !reg->idx[0].rel_addr);
-    assert(reg->idx_count < 2 || !reg->idx[1].rel_addr);
 
-    shader_signature = reg->type == VKD3DSPR_PATCHCONST
+    shader_signature = reg_type == VKD3DSPR_PATCHCONST
             ? &compiler->patch_constant_signature : &compiler->input_signature;
 
-    element_idx = reg->idx[reg->idx_count - 1].offset;
     signature_element = &shader_signature->elements[element_idx];
     sysval = signature_element->sysval_semantic;
     /* The Vulkan spec does not explicitly forbid passing varyings from the
      * TCS to the TES via builtins. However, Mesa doesn't seem to handle it
      * well, and we don't actually need them to be in builtins. */
-    if (compiler->shader_type == VKD3D_SHADER_TYPE_DOMAIN && reg->type != VKD3DSPR_PATCHCONST)
+    if (compiler->shader_type == VKD3D_SHADER_TYPE_DOMAIN && reg_type != VKD3DSPR_PATCHCONST)
         sysval = VKD3D_SHADER_SV_NONE;
 
     builtin = get_spirv_builtin_for_sysval(compiler, sysval);
 
-    array_sizes[0] = (reg->type == VKD3DSPR_PATCHCONST ? 0 : compiler->input_control_point_count);
+    array_sizes[0] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->input_control_point_count);
     array_sizes[1] = signature_element->register_count;
     if (array_sizes[1] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic)
             && (!vsir_sysval_semantic_is_clip_cull(signature_element->sysval_semantic) || array_sizes[0]))
@@ -4731,8 +4852,8 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
     else
     {
         component_type = signature_element->component_type;
-        input_component_count = vkd3d_write_mask_component_count(signature_element->mask);
-        component_idx = vkd3d_write_mask_get_component_idx(signature_element->mask);
+        input_component_count = vsir_write_mask_component_count(signature_element->mask);
+        component_idx = vsir_write_mask_get_component_idx(signature_element->mask);
     }
 
     if (needs_private_io_variable(builtin))
@@ -4742,13 +4863,13 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
     }
     else
     {
-        component_idx = vkd3d_write_mask_get_component_idx(write_mask);
+        component_idx = vsir_write_mask_get_component_idx(write_mask);
         reg_write_mask = write_mask >> component_idx;
     }
 
     storage_class = SpvStorageClassInput;
 
-    vkd3d_symbol_make_register(&reg_symbol, reg);
+    vkd3d_symbol_make_io(&reg_symbol, reg_type, element_idx);
 
     if ((entry = rb_get(&compiler->symbol_table, &reg_symbol)))
     {
@@ -4756,7 +4877,7 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
          * duplicate declarations are: a single register split into multiple declarations having
          * different components, which should have been merged, and declarations in one phase
          * being repeated in another (i.e. vcp/vocp), which should have been deleted. */
-        if (reg->type != VKD3DSPR_INPUT || !is_in_fork_or_join_phase(compiler))
+        if (reg_type != VKD3DSPR_INPUT || !is_in_fork_or_join_phase(compiler))
             FIXME("Duplicate input definition found.\n");
         symbol = RB_ENTRY_VALUE(entry, struct vkd3d_symbol, entry);
         return symbol->id;
@@ -4765,7 +4886,7 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
     if (builtin)
     {
         input_id = spirv_compiler_emit_builtin_variable_v(compiler, builtin, storage_class, array_sizes, 2);
-        if (reg->type == VKD3DSPR_PATCHCONST)
+        if (reg_type == VKD3DSPR_PATCHCONST)
             vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0);
     }
     else
@@ -4775,7 +4896,7 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
         input_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream,
                 storage_class, component_type, input_component_count, array_sizes, 2);
         vkd3d_spirv_add_iface_variable(builder, input_id);
-        if (reg->type == VKD3DSPR_PATCHCONST)
+        if (reg_type == VKD3DSPR_PATCHCONST)
         {
             vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0);
             location += shader_signature_next_location(&compiler->input_signature);
@@ -4784,7 +4905,8 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
         if (component_idx)
             vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationComponent, component_idx);
 
-        spirv_compiler_emit_interpolation_decorations(compiler, input_id, signature_element->interpolation_mode);
+        spirv_compiler_emit_interpolation_decorations(compiler, component_type, input_id,
+                signature_element->interpolation_mode);
     }
 
     var_id = input_id;
@@ -4802,12 +4924,14 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
     assert(!builtin || !builtin->spirv_array_size || use_private_var || array_sizes[0] || array_sizes[1]);
     spirv_compiler_put_symbol(compiler, &reg_symbol);
 
-    spirv_compiler_emit_register_debug_name(builder, var_id, reg);
+    vkd3d_spirv_build_op_name(builder, var_id, reg_type == VKD3DSPR_PATCHCONST ? "vpc%u" : "v%u", element_idx);
 
     if (use_private_var)
     {
-        struct vkd3d_shader_register dst_reg = *reg;
-        dst_reg.data_type = VKD3D_DATA_FLOAT;
+        struct vkd3d_shader_register dst_reg;
+
+        vsir_register_init(&dst_reg, reg_type, VKD3D_DATA_FLOAT, 1);
+        dst_reg.idx[0].offset = element_idx;
 
         type_id = vkd3d_spirv_get_type_id(builder, component_type, input_component_count);
 
@@ -4824,9 +4948,9 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler,
 
         val_id = spirv_compiler_emit_swizzle(compiler, val_id,
                 vkd3d_write_mask_from_component_count(input_component_count),
-                VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_SHADER_NO_SWIZZLE, dst->write_mask >> component_idx);
+                VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_SHADER_NO_SWIZZLE, signature_element->mask >> component_idx);
 
-        spirv_compiler_emit_store_reg(compiler, &dst_reg, dst->write_mask, val_id);
+        spirv_compiler_emit_store_reg(compiler, &dst_reg, signature_element->mask, val_id);
     }
 
     return input_id;
@@ -4896,7 +5020,7 @@ static void calculate_clip_or_cull_distance_mask(const struct signature_element
         return;
     }
 
-    write_mask = e->mask >> vkd3d_write_mask_get_component_idx(e->mask);
+    write_mask = e->mask >> vsir_write_mask_get_component_idx(e->mask);
     *mask |= (write_mask & VKD3DSP_WRITEMASK_ALL) << (VKD3D_VEC4_SIZE * e->semantic_index);
 }
 
@@ -4995,7 +5119,7 @@ static void spirv_compiler_emit_output_register(struct spirv_compiler *compiler,
             SpvStorageClassOutput, builtin->component_type, write_mask);
     reg_symbol.info.reg.is_aggregate = builtin->spirv_array_size;
     spirv_compiler_put_symbol(compiler, &reg_symbol);
-    spirv_compiler_emit_register_execution_mode(compiler, reg);
+    spirv_compiler_emit_register_execution_mode(compiler, reg->type);
     spirv_compiler_emit_register_debug_name(builder, output_id, reg);
 }
 
@@ -5024,46 +5148,44 @@ static uint32_t spirv_compiler_emit_shader_phase_builtin_variable(struct spirv_c
     return id;
 }
 
-static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const struct vkd3d_shader_dst_param *dst)
+static void spirv_compiler_emit_output(struct spirv_compiler *compiler,
+        enum vkd3d_shader_register_type reg_type, unsigned int element_idx)
 {
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
-    const struct vkd3d_shader_register *reg = &dst->reg;
     unsigned int component_idx, output_component_count;
     const struct signature_element *signature_element;
     enum vkd3d_shader_component_type component_type;
     const struct shader_signature *shader_signature;
     const struct vkd3d_spirv_builtin *builtin;
     enum vkd3d_shader_sysval_semantic sysval;
-    unsigned int write_mask, reg_write_mask;
+    uint32_t write_mask, reg_write_mask;
     bool use_private_variable = false;
     struct vkd3d_symbol reg_symbol;
     SpvStorageClass storage_class;
     unsigned int array_sizes[2];
-    unsigned int element_idx;
     bool is_patch_constant;
     uint32_t id, var_id;
 
-    is_patch_constant = is_in_fork_or_join_phase(compiler);
+    is_patch_constant = (reg_type == VKD3DSPR_PATCHCONST);
 
     shader_signature = is_patch_constant ? &compiler->patch_constant_signature : &compiler->output_signature;
 
-    element_idx = reg->idx[reg->idx_count - 1].offset;
     signature_element = &shader_signature->elements[element_idx];
     sysval = signature_element->sysval_semantic;
     /* Don't use builtins for TCS -> TES varyings. See spirv_compiler_emit_input(). */
     if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL && !is_patch_constant)
         sysval = VKD3D_SHADER_SV_NONE;
-    array_sizes[0] = (reg->type == VKD3DSPR_PATCHCONST ? 0 : compiler->output_control_point_count);
+    array_sizes[0] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->output_control_point_count);
     array_sizes[1] = signature_element->register_count;
     if (array_sizes[1] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic))
         array_sizes[1] = 0;
 
-    builtin = vkd3d_get_spirv_builtin(compiler, dst->reg.type, sysval);
+    builtin = vkd3d_get_spirv_builtin(compiler, reg_type, sysval);
 
     write_mask = signature_element->mask;
 
-    component_idx = vkd3d_write_mask_get_component_idx(write_mask);
-    output_component_count = vkd3d_write_mask_component_count(write_mask);
+    component_idx = vsir_write_mask_get_component_idx(write_mask);
+    output_component_count = vsir_write_mask_component_count(write_mask);
     if (builtin)
     {
         component_type = builtin->component_type;
@@ -5077,15 +5199,18 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const st
 
     storage_class = SpvStorageClassOutput;
 
-    if (get_shader_output_swizzle(compiler, signature_element->register_index) != VKD3D_SHADER_NO_SWIZZLE
-            || (compiler->output_info[element_idx].id && compiler->output_info[element_idx].array_element_mask)
-            || needs_private_io_variable(builtin))
+    if (needs_private_io_variable(builtin))
+        use_private_variable = true;
+
+    if (!is_patch_constant
+            && (get_shader_output_swizzle(compiler, signature_element->register_index) != VKD3D_SHADER_NO_SWIZZLE
+            || (compiler->output_info[element_idx].id && compiler->output_info[element_idx].array_element_mask)))
     {
         use_private_variable = true;
     }
 
     reg_write_mask = write_mask >> component_idx;
-    vkd3d_symbol_make_register(&reg_symbol, reg);
+    vkd3d_symbol_make_io(&reg_symbol, reg_type, element_idx);
 
     if (rb_get(&compiler->symbol_table, &reg_symbol))
     {
@@ -5094,7 +5219,7 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const st
         return;
     }
 
-    if (compiler->output_info[element_idx].id)
+    if (!is_patch_constant && compiler->output_info[element_idx].id)
     {
         id = compiler->output_info[element_idx].id;
     }
@@ -5105,7 +5230,7 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const st
         else
             id = spirv_compiler_emit_builtin_variable_v(compiler, builtin, storage_class, array_sizes, 2);
 
-        spirv_compiler_emit_register_execution_mode(compiler, &dst->reg);
+        spirv_compiler_emit_register_execution_mode(compiler, reg_type);
     }
     else if (signature_element->target_location == SIGNATURE_TARGET_LOCATION_UNUSED)
     {
@@ -5146,8 +5271,11 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const st
 
     spirv_compiler_decorate_xfb_output(compiler, id, output_component_count, signature_element);
 
-    compiler->output_info[element_idx].id = id;
-    compiler->output_info[element_idx].component_type = component_type;
+    if (!is_patch_constant)
+    {
+        compiler->output_info[element_idx].id = id;
+        compiler->output_info[element_idx].component_type = component_type;
+    }
 
     var_id = id;
     if (use_private_variable)
@@ -5165,8 +5293,7 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, const st
 
     spirv_compiler_put_symbol(compiler, &reg_symbol);
 
-    if (!is_patch_constant)
-        spirv_compiler_emit_register_debug_name(builder, var_id, reg);
+    vkd3d_spirv_build_op_name(builder, var_id, reg_type == VKD3DSPR_PATCHCONST ? "vpc%u" : "o%u", element_idx);
 
     if (use_private_variable)
     {
@@ -5198,9 +5325,9 @@ static uint32_t spirv_compiler_get_output_array_index(struct spirv_compiler *com
 static void spirv_compiler_emit_store_shader_output(struct spirv_compiler *compiler,
         const struct shader_signature *signature, const struct signature_element *output,
         const struct vkd3d_shader_output_info *output_info,
-        uint32_t output_index_id, uint32_t val_id, unsigned int write_mask)
+        uint32_t output_index_id, uint32_t val_id, uint32_t write_mask)
 {
-    unsigned int dst_write_mask, use_mask, uninit_mask, swizzle, mask;
+    uint32_t dst_write_mask, use_mask, uninit_mask, swizzle, mask;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     uint32_t type_id, zero_id, ptr_type_id, chain_id, object_id;
     const struct signature_element *element;
@@ -5222,7 +5349,7 @@ static void spirv_compiler_emit_store_shader_output(struct spirv_compiler *compi
             use_mask |= element->used_mask;
         }
     }
-    index = vkd3d_write_mask_get_component_idx(output->mask);
+    index = vsir_write_mask_get_component_idx(output->mask);
     dst_write_mask >>= index;
     use_mask >>= index;
     write_mask &= dst_write_mask;
@@ -5246,7 +5373,7 @@ static void spirv_compiler_emit_store_shader_output(struct spirv_compiler *compi
                 output_info->component_type, VKD3D_VEC4_SIZE, 0);
         val_id = spirv_compiler_emit_vector_shuffle(compiler,
                 zero_id, val_id, swizzle, uninit_mask, output_info->component_type,
-                vkd3d_write_mask_component_count(write_mask));
+                vsir_write_mask_component_count(write_mask));
     }
     else
     {
@@ -5258,7 +5385,7 @@ static void spirv_compiler_emit_store_shader_output(struct spirv_compiler *compi
     if (output_index_id)
     {
         type_id = vkd3d_spirv_get_type_id(builder,
-                output_info->component_type, vkd3d_write_mask_component_count(dst_write_mask));
+                output_info->component_type, vsir_write_mask_component_count(dst_write_mask));
         ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassOutput, type_id);
         output_id = vkd3d_spirv_build_op_access_chain1(builder, ptr_type_id, output_id, output_index_id);
     }
@@ -5444,6 +5571,21 @@ static void spirv_compiler_emit_dcl_global_flags(struct spirv_compiler *compiler
         flags &= ~(VKD3DSGF_ENABLE_DOUBLE_PRECISION_FLOAT_OPS | VKD3DSGF_ENABLE_11_1_DOUBLE_EXTENSIONS);
     }
 
+    if (flags & VKD3DSGF_ENABLE_INT64)
+    {
+        if (compiler->features & VKD3D_SHADER_COMPILE_OPTION_FEATURE_INT64)
+        {
+            vkd3d_spirv_enable_capability(&compiler->spirv_builder, SpvCapabilityInt64);
+        }
+        else
+        {
+            WARN("Unsupported 64-bit integer ops.\n");
+            spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_UNSUPPORTED_FEATURE,
+                    "The target environment does not support 64-bit integers.");
+        }
+        flags &= ~VKD3DSGF_ENABLE_INT64;
+    }
+
     if (flags & ~(VKD3DSGF_REFACTORING_ALLOWED | VKD3DSGF_ENABLE_RAW_AND_STRUCTURED_BUFFERS))
         FIXME("Unhandled global flags %#"PRIx64".\n", (uint64_t)flags);
     else
@@ -5493,39 +5635,40 @@ static void spirv_compiler_emit_dcl_indexable_temp(struct spirv_compiler *compil
 {
     const struct vkd3d_shader_indexable_temp *temp = &instruction->declaration.indexable_temp;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    uint32_t id, type_id, length_id, ptr_type_id, init_id = 0;
+    enum vkd3d_shader_component_type component_type;
     struct vkd3d_shader_register reg;
     struct vkd3d_symbol reg_symbol;
+    SpvStorageClass storage_class;
     size_t function_location;
-    uint32_t id;
 
-    if (temp->component_count != 4)
-        FIXME("Unhandled component count %u.\n", temp->component_count);
+    storage_class = SpvStorageClassFunction;
 
     vsir_register_init(&reg, VKD3DSPR_IDXTEMP, VKD3D_DATA_FLOAT, 1);
     reg.idx[0].offset = temp->register_idx;
 
     if (temp->alignment)
         WARN("Ignoring alignment %u.\n", temp->alignment);
-    if (temp->initialiser)
-    {
-        FIXME("Initialisers are not supported.\n");
-        spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_NOT_IMPLEMENTED,
-                "Initialisers for indexable temps are not supported.");
-    }
 
     function_location = spirv_compiler_get_current_function_location(compiler);
     vkd3d_spirv_begin_function_stream_insertion(builder, function_location);
 
-    id = spirv_compiler_emit_array_variable(compiler, &builder->function_stream,
-            SpvStorageClassFunction, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, &temp->register_size, 1);
+    component_type = vkd3d_component_type_from_data_type(temp->data_type);
+    type_id = vkd3d_spirv_get_type_id(builder, component_type, temp->component_count);
+    length_id = spirv_compiler_get_constant_uint(compiler, temp->register_size);
+    type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id);
+    ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, storage_class, type_id);
+    if (temp->initialiser)
+        init_id = spirv_compiler_emit_constant_array(compiler, temp->initialiser, NULL);
+    id = vkd3d_spirv_build_op_variable(builder, &builder->function_stream, ptr_type_id, storage_class, init_id);
 
     spirv_compiler_emit_register_debug_name(builder, id, &reg);
 
     vkd3d_spirv_end_function_stream_insertion(builder);
 
     vkd3d_symbol_make_register(&reg_symbol, &reg);
-    vkd3d_symbol_set_register_info(&reg_symbol, id,
-            SpvStorageClassFunction, VKD3D_SHADER_COMPONENT_FLOAT, VKD3DSP_WRITEMASK_ALL);
+    vkd3d_symbol_set_register_info(&reg_symbol, id, storage_class,
+            component_type, vkd3d_write_mask_from_component_count(temp->component_count));
     spirv_compiler_put_symbol(compiler, &reg_symbol);
 }
 
@@ -5743,34 +5886,24 @@ static void spirv_compiler_emit_dcl_immediate_constant_buffer(struct spirv_compi
         const struct vkd3d_shader_instruction *instruction)
 {
     const struct vkd3d_shader_immediate_constant_buffer *icb = instruction->declaration.icb;
-    uint32_t *elements, length_id, type_id, const_id, ptr_type_id, icb_id;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    uint32_t type_id, const_id, ptr_type_id, icb_id;
     struct vkd3d_shader_register reg;
     struct vkd3d_symbol reg_symbol;
-    unsigned int i;
 
-    assert(icb->data_type == VKD3D_DATA_FLOAT);
-    assert(icb->component_count == VKD3D_VEC4_SIZE);
-
-    if (!(elements = vkd3d_calloc(icb->element_count, sizeof(*elements))))
-        return;
-    for (i = 0; i < icb->element_count; ++i)
-        elements[i] = spirv_compiler_get_constant(compiler,
-                VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, &icb->data[4 * i]);
-    type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE);
-    length_id = spirv_compiler_get_constant_uint(compiler, icb->element_count);
-    type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id);
-    const_id = vkd3d_spirv_build_op_constant_composite(builder, type_id, elements, icb->element_count);
+    const_id = spirv_compiler_emit_constant_array(compiler, icb, &type_id);
     ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassPrivate, type_id);
     icb_id = vkd3d_spirv_build_op_variable(builder, &builder->global_stream,
             ptr_type_id, SpvStorageClassPrivate, const_id);
     vkd3d_spirv_build_op_name(builder, icb_id, "icb");
-    vkd3d_free(elements);
 
-    vsir_register_init(&reg, VKD3DSPR_IMMCONSTBUFFER, VKD3D_DATA_FLOAT, 0);
+    /* Set an index count of 2 so vkd3d_symbol_make_register() uses idx[0] as a buffer id. */
+    vsir_register_init(&reg, VKD3DSPR_IMMCONSTBUFFER, VKD3D_DATA_FLOAT, 2);
+    reg.idx[0].offset = icb->register_idx;
     vkd3d_symbol_make_register(&reg_symbol, &reg);
     vkd3d_symbol_set_register_info(&reg_symbol, icb_id, SpvStorageClassPrivate,
-            VKD3D_SHADER_COMPONENT_FLOAT, VKD3DSP_WRITEMASK_ALL);
+                vkd3d_component_type_from_data_type(icb->data_type),
+                vkd3d_write_mask_from_component_count(icb->component_count));
     spirv_compiler_put_symbol(compiler, &reg_symbol);
 }
 
@@ -6147,37 +6280,22 @@ static void spirv_compiler_emit_dcl_input(struct spirv_compiler *compiler,
 {
     const struct vkd3d_shader_dst_param *dst = &instruction->declaration.dst;
 
-    /* OUTPOINTID is handled in spirv_compiler_emit_hull_shader_builtins(). */
-    if (dst->reg.type == VKD3DSPR_INPUT || dst->reg.type == VKD3DSPR_PATCHCONST)
-        spirv_compiler_emit_input(compiler, dst);
-    else if (dst->reg.type != VKD3DSPR_OUTPOINTID)
+    /* INPUT and PATCHCONST are handled in spirv_compiler_emit_io_declarations().
+     * OUTPOINTID is handled in spirv_compiler_emit_hull_shader_builtins(). */
+    if (dst->reg.type != VKD3DSPR_INPUT && dst->reg.type != VKD3DSPR_PATCHCONST
+            && dst->reg.type != VKD3DSPR_OUTPOINTID)
         spirv_compiler_emit_input_register(compiler, dst);
 }
 
-static void spirv_compiler_emit_dcl_input_sysval(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_instruction *instruction)
-{
-    spirv_compiler_emit_input(compiler, &instruction->declaration.register_semantic.reg);
-}
-
 static void spirv_compiler_emit_dcl_output(struct spirv_compiler *compiler,
         const struct vkd3d_shader_instruction *instruction)
 {
     const struct vkd3d_shader_dst_param *dst = &instruction->declaration.dst;
 
-    if (dst->reg.type == VKD3DSPR_OUTPUT
-            || (is_in_fork_or_join_phase(compiler) && dst->reg.type == VKD3DSPR_PATCHCONST))
-        spirv_compiler_emit_output(compiler, dst);
-    else
+    if (dst->reg.type != VKD3DSPR_OUTPUT && dst->reg.type != VKD3DSPR_PATCHCONST)
         spirv_compiler_emit_output_register(compiler, dst);
 }
 
-static void spirv_compiler_emit_dcl_output_siv(struct spirv_compiler *compiler,
-        const struct vkd3d_shader_instruction *instruction)
-{
-    spirv_compiler_emit_output(compiler, &instruction->declaration.register_semantic.reg);
-}
-
 static void spirv_compiler_emit_dcl_stream(struct spirv_compiler *compiler,
         const struct vkd3d_shader_instruction *instruction)
 {
@@ -6377,7 +6495,6 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile
 
 static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler)
 {
-    const struct shader_signature *signature = &compiler->output_signature;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
 
     if (is_in_control_point_phase(compiler) && compiler->emit_default_control_point_phase)
@@ -6395,7 +6512,6 @@ static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler)
 
         /* Fork and join phases share output registers (patch constants).
          * Control point phase has separate output registers. */
-        memset(compiler->output_info, 0, signature->element_count * sizeof(*compiler->output_info));
         memset(compiler->private_output_variable, 0, sizeof(compiler->private_output_variable));
         memset(compiler->private_output_variable_write_mask, 0, sizeof(compiler->private_output_variable_write_mask));
     }
@@ -6438,15 +6554,12 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile
 {
     const struct shader_signature *output_signature = &compiler->output_signature;
     const struct shader_signature *input_signature = &compiler->input_signature;
+    uint32_t type_id, output_ptr_type_id, input_id, dst_id, invocation_id;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
     enum vkd3d_shader_component_type component_type;
     struct vkd3d_shader_src_param invocation;
     struct vkd3d_shader_register input_reg;
-    uint32_t type_id, output_ptr_type_id;
-    uint32_t input_id, output_id, dst_id;
     unsigned int component_count;
-    unsigned int array_sizes[2];
-    uint32_t invocation_id;
     unsigned int i;
 
     invocation_id = spirv_compiler_emit_load_invocation_id(compiler);
@@ -6466,6 +6579,8 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile
     {
         const struct signature_element *output = &output_signature->elements[i];
         const struct signature_element *input = &input_signature->elements[i];
+        struct vkd3d_shader_register_info output_reg_info;
+        struct vkd3d_shader_register output_reg;
 
         assert(input->mask == output->mask);
         assert(input->component_type == output->component_type);
@@ -6473,22 +6588,16 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile
         input_reg.idx[1].offset = i;
         input_id = spirv_compiler_get_register_id(compiler, &input_reg);
 
+        vsir_register_init(&output_reg, VKD3DSPR_OUTPUT, VKD3D_DATA_FLOAT, 1);
+        output_reg.idx[0].offset = i;
+        spirv_compiler_get_register_info(compiler, &output_reg, &output_reg_info);
+
         component_type = output->component_type;
-        component_count = vkd3d_write_mask_component_count(output->mask);
+        component_count = vsir_write_mask_component_count(output->mask);
         type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count);
-        if ((array_sizes[0] = (input->register_count > 1) ? input->register_count : 0))
-            type_id = vkd3d_spirv_get_op_type_array(builder, type_id, spirv_compiler_get_constant_uint(compiler,
-                    array_sizes[0]));
-
-        array_sizes[1] = compiler->output_control_point_count;
-        output_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, SpvStorageClassOutput,
-                component_type, component_count, array_sizes, 2);
-        vkd3d_spirv_add_iface_variable(builder, output_id);
-        vkd3d_spirv_build_op_decorate1(builder, output_id, SpvDecorationLocation, output->register_index);
-        vkd3d_spirv_build_op_name(builder, output_id, "vocp%u", output->register_index);
-
         output_ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassOutput, type_id);
-        dst_id = vkd3d_spirv_build_op_access_chain1(builder, output_ptr_type_id, output_id, invocation_id);
+
+        dst_id = vkd3d_spirv_build_op_access_chain1(builder, output_ptr_type_id, output_reg_info.id, invocation_id);
 
         vkd3d_spirv_build_op_copy_memory(builder, dst_id, input_id, SpvMemoryAccessMaskNone);
     }
@@ -6593,12 +6702,14 @@ static SpvOp spirv_compiler_map_alu_instruction(const struct vkd3d_shader_instru
         {VKD3DSIH_ISHR,       SpvOpShiftRightArithmetic},
         {VKD3DSIH_ITOD,       SpvOpConvertSToF},
         {VKD3DSIH_ITOF,       SpvOpConvertSToF},
+        {VKD3DSIH_ITOI,       SpvOpSConvert},
         {VKD3DSIH_MUL,        SpvOpFMul},
         {VKD3DSIH_NOT,        SpvOpNot},
         {VKD3DSIH_OR,         SpvOpBitwiseOr},
         {VKD3DSIH_USHR,       SpvOpShiftRightLogical},
         {VKD3DSIH_UTOD,       SpvOpConvertUToF},
         {VKD3DSIH_UTOF,       SpvOpConvertUToF},
+        {VKD3DSIH_UTOU,       SpvOpUConvert},
         {VKD3DSIH_XOR,        SpvOpBitwiseXor},
     };
     unsigned int i;
@@ -6650,6 +6761,10 @@ static void spirv_compiler_emit_bool_cast(struct spirv_compiler *compiler,
     {
         val_id = spirv_compiler_emit_bool_to_int(compiler, 1, val_id, instruction->handler_idx == VKD3DSIH_ITOI);
     }
+    else if (dst->reg.data_type == VKD3D_DATA_UINT64)
+    {
+        val_id = spirv_compiler_emit_bool_to_int64(compiler, 1, val_id, instruction->handler_idx == VKD3DSIH_ITOI);
+    }
     else
     {
         WARN("Unhandled data type %u.\n", dst->reg.data_type);
@@ -6715,11 +6830,11 @@ static enum vkd3d_result spirv_compiler_emit_alu_instruction(struct spirv_compil
      * Microsoft fxc will compile immediate constants larger than 5 bits.
      * Fixing up the constants would be more elegant, but the simplest way is
      * to let this handle constants too. */
-    if (instruction->handler_idx == VKD3DSIH_ISHL || instruction->handler_idx == VKD3DSIH_ISHR
-            || instruction->handler_idx == VKD3DSIH_USHR)
+    if (!(instruction->flags & VKD3DSI_SHIFT_UNMASKED) && (instruction->handler_idx == VKD3DSIH_ISHL
+            || instruction->handler_idx == VKD3DSIH_ISHR || instruction->handler_idx == VKD3DSIH_USHR))
     {
         uint32_t mask_id = spirv_compiler_get_constant_vector(compiler,
-                VKD3D_SHADER_COMPONENT_UINT, vkd3d_write_mask_component_count(dst->write_mask), 0x1f);
+                VKD3D_SHADER_COMPONENT_UINT, vsir_write_mask_component_count(dst->write_mask), 0x1f);
         src_ids[1] = vkd3d_spirv_build_op_and(builder, type_id, src_ids[1], mask_id);
     }
 
@@ -6849,7 +6964,7 @@ static void spirv_compiler_emit_mov(struct spirv_compiler *compiler,
         return;
     }
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     if (component_count != 1 && component_count != VKD3D_VEC4_SIZE
             && dst_reg_info.write_mask == VKD3DSP_WRITEMASK_ALL)
     {
@@ -6863,7 +6978,7 @@ static void spirv_compiler_emit_mov(struct spirv_compiler *compiler,
         for (i = 0; i < ARRAY_SIZE(components); ++i)
         {
             if (dst->write_mask & (VKD3DSP_WRITEMASK_0 << i))
-                components[i] = VKD3D_VEC4_SIZE + vkd3d_swizzle_get_component(src->swizzle, i);
+                components[i] = VKD3D_VEC4_SIZE + vsir_swizzle_get_component(src->swizzle, i);
             else
                 components[i] = i;
         }
@@ -6877,6 +6992,11 @@ static void spirv_compiler_emit_mov(struct spirv_compiler *compiler,
 
 general_implementation:
     val_id = spirv_compiler_emit_load_src(compiler, src, dst->write_mask);
+    if (dst->reg.data_type != src->reg.data_type)
+    {
+        val_id = vkd3d_spirv_build_op_bitcast(builder, vkd3d_spirv_get_type_id_for_data_type(builder,
+                dst->reg.data_type, vsir_write_mask_component_count(dst->write_mask)), val_id);
+    }
     spirv_compiler_emit_store_dst(compiler, dst, val_id);
 }
 
@@ -6893,12 +7013,12 @@ static void spirv_compiler_emit_movc(struct spirv_compiler *compiler,
     src1_id = spirv_compiler_emit_load_src(compiler, &src[1], dst->write_mask);
     src2_id = spirv_compiler_emit_load_src(compiler, &src[2], dst->write_mask);
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     type_id = spirv_compiler_get_type_id_for_dst(compiler, dst);
 
     if (src[0].reg.data_type != VKD3D_DATA_BOOL)
         condition_id = spirv_compiler_emit_int_to_bool(compiler,
-                VKD3D_SHADER_CONDITIONAL_OP_NZ, component_count, condition_id);
+                VKD3D_SHADER_CONDITIONAL_OP_NZ, src[0].reg.data_type, component_count, condition_id);
     val_id = vkd3d_spirv_build_op_select(builder, type_id, condition_id, src1_id, src2_id);
 
     spirv_compiler_emit_store_dst(compiler, dst, val_id);
@@ -6919,11 +7039,11 @@ static void spirv_compiler_emit_swapc(struct spirv_compiler *compiler,
     src1_id = spirv_compiler_emit_load_src(compiler, &src[1], dst->write_mask);
     src2_id = spirv_compiler_emit_load_src(compiler, &src[2], dst->write_mask);
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_FLOAT, component_count);
 
     condition_id = spirv_compiler_emit_int_to_bool(compiler,
-            VKD3D_SHADER_CONDITIONAL_OP_NZ, component_count, condition_id);
+            VKD3D_SHADER_CONDITIONAL_OP_NZ, src[0].reg.data_type, component_count, condition_id);
 
     val_id = vkd3d_spirv_build_op_select(builder, type_id, condition_id, src2_id, src1_id);
     spirv_compiler_emit_store_dst(compiler, &dst[0], val_id);
@@ -6942,7 +7062,7 @@ static void spirv_compiler_emit_dot(struct spirv_compiler *compiler,
     unsigned int component_count, i;
     DWORD write_mask;
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     component_type = vkd3d_component_type_from_data_type(dst->reg.data_type);
 
     if (instruction->handler_idx == VKD3DSIH_DP4)
@@ -6980,7 +7100,7 @@ static void spirv_compiler_emit_rcp(struct spirv_compiler *compiler,
     uint32_t type_id, src_id, val_id, div_id;
     unsigned int component_count;
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     type_id = spirv_compiler_get_type_id_for_dst(compiler, dst);
 
     src_id = spirv_compiler_emit_load_src(compiler, src, dst->write_mask);
@@ -7060,7 +7180,7 @@ static void spirv_compiler_emit_imad(struct spirv_compiler *compiler,
     uint32_t type_id, val_id, src_ids[3];
     unsigned int i, component_count;
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_INT, component_count);
 
     for (i = 0; i < ARRAY_SIZE(src_ids); ++i)
@@ -7087,16 +7207,18 @@ static void spirv_compiler_emit_int_div(struct spirv_compiler *compiler,
 
     if (dst[0].reg.type != VKD3DSPR_NULL)
     {
-        component_count = vkd3d_write_mask_component_count(dst[0].write_mask);
+        component_count = vsir_write_mask_component_count(dst[0].write_mask);
         type_id = spirv_compiler_get_type_id_for_dst(compiler, &dst[0]);
 
         src0_id = spirv_compiler_emit_load_src(compiler, &src[0], dst[0].write_mask);
         src1_id = spirv_compiler_emit_load_src(compiler, &src[1], dst[0].write_mask);
 
         condition_id = spirv_compiler_emit_int_to_bool(compiler,
-                VKD3D_SHADER_CONDITIONAL_OP_NZ, component_count, src1_id);
-        uint_max_id = spirv_compiler_get_constant_uint_vector(compiler,
-                0xffffffff, component_count);
+                VKD3D_SHADER_CONDITIONAL_OP_NZ, src[1].reg.data_type, component_count, src1_id);
+        if (dst[0].reg.data_type == VKD3D_DATA_UINT64)
+            uint_max_id = spirv_compiler_get_constant_uint64_vector(compiler, UINT64_MAX, component_count);
+        else
+            uint_max_id = spirv_compiler_get_constant_uint_vector(compiler, 0xffffffff, component_count);
 
         val_id = vkd3d_spirv_build_op_tr2(builder, &builder->function_stream, div_op, type_id, src0_id, src1_id);
         /* The SPIR-V spec says: "The resulting value is undefined if Operand 2 is 0." */
@@ -7109,16 +7231,18 @@ static void spirv_compiler_emit_int_div(struct spirv_compiler *compiler,
     {
         if (!component_count || dst[0].write_mask != dst[1].write_mask)
         {
-            component_count = vkd3d_write_mask_component_count(dst[1].write_mask);
+            component_count = vsir_write_mask_component_count(dst[1].write_mask);
             type_id = spirv_compiler_get_type_id_for_dst(compiler, &dst[1]);
 
             src0_id = spirv_compiler_emit_load_src(compiler, &src[0], dst[1].write_mask);
             src1_id = spirv_compiler_emit_load_src(compiler, &src[1], dst[1].write_mask);
 
             condition_id = spirv_compiler_emit_int_to_bool(compiler,
-                    VKD3D_SHADER_CONDITIONAL_OP_NZ, component_count, src1_id);
-            uint_max_id = spirv_compiler_get_constant_uint_vector(compiler,
-                    0xffffffff, component_count);
+                    VKD3D_SHADER_CONDITIONAL_OP_NZ, src[1].reg.data_type, component_count, src1_id);
+            if (dst[1].reg.data_type == VKD3D_DATA_UINT64)
+                uint_max_id = spirv_compiler_get_constant_uint64_vector(compiler, UINT64_MAX, component_count);
+            else
+                uint_max_id = spirv_compiler_get_constant_uint_vector(compiler, 0xffffffff, component_count);
         }
 
         val_id = vkd3d_spirv_build_op_tr2(builder, &builder->function_stream, mod_op, type_id, src0_id, src1_id);
@@ -7147,7 +7271,7 @@ static void spirv_compiler_emit_ftoi(struct spirv_compiler *compiler,
      * as a signed integer, but Direct3D expects the result to saturate,
      * and for NaN to yield zero. */
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     src_type_id = spirv_compiler_get_type_id_for_reg(compiler, &src->reg, dst->write_mask);
     dst_type_id = spirv_compiler_get_type_id_for_dst(compiler, dst);
     src_id = spirv_compiler_emit_load_src(compiler, src, dst->write_mask);
@@ -7200,7 +7324,7 @@ static void spirv_compiler_emit_ftou(struct spirv_compiler *compiler,
      * as an unsigned integer, but Direct3D expects the result to saturate,
      * and for NaN to yield zero. */
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     src_type_id = spirv_compiler_get_type_id_for_reg(compiler, &src->reg, dst->write_mask);
     dst_type_id = spirv_compiler_get_type_id_for_dst(compiler, dst);
     src_id = spirv_compiler_emit_load_src(compiler, src, dst->write_mask);
@@ -7387,7 +7511,7 @@ static void spirv_compiler_emit_comparison_instruction(struct spirv_compiler *co
             return;
     }
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
 
     src0_id = spirv_compiler_emit_load_src(compiler, &src[0], dst->write_mask);
     src1_id = spirv_compiler_emit_load_src(compiler, &src[1], dst->write_mask);
@@ -7409,7 +7533,7 @@ static uint32_t spirv_compiler_emit_conditional_branch(struct spirv_compiler *co
     uint32_t condition_id, merge_block_id;
 
     condition_id = spirv_compiler_emit_load_src(compiler, src, VKD3DSP_WRITEMASK_0);
-    condition_id = spirv_compiler_emit_int_to_bool(compiler, instruction->flags, 1, condition_id);
+    condition_id = spirv_compiler_emit_int_to_bool(compiler, instruction->flags, src->reg.data_type, 1, condition_id);
 
     merge_block_id = vkd3d_spirv_alloc_id(builder);
 
@@ -7542,7 +7666,7 @@ static int spirv_compiler_emit_control_flow_instruction(struct spirv_compiler *c
                 return VKD3D_ERROR_OUT_OF_MEMORY;
 
             val_id = spirv_compiler_emit_load_src(compiler, src, VKD3DSP_WRITEMASK_0);
-            condition_id = spirv_compiler_emit_int_to_bool(compiler, instruction->flags, 1, val_id);
+            condition_id = spirv_compiler_emit_int_to_bool(compiler, instruction->flags, src->reg.data_type, 1, val_id);
 
             true_label = vkd3d_spirv_alloc_id(builder);
             merge_block_id = vkd3d_spirv_alloc_id(builder);
@@ -7705,7 +7829,7 @@ static int spirv_compiler_emit_control_flow_instruction(struct spirv_compiler *c
                         "The swizzle for a switch case value is not scalar.");
             }
             assert(src->reg.type == VKD3DSPR_IMMCONST);
-            value = *src->reg.u.immconst_uint;
+            value = *src->reg.u.immconst_u32;
 
             if (!vkd3d_array_reserve((void **)&cf_info->u.switch_.case_blocks, &cf_info->u.switch_.case_blocks_size,
                     2 * (cf_info->u.switch_.case_block_count + 1), sizeof(*cf_info->u.switch_.case_blocks)))
@@ -8325,7 +8449,7 @@ static void spirv_compiler_emit_gather4(struct spirv_compiler *compiler,
     }
     else
     {
-        component_idx = vkd3d_swizzle_get_component(sampler->swizzle, 0);
+        component_idx = vsir_swizzle_get_component(sampler->swizzle, 0);
         /* Nvidia driver requires signed integer type. */
         component_id = spirv_compiler_get_constant(compiler,
                 VKD3D_SHADER_COMPONENT_INT, 1, &component_idx);
@@ -8403,7 +8527,7 @@ static void spirv_compiler_emit_ld_raw_structured_srv_uav(struct spirv_compiler
             if (!(dst->write_mask & (VKD3DSP_WRITEMASK_0 << i)))
                 continue;
 
-            component_idx = vkd3d_swizzle_get_component(resource->swizzle, i);
+            component_idx = vsir_swizzle_get_component(resource->swizzle, i);
             coordinate_id = base_coordinate_id;
             if (component_idx)
                 coordinate_id = vkd3d_spirv_build_op_iadd(builder, type_id,
@@ -8435,7 +8559,7 @@ static void spirv_compiler_emit_ld_raw_structured_srv_uav(struct spirv_compiler
             if (!(dst->write_mask & (VKD3DSP_WRITEMASK_0 << i)))
                 continue;
 
-            component_idx = vkd3d_swizzle_get_component(resource->swizzle, i);
+            component_idx = vsir_swizzle_get_component(resource->swizzle, i);
             coordinate_id = base_coordinate_id;
             if (component_idx)
                 coordinate_id = vkd3d_spirv_build_op_iadd(builder, type_id,
@@ -8479,7 +8603,7 @@ static void spirv_compiler_emit_ld_tgsm(struct spirv_compiler *compiler,
         if (!(dst->write_mask & (VKD3DSP_WRITEMASK_0 << i)))
             continue;
 
-        component_idx = vkd3d_swizzle_get_component(resource->swizzle, i);
+        component_idx = vsir_swizzle_get_component(resource->swizzle, i);
         coordinate_id = base_coordinate_id;
         if (component_idx)
             coordinate_id = vkd3d_spirv_build_op_iadd(builder, type_id,
@@ -8540,7 +8664,7 @@ static void spirv_compiler_emit_store_uav_raw_structured(struct spirv_compiler *
         assert(data->reg.data_type == VKD3D_DATA_UINT);
         val_id = spirv_compiler_emit_load_src(compiler, data, dst->write_mask);
 
-        component_count = vkd3d_write_mask_component_count(dst->write_mask);
+        component_count = vsir_write_mask_component_count(dst->write_mask);
         for (component_idx = 0; component_idx < component_count; ++component_idx)
         {
             data_id = component_count > 1 ?
@@ -8569,7 +8693,7 @@ static void spirv_compiler_emit_store_uav_raw_structured(struct spirv_compiler *
         assert(data->reg.data_type == VKD3D_DATA_UINT);
         val_id = spirv_compiler_emit_load_src(compiler, data, dst->write_mask);
 
-        component_count = vkd3d_write_mask_component_count(dst->write_mask);
+        component_count = vsir_write_mask_component_count(dst->write_mask);
         for (component_idx = 0; component_idx < component_count; ++component_idx)
         {
             /* Mesa Vulkan drivers require the texel parameter to be a vector. */
@@ -8613,7 +8737,7 @@ static void spirv_compiler_emit_store_tgsm(struct spirv_compiler *compiler,
     assert(data->reg.data_type == VKD3D_DATA_UINT);
     val_id = spirv_compiler_emit_load_src(compiler, data, dst->write_mask);
 
-    component_count = vkd3d_write_mask_component_count(dst->write_mask);
+    component_count = vsir_write_mask_component_count(dst->write_mask);
     for (component_idx = 0; component_idx < component_count; ++component_idx)
     {
         data_id = component_count > 1 ?
@@ -9275,7 +9399,7 @@ static void spirv_compiler_emit_eval_attrib(struct spirv_compiler *compiler,
     }
 
     type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_FLOAT,
-            vkd3d_write_mask_component_count(register_info.write_mask));
+            vsir_write_mask_component_count(register_info.write_mask));
 
     instr_set_id = vkd3d_spirv_get_glsl_std450_instr_set(builder);
     val_id = vkd3d_spirv_build_op_ext_inst(builder, type_id, instr_set_id, op, src_ids, src_count);
@@ -9425,18 +9549,9 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler,
         case VKD3DSIH_DCL_INPUT:
             spirv_compiler_emit_dcl_input(compiler, instruction);
             break;
-        case VKD3DSIH_DCL_INPUT_PS_SGV:
-        case VKD3DSIH_DCL_INPUT_PS_SIV:
-        case VKD3DSIH_DCL_INPUT_SGV:
-        case VKD3DSIH_DCL_INPUT_SIV:
-            spirv_compiler_emit_dcl_input_sysval(compiler, instruction);
-            break;
         case VKD3DSIH_DCL_OUTPUT:
             spirv_compiler_emit_dcl_output(compiler, instruction);
             break;
-        case VKD3DSIH_DCL_OUTPUT_SIV:
-            spirv_compiler_emit_dcl_output_siv(compiler, instruction);
-            break;
         case VKD3DSIH_DCL_STREAM:
             spirv_compiler_emit_dcl_stream(compiler, instruction);
             break;
@@ -9719,6 +9834,11 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler,
         case VKD3DSIH_DCL_CONSTANT_BUFFER:
         case VKD3DSIH_DCL_HS_MAX_TESSFACTOR:
         case VKD3DSIH_DCL_INPUT_CONTROL_POINT_COUNT:
+        case VKD3DSIH_DCL_INPUT_SGV:
+        case VKD3DSIH_DCL_INPUT_SIV:
+        case VKD3DSIH_DCL_INPUT_PS_SGV:
+        case VKD3DSIH_DCL_INPUT_PS_SIV:
+        case VKD3DSIH_DCL_OUTPUT_SIV:
         case VKD3DSIH_DCL_RESOURCE_RAW:
         case VKD3DSIH_DCL_RESOURCE_STRUCTURED:
         case VKD3DSIH_DCL_SAMPLER:
@@ -9740,6 +9860,30 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler,
     return ret;
 }
 
+static void spirv_compiler_emit_io_declarations(struct spirv_compiler *compiler)
+{
+    for (unsigned int i = 0; i < compiler->input_signature.element_count; ++i)
+        spirv_compiler_emit_input(compiler, VKD3DSPR_INPUT, i);
+
+    for (unsigned int i = 0; i < compiler->output_signature.element_count; ++i)
+    {
+        /* PS outputs other than TARGET have dedicated registers and therefore
+         * go through spirv_compiler_emit_dcl_output() for now. */
+        if (compiler->shader_type == VKD3D_SHADER_TYPE_PIXEL
+                && compiler->output_signature.elements[i].sysval_semantic != VKD3D_SHADER_SV_TARGET)
+            continue;
+        spirv_compiler_emit_output(compiler, VKD3DSPR_OUTPUT, i);
+    }
+
+    for (unsigned int i = 0; i < compiler->patch_constant_signature.element_count; ++i)
+    {
+        if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL)
+            spirv_compiler_emit_output(compiler, VKD3DSPR_PATCHCONST, i);
+        else
+            spirv_compiler_emit_input(compiler, VKD3DSPR_PATCHCONST, i);
+    }
+}
+
 static void spirv_compiler_emit_descriptor_declarations(struct spirv_compiler *compiler)
 {
     unsigned int i;
@@ -9826,6 +9970,8 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler,
     if (compiler->shader_type != VKD3D_SHADER_TYPE_HULL)
         spirv_compiler_emit_shader_signature_outputs(compiler);
 
+    spirv_compiler_emit_io_declarations(compiler);
+
     for (i = 0; i < instructions.count && result >= 0; ++i)
     {
         compiler->location.line = i + 1;
@@ -9874,11 +10020,28 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler,
     if (!vkd3d_spirv_compile_module(builder, spirv, spirv_compiler_get_entry_point_name(compiler)))
         return VKD3D_ERROR;
 
-    if (TRACE_ON())
+    if (TRACE_ON() || parser->config_flags & VKD3D_SHADER_CONFIG_FLAG_FORCE_VALIDATION)
     {
         enum vkd3d_shader_spirv_environment environment = spirv_compiler_get_target_environment(compiler);
-        vkd3d_spirv_dump(spirv, environment);
-        vkd3d_spirv_validate(spirv, environment);
+        struct vkd3d_string_buffer buffer;
+
+        if (TRACE_ON())
+            vkd3d_spirv_dump(spirv, environment);
+
+        vkd3d_string_buffer_init(&buffer);
+        if (!vkd3d_spirv_validate(&buffer, spirv, environment))
+        {
+            FIXME("Failed to validate SPIR-V binary.\n");
+            vkd3d_shader_trace_text(buffer.buffer, buffer.content_size);
+
+            if (compiler->config_flags & VKD3D_SHADER_CONFIG_FLAG_FORCE_VALIDATION)
+            {
+                spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_INVALID_SHADER,
+                        "Execution generated an invalid shader, failing compilation:\n%s",
+                        buffer.buffer);
+            }
+        }
+        vkd3d_string_buffer_cleanup(&buffer);
     }
 
     if (compiler->failed)
@@ -9905,7 +10068,7 @@ int spirv_compile(struct vkd3d_shader_parser *parser,
     int ret;
 
     if (!(spirv_compiler = spirv_compiler_create(&parser->shader_version, &parser->shader_desc,
-            compile_info, scan_descriptor_info, message_context, &parser->location)))
+            compile_info, scan_descriptor_info, message_context, &parser->location, parser->config_flags)))
     {
         ERR("Failed to create SPIR-V compiler.\n");
         return VKD3D_ERROR;
diff --git a/libs/vkd3d/libs/vkd3d-shader/tpf.c b/libs/vkd3d/libs/vkd3d-shader/tpf.c
index 80f8ab98c08..f859e758d8e 100644
--- a/libs/vkd3d/libs/vkd3d-shader/tpf.c
+++ b/libs/vkd3d/libs/vkd3d-shader/tpf.c
@@ -792,9 +792,11 @@ static void shader_sm4_read_shader_data(struct vkd3d_shader_instruction *ins, ui
         ins->handler_idx = VKD3DSIH_INVALID;
         return;
     }
+    icb->register_idx = 0;
     icb->data_type = VKD3D_DATA_FLOAT;
     icb->component_count = VKD3D_VEC4_SIZE;
     icb->element_count = icb_size / VKD3D_VEC4_SIZE;
+    icb->is_null = false;
     memcpy(icb->data, tokens, sizeof(*tokens) * icb_size);
     shader_instruction_array_add_icb(&priv->p.instructions, icb);
     ins->declaration.icb = icb;
@@ -918,10 +920,11 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins
         uint32_t opcode_token, const uint32_t *tokens, unsigned int token_count, struct vkd3d_shader_sm4_parser *priv)
 {
     struct vkd3d_shader_index_range *index_range = &ins->declaration.index_range;
-    unsigned int i, register_idx, register_count, write_mask;
+    unsigned int i, register_idx, register_count;
     enum vkd3d_shader_register_type type;
     struct sm4_index_range_array *ranges;
     unsigned int *io_masks;
+    uint32_t write_mask;
 
     shader_sm4_read_dst_param(priv, &tokens, &tokens[token_count], VKD3D_DATA_OPAQUE,
             &index_range->dst);
@@ -931,7 +934,7 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins
     register_count = index_range->register_count;
     write_mask = index_range->dst.write_mask;
 
-    if (vkd3d_write_mask_component_count(write_mask) != 1)
+    if (vsir_write_mask_component_count(write_mask) != 1)
     {
         WARN("Unhandled write mask %#x.\n", write_mask);
         vkd3d_shader_parser_warning(&priv->p, VKD3D_SHADER_WARNING_TPF_UNHANDLED_INDEX_RANGE_MASK,
@@ -1900,19 +1903,19 @@ static bool shader_sm4_read_param(struct vkd3d_shader_sm4_parser *priv, const ui
 
     if (register_type == VKD3D_SM4_RT_IMMCONST || register_type == VKD3D_SM4_RT_IMMCONST64)
     {
-        unsigned int dword_count;
+        unsigned int u32_count;
 
         switch (param->dimension)
         {
             case VSIR_DIMENSION_SCALAR:
-                dword_count = 1 + (register_type == VKD3D_SM4_RT_IMMCONST64);
-                if (end - *ptr < dword_count)
+                u32_count = 1 + (register_type == VKD3D_SM4_RT_IMMCONST64);
+                if (end - *ptr < u32_count)
                 {
                     WARN("Invalid ptr %p, end %p.\n", *ptr, end);
                     return false;
                 }
-                memcpy(param->u.immconst_uint, *ptr, dword_count * sizeof(DWORD));
-                *ptr += dword_count;
+                memcpy(param->u.immconst_u32, *ptr, u32_count * sizeof(uint32_t));
+                *ptr += u32_count;
                 break;
 
             case VSIR_DIMENSION_VEC4:
@@ -1921,7 +1924,7 @@ static bool shader_sm4_read_param(struct vkd3d_shader_sm4_parser *priv, const ui
                     WARN("Invalid ptr %p, end %p.\n", *ptr, end);
                     return false;
                 }
-                memcpy(param->u.immconst_uint, *ptr, VKD3D_VEC4_SIZE * sizeof(DWORD));
+                memcpy(param->u.immconst_u32, *ptr, VKD3D_VEC4_SIZE * sizeof(uint32_t));
                 *ptr += 4;
                 break;
 
@@ -1930,6 +1933,15 @@ static bool shader_sm4_read_param(struct vkd3d_shader_sm4_parser *priv, const ui
                 break;
         }
     }
+    else if (register_type == VKD3D_SM4_RT_IMMCONSTBUFFER)
+    {
+        if (param->idx_count != 1)
+        {
+            WARN("Unexpected idx count %u.\n", param->idx_count);
+            vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT,
+                    "Invalid index count %u for immediate const buffer register; expected count 1.", param->idx_count);
+        }
+    }
     else if (!shader_is_sm_5_1(priv) && vsir_register_is_descriptor(param))
     {
         /* SM5.1 places a symbol identifier in idx[0] and moves
@@ -1970,10 +1982,10 @@ static uint32_t swizzle_from_sm4(uint32_t s)
 static uint32_t swizzle_to_sm4(uint32_t s)
 {
     uint32_t ret = 0;
-    ret |= ((vkd3d_swizzle_get_component(s, 0)) & 0x3);
-    ret |= ((vkd3d_swizzle_get_component(s, 1)) & 0x3) << 2;
-    ret |= ((vkd3d_swizzle_get_component(s, 2)) & 0x3) << 4;
-    ret |= ((vkd3d_swizzle_get_component(s, 3)) & 0x3) << 6;
+    ret |= ((vsir_swizzle_get_component(s, 0)) & 0x3);
+    ret |= ((vsir_swizzle_get_component(s, 1)) & 0x3) << 2;
+    ret |= ((vsir_swizzle_get_component(s, 2)) & 0x3) << 4;
+    ret |= ((vsir_swizzle_get_component(s, 3)) & 0x3) << 6;
     return ret;
 }
 
@@ -2002,12 +2014,12 @@ static bool register_is_control_point_input(const struct vkd3d_shader_register *
             || priv->p.shader_version.type == VKD3D_SHADER_TYPE_GEOMETRY));
 }
 
-static unsigned int mask_from_swizzle(unsigned int swizzle)
+static uint32_t mask_from_swizzle(uint32_t swizzle)
 {
-    return (1u << vkd3d_swizzle_get_component(swizzle, 0))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 1))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 2))
-            | (1u << vkd3d_swizzle_get_component(swizzle, 3));
+    return (1u << vsir_swizzle_get_component(swizzle, 0))
+            | (1u << vsir_swizzle_get_component(swizzle, 1))
+            | (1u << vsir_swizzle_get_component(swizzle, 2))
+            | (1u << vsir_swizzle_get_component(swizzle, 3));
 }
 
 static bool shader_sm4_validate_input_output_register(struct vkd3d_shader_sm4_parser *priv,
@@ -2219,7 +2231,7 @@ static bool shader_sm4_read_dst_param(struct vkd3d_shader_sm4_parser *priv, cons
     }
 
     if (data_type == VKD3D_DATA_DOUBLE)
-        dst_param->write_mask = vkd3d_write_mask_64_from_32(dst_param->write_mask);
+        dst_param->write_mask = vsir_write_mask_64_from_32(dst_param->write_mask);
     /* Some 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;
@@ -2533,6 +2545,16 @@ static bool shader_sm4_init(struct vkd3d_shader_sm4_parser *sm4, const uint32_t
     return true;
 }
 
+static void uninvert_used_masks(struct shader_signature *signature)
+{
+    for (unsigned int i = 0; i < signature->element_count; ++i)
+    {
+        struct signature_element *e = &signature->elements[i];
+
+        e->used_mask = e->mask & ~e->used_mask;
+    }
+}
+
 static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser *sm4,
         const struct shader_signature *signature, unsigned int *masks, const char *name)
 {
@@ -2628,6 +2650,12 @@ int vkd3d_shader_sm4_parser_create(const struct vkd3d_shader_compile_info *compi
         return VKD3D_ERROR_INVALID_ARGUMENT;
     }
 
+    /* DXBC stores used masks inverted for output signatures, for some reason.
+     * We return them un-inverted. */
+    uninvert_used_masks(&shader_desc->output_signature);
+    if (sm4->p.shader_version.type == VKD3D_SHADER_TYPE_HULL)
+        uninvert_used_masks(&shader_desc->patch_constant_signature);
+
     if (!shader_sm4_parser_validate_signature(sm4, &shader_desc->input_signature,
             sm4->input_register_masks, "Input")
             || !shader_sm4_parser_validate_signature(sm4, &shader_desc->output_signature,
@@ -3842,7 +3870,7 @@ static void sm4_src_from_constant_value(struct vkd3d_shader_src_param *src,
     if (width == 1)
     {
         src->reg.dimension = VSIR_DIMENSION_SCALAR;
-        src->reg.u.immconst_uint[0] = value->u[0].u;
+        src->reg.u.immconst_u32[0] = value->u[0].u;
     }
     else
     {
@@ -3852,9 +3880,9 @@ static void sm4_src_from_constant_value(struct vkd3d_shader_src_param *src,
         for (i = 0; i < 4; ++i)
         {
             if ((map_writemask & (1u << i)) && (j < width))
-                src->reg.u.immconst_uint[i] = value->u[j++].u;
+                src->reg.u.immconst_u32[i] = value->u[j++].u;
             else
-                src->reg.u.immconst_uint[i] = 0;
+                src->reg.u.immconst_u32[i] = 0;
         }
     }
 }
@@ -4049,12 +4077,12 @@ static void sm4_write_src_register(const struct tpf_writer *tpf, const struct vk
 
     if (src->reg.type == VKD3DSPR_IMMCONST)
     {
-        put_u32(buffer, src->reg.u.immconst_uint[0]);
+        put_u32(buffer, src->reg.u.immconst_u32[0]);
         if (src->reg.dimension == VSIR_DIMENSION_VEC4)
         {
-            put_u32(buffer, src->reg.u.immconst_uint[1]);
-            put_u32(buffer, src->reg.u.immconst_uint[2]);
-            put_u32(buffer, src->reg.u.immconst_uint[3]);
+            put_u32(buffer, src->reg.u.immconst_u32[1]);
+            put_u32(buffer, src->reg.u.immconst_u32[2]);
+            put_u32(buffer, src->reg.u.immconst_u32[3]);
         }
     }
 }
@@ -4589,7 +4617,7 @@ static void write_sm4_ld(const struct tpf_writer *tpf, const struct hlsl_ir_node
             memset(&instr.srcs[2], 0, sizeof(instr.srcs[2]));
             reg->type = VKD3DSPR_IMMCONST;
             reg->dimension = VSIR_DIMENSION_SCALAR;
-            reg->u.immconst_uint[0] = index->value.u[0].u;
+            reg->u.immconst_u32[0] = index->value.u[0].u;
         }
         else if (tpf->ctx->profile->major_version == 4 && tpf->ctx->profile->minor_version == 0)
         {
@@ -4750,7 +4778,7 @@ static void write_sm4_cast_from_bool(const struct tpf_writer *tpf, const struct
     sm4_src_from_node(tpf, &instr.srcs[0], arg, instr.dsts[0].write_mask);
     instr.srcs[1].reg.type = VKD3DSPR_IMMCONST;
     instr.srcs[1].reg.dimension = VSIR_DIMENSION_SCALAR;
-    instr.srcs[1].reg.u.immconst_uint[0] = mask;
+    instr.srcs[1].reg.u.immconst_u32[0] = mask;
     instr.src_count = 2;
 
     write_sm4_instruction(tpf, &instr);
diff --git a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_main.c
index 8fd8945151f..8fd98d2a551 100644
--- a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_main.c
+++ b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_main.c
@@ -23,6 +23,8 @@
 #include <stdio.h>
 #include <math.h>
 
+/* VKD3D_DEBUG_ENV_NAME("VKD3D_SHADER_DEBUG"); */
+
 static inline int char_to_int(char c)
 {
     if ('0' <= c && c <= '9')
@@ -305,6 +307,16 @@ void vkd3d_shader_vwarning(struct vkd3d_shader_message_context *context, const s
     vkd3d_string_buffer_printf(&context->messages, "\n");
 }
 
+void vkd3d_shader_warning(struct vkd3d_shader_message_context *context, const struct vkd3d_shader_location *location,
+        enum vkd3d_shader_error error, const char *format, ...)
+{
+    va_list args;
+
+    va_start(args, format);
+    vkd3d_shader_vwarning(context, location, error, format, args);
+    va_end(args);
+}
+
 void vkd3d_shader_verror(struct vkd3d_shader_message_context *context, const struct vkd3d_shader_location *location,
         enum vkd3d_shader_error error, const char *format, va_list args)
 {
@@ -888,6 +900,21 @@ static void vkd3d_shader_scan_combined_sampler_declaration(
             &semantic->resource.range, semantic->resource_type, VKD3D_SHADER_RESOURCE_DATA_FLOAT);
 }
 
+static const struct vkd3d_shader_descriptor_info1 *find_descriptor(
+        const struct vkd3d_shader_scan_descriptor_info1 *info,
+        enum vkd3d_shader_descriptor_type type, unsigned int register_id)
+{
+    for (unsigned int i = 0; i < info->descriptor_count; ++i)
+    {
+        const struct vkd3d_shader_descriptor_info1 *d = &info->descriptors[i];
+
+        if (d->type == type && d->register_id == register_id)
+            return d;
+    }
+
+    return NULL;
+}
+
 static void vkd3d_shader_scan_combined_sampler_usage(struct vkd3d_shader_scan_context *context,
         const struct vkd3d_shader_register *resource, const struct vkd3d_shader_register *sampler)
 {
@@ -913,7 +940,6 @@ static void vkd3d_shader_scan_combined_sampler_usage(struct vkd3d_shader_scan_co
 
     if (vkd3d_shader_ver_ge(context->version, 5, 1))
     {
-        const struct vkd3d_shader_scan_descriptor_info1 *info = context->scan_descriptor_info;
         const struct vkd3d_shader_descriptor_info1 *d;
         bool dynamic_resource, dynamic_sampler;
 
@@ -928,30 +954,13 @@ static void vkd3d_shader_scan_combined_sampler_usage(struct vkd3d_shader_scan_co
         if (dynamic_resource || dynamic_sampler)
             return;
 
-        for (i = 0; i < info->descriptor_count; ++i)
-        {
-            d = &info->descriptors[i];
-            if (d->type != VKD3D_SHADER_DESCRIPTOR_TYPE_SRV)
-                continue;
-            if (d->register_id != resource->idx[0].offset)
-                continue;
+        if ((d = find_descriptor(context->scan_descriptor_info,
+                VKD3D_SHADER_DESCRIPTOR_TYPE_SRV, resource->idx[0].offset)))
             resource_space = d->register_space;
-            break;
-        }
 
-        if (sampler)
-        {
-            for (i = 0; i < info->descriptor_count; ++i)
-            {
-                d = &info->descriptors[i];
-                if (d->type != VKD3D_SHADER_DESCRIPTOR_TYPE_SAMPLER)
-                    continue;
-                if (d->register_id != sampler->idx[0].offset)
-                    continue;
-                sampler_space = d->register_space;
-                break;
-            }
-        }
+        if (sampler && (d = find_descriptor(context->scan_descriptor_info,
+                VKD3D_SHADER_DESCRIPTOR_TYPE_SAMPLER, sampler->idx[0].offset)))
+            sampler_space = d->register_space;
     }
 
     for (i = 0; i < info->combined_sampler_count; ++i)
@@ -1074,6 +1083,9 @@ static int vkd3d_shader_scan_instruction(struct vkd3d_shader_scan_context *conte
             vkd3d_shader_scan_sampler_declaration(context, instruction);
             break;
         case VKD3DSIH_DCL:
+            if (instruction->declaration.semantic.resource_type == VKD3D_SHADER_RESOURCE_NONE)
+                break;
+
             if (instruction->declaration.semantic.resource.reg.reg.type == VKD3DSPR_COMBINED_SAMPLER)
             {
                 vkd3d_shader_scan_combined_sampler_declaration(context, &instruction->declaration.semantic);
@@ -1541,16 +1553,15 @@ static int vkd3d_shader_parser_compile(struct vkd3d_shader_parser *parser,
 
     scan_info = *compile_info;
 
-    if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0)
-        return ret;
-
     switch (compile_info->target_type)
     {
         case VKD3D_SHADER_TARGET_D3D_ASM:
-            ret = vkd3d_dxbc_binary_to_text(&parser->instructions, &parser->shader_version, compile_info, out);
+            ret = vkd3d_dxbc_binary_to_text(&parser->instructions, &parser->shader_version, compile_info, out, VSIR_ASM_D3D);
             break;
 
         case VKD3D_SHADER_TARGET_GLSL:
+            if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0)
+                return ret;
             if (!(glsl_generator = vkd3d_glsl_generator_create(&parser->shader_version,
                     message_context, &parser->location)))
             {
@@ -1561,19 +1572,22 @@ static int vkd3d_shader_parser_compile(struct vkd3d_shader_parser *parser,
 
             ret = vkd3d_glsl_generator_generate(glsl_generator, parser, out);
             vkd3d_glsl_generator_destroy(glsl_generator);
+            vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info);
             break;
 
         case VKD3D_SHADER_TARGET_SPIRV_BINARY:
         case VKD3D_SHADER_TARGET_SPIRV_TEXT:
+            if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0)
+                return ret;
             ret = spirv_compile(parser, &scan_descriptor_info, compile_info, out, message_context);
+            vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info);
             break;
 
         default:
             /* Validation should prevent us from reaching this. */
-            assert(0);
+            vkd3d_unreachable();
     }
 
-    vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info);
     return ret;
 }
 
@@ -1622,14 +1636,10 @@ static int compile_d3d_bytecode(const struct vkd3d_shader_compile_info *compile_
         return ret;
     }
 
-    if (compile_info->target_type == VKD3D_SHADER_TARGET_D3D_ASM)
-    {
-        ret = vkd3d_dxbc_binary_to_text(&parser->instructions, &parser->shader_version, compile_info, out);
-        vkd3d_shader_parser_destroy(parser);
-        return ret;
-    }
+    ret = vkd3d_shader_parser_compile(parser, compile_info, out, message_context);
 
-    return VKD3D_ERROR;
+    vkd3d_shader_parser_destroy(parser);
+    return ret;
 }
 
 static int compile_dxbc_dxil(const struct vkd3d_shader_compile_info *compile_info,
@@ -1907,6 +1917,10 @@ const enum vkd3d_shader_target_type *vkd3d_shader_get_supported_target_types(
 
     static const enum vkd3d_shader_target_type d3dbc_types[] =
     {
+        VKD3D_SHADER_TARGET_SPIRV_BINARY,
+#ifdef HAVE_SPIRV_TOOLS
+        VKD3D_SHADER_TARGET_SPIRV_TEXT,
+#endif
         VKD3D_SHADER_TARGET_D3D_ASM,
     };
 
diff --git a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h
index a93fa7160f7..47b245bd7ee 100644
--- a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h
+++ b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h
@@ -91,11 +91,12 @@ enum vkd3d_shader_error
     VKD3D_SHADER_ERROR_SPV_INVALID_REGISTER_TYPE        = 2001,
     VKD3D_SHADER_ERROR_SPV_INVALID_DESCRIPTOR_BINDING   = 2002,
     VKD3D_SHADER_ERROR_SPV_DESCRIPTOR_IDX_UNSUPPORTED   = 2003,
-    VKD3D_SHADER_ERROR_SPV_STENCIL_EXPORT_UNSUPPORTED   = 2004,
+    VKD3D_SHADER_ERROR_SPV_UNSUPPORTED_FEATURE          = 2004,
     VKD3D_SHADER_ERROR_SPV_OUT_OF_MEMORY                = 2005,
     VKD3D_SHADER_ERROR_SPV_INVALID_TYPE                 = 2006,
     VKD3D_SHADER_ERROR_SPV_INVALID_HANDLER              = 2007,
     VKD3D_SHADER_ERROR_SPV_NOT_IMPLEMENTED              = 2008,
+    VKD3D_SHADER_ERROR_SPV_INVALID_SHADER               = 2009,
 
     VKD3D_SHADER_WARNING_SPV_INVALID_SWIZZLE            = 2300,
 
@@ -196,6 +197,7 @@ enum vkd3d_shader_error
     VKD3D_SHADER_WARNING_DXIL_ENTRY_POINT_MISMATCH      = 8306,
     VKD3D_SHADER_WARNING_DXIL_INVALID_MASK              = 8307,
     VKD3D_SHADER_WARNING_DXIL_INVALID_OPERATION         = 8308,
+    VKD3D_SHADER_WARNING_DXIL_IGNORING_ATTACHMENT       = 8309,
 
     VKD3D_SHADER_ERROR_VSIR_NOT_IMPLEMENTED             = 9000,
     VKD3D_SHADER_ERROR_VSIR_INVALID_HANDLER             = 9001,
@@ -618,6 +620,11 @@ static inline bool data_type_is_bool(enum vkd3d_data_type data_type)
     return data_type == VKD3D_DATA_BOOL;
 }
 
+static inline bool data_type_is_64_bit(enum vkd3d_data_type data_type)
+{
+    return data_type == VKD3D_DATA_DOUBLE || data_type == VKD3D_DATA_UINT64;
+}
+
 enum vsir_dimension
 {
     VSIR_DIMENSION_NONE,
@@ -740,6 +747,7 @@ enum vkd3d_tessellator_domain
 #define VKD3DSI_RESINFO_UINT            0x2
 #define VKD3DSI_SAMPLE_INFO_UINT        0x1
 #define VKD3DSI_SAMPLER_COMPARISON_MODE 0x1
+#define VKD3DSI_SHIFT_UNMASKED          0x1
 
 #define VKD3DSI_PRECISE_X         0x100
 #define VKD3DSI_PRECISE_Y         0x200
@@ -793,10 +801,12 @@ struct vkd3d_shader_version
 
 struct vkd3d_shader_immediate_constant_buffer
 {
+    unsigned int register_idx;
     enum vkd3d_data_type data_type;
     /* total count is element_count * component_count */
     unsigned int element_count;
     unsigned int component_count;
+    bool is_null;
     uint32_t data[];
 };
 
@@ -831,10 +841,10 @@ struct vkd3d_shader_register
     unsigned int alignment;
     union
     {
-        DWORD immconst_uint[VKD3D_VEC4_SIZE];
-        float immconst_float[VKD3D_VEC4_SIZE];
-        uint64_t immconst_uint64[VKD3D_DVEC2_SIZE];
-        double immconst_double[VKD3D_DVEC2_SIZE];
+        uint32_t immconst_u32[VKD3D_VEC4_SIZE];
+        float immconst_f32[VKD3D_VEC4_SIZE];
+        uint64_t immconst_u64[VKD3D_DVEC2_SIZE];
+        double immconst_f64[VKD3D_DVEC2_SIZE];
         unsigned fp_body_idx;
     } u;
 };
@@ -1311,9 +1321,15 @@ struct vkd3d_string_buffer_cache
     size_t count, max_count, capacity;
 };
 
+enum vsir_asm_dialect
+{
+    VSIR_ASM_VSIR,
+    VSIR_ASM_D3D,
+};
+
 enum vkd3d_result vkd3d_dxbc_binary_to_text(const struct vkd3d_shader_instruction_array *instructions,
         const struct vkd3d_shader_version *shader_version, const struct vkd3d_shader_compile_info *compile_info,
-        struct vkd3d_shader_code *out);
+        struct vkd3d_shader_code *out, enum vsir_asm_dialect dialect);
 void vkd3d_string_buffer_cleanup(struct vkd3d_string_buffer *buffer);
 struct vkd3d_string_buffer *vkd3d_string_buffer_get(struct vkd3d_string_buffer_cache *list);
 void vkd3d_string_buffer_init(struct vkd3d_string_buffer *buffer);
@@ -1382,6 +1398,8 @@ void vkd3d_shader_verror(struct vkd3d_shader_message_context *context, const str
         enum vkd3d_shader_error error, const char *format, va_list args);
 void vkd3d_shader_vnote(struct vkd3d_shader_message_context *context, const struct vkd3d_shader_location *location,
         enum vkd3d_shader_log_level level, const char *format, va_list args);
+void vkd3d_shader_warning(struct vkd3d_shader_message_context *context, const struct vkd3d_shader_location *location,
+        enum vkd3d_shader_error error, const char *format, ...) VKD3D_PRINTF_FUNC(4, 5);
 void vkd3d_shader_vwarning(struct vkd3d_shader_message_context *context, const struct vkd3d_shader_location *location,
         enum vkd3d_shader_error error, const char *format, va_list args);
 
@@ -1444,6 +1462,8 @@ static inline enum vkd3d_shader_component_type vkd3d_component_type_from_data_ty
             return VKD3D_SHADER_COMPONENT_INT;
         case VKD3D_DATA_DOUBLE:
             return VKD3D_SHADER_COMPONENT_DOUBLE;
+        case VKD3D_DATA_UINT64:
+            return VKD3D_SHADER_COMPONENT_UINT64;
         case VKD3D_DATA_BOOL:
             return VKD3D_SHADER_COMPONENT_BOOL;
         default:
@@ -1505,7 +1525,7 @@ static inline enum vkd3d_shader_input_sysval_semantic vkd3d_siv_from_sysval(enum
     return vkd3d_siv_from_sysval_indexed(sysval, 0);
 }
 
-static inline unsigned int vkd3d_write_mask_get_component_idx(DWORD write_mask)
+static inline unsigned int vsir_write_mask_get_component_idx(uint32_t write_mask)
 {
     unsigned int i;
 
@@ -1520,7 +1540,7 @@ static inline unsigned int vkd3d_write_mask_get_component_idx(DWORD write_mask)
     return 0;
 }
 
-static inline unsigned int vkd3d_write_mask_component_count(DWORD write_mask)
+static inline unsigned int vsir_write_mask_component_count(uint32_t write_mask)
 {
     unsigned int count = vkd3d_popcount(write_mask & VKD3DSP_WRITEMASK_ALL);
     assert(1 <= count && count <= VKD3D_VEC4_SIZE);
@@ -1533,32 +1553,30 @@ 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)
+static inline uint32_t vsir_write_mask_64_from_32(uint32_t write_mask32)
 {
-    unsigned int write_mask64 = write_mask32 | (write_mask32 >> 1);
+    uint32_t 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)
+static inline uint32_t vsir_write_mask_32_from_64(uint32_t write_mask64)
 {
-    unsigned int write_mask32 = (write_mask64 | (write_mask64 << 1))
+    uint32_t write_mask32 = (write_mask64 | (write_mask64 << 1))
             & (VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_2);
     return write_mask32 | (write_mask32 << 1);
 }
 
-static inline unsigned int vkd3d_swizzle_get_component(DWORD swizzle,
-        unsigned int idx)
+static inline unsigned int vsir_swizzle_get_component(uint32_t swizzle, unsigned int idx)
 {
     return (swizzle >> VKD3D_SHADER_SWIZZLE_SHIFT(idx)) & VKD3D_SHADER_SWIZZLE_MASK;
 }
 
-static inline unsigned int vkd3d_swizzle_get_component64(DWORD swizzle,
-        unsigned int idx)
+static inline unsigned int vsir_swizzle_get_component64(uint32_t swizzle, unsigned int idx)
 {
     return ((swizzle >> VKD3D_SHADER_SWIZZLE_SHIFT(idx * 2)) & VKD3D_SHADER_SWIZZLE_MASK) / 2u;
 }
 
-static inline unsigned int vkd3d_compact_swizzle(unsigned int swizzle, unsigned int write_mask)
+static inline unsigned int vkd3d_compact_swizzle(uint32_t swizzle, uint32_t write_mask)
 {
     unsigned int i, compacted_swizzle = 0;
 
@@ -1567,7 +1585,7 @@ static inline unsigned int vkd3d_compact_swizzle(unsigned int swizzle, unsigned
         if (write_mask & (VKD3DSP_WRITEMASK_0 << i))
         {
             compacted_swizzle <<= VKD3D_SHADER_SWIZZLE_SHIFT(1);
-            compacted_swizzle |= vkd3d_swizzle_get_component(swizzle, i);
+            compacted_swizzle |= vsir_swizzle_get_component(swizzle, i);
         }
     }
 
diff --git a/libs/vkd3d/libs/vkd3d/command.c b/libs/vkd3d/libs/vkd3d/command.c
index 15c8317b191..549f6a45ffb 100644
--- a/libs/vkd3d/libs/vkd3d/command.c
+++ b/libs/vkd3d/libs/vkd3d/command.c
@@ -2644,6 +2644,8 @@ static bool d3d12_command_list_update_compute_pipeline(struct d3d12_command_list
 {
     const struct vkd3d_vk_device_procs *vk_procs = &list->device->vk_procs;
 
+    vkd3d_cond_signal(&list->device->worker_cond);
+
     if (list->current_pipeline != VK_NULL_HANDLE)
         return true;
 
@@ -2665,6 +2667,8 @@ static bool d3d12_command_list_update_graphics_pipeline(struct d3d12_command_lis
     VkRenderPass vk_render_pass;
     VkPipeline vk_pipeline;
 
+    vkd3d_cond_signal(&list->device->worker_cond);
+
     if (list->current_pipeline != VK_NULL_HANDLE)
         return true;
 
diff --git a/libs/vkd3d/libs/vkd3d/device.c b/libs/vkd3d/libs/vkd3d/device.c
index 5c801ca4676..69a46e9188a 100644
--- a/libs/vkd3d/libs/vkd3d/device.c
+++ b/libs/vkd3d/libs/vkd3d/device.c
@@ -2056,7 +2056,7 @@ static HRESULT d3d12_device_init_pipeline_cache(struct d3d12_device *device)
     VkPipelineCacheCreateInfo cache_info;
     VkResult vr;
 
-    vkd3d_mutex_init(&device->mutex);
+    vkd3d_mutex_init(&device->pipeline_cache_mutex);
 
     cache_info.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;
     cache_info.pNext = NULL;
@@ -2080,7 +2080,7 @@ static void d3d12_device_destroy_pipeline_cache(struct d3d12_device *device)
     if (device->vk_pipeline_cache)
         VK_CALL(vkDestroyPipelineCache(device->vk_device, device->vk_pipeline_cache, NULL));
 
-    vkd3d_mutex_destroy(&device->mutex);
+    vkd3d_mutex_destroy(&device->pipeline_cache_mutex);
 }
 
 #define VKD3D_VA_FALLBACK_BASE      0x8000000000000000ull
@@ -2495,6 +2495,28 @@ static ULONG STDMETHODCALLTYPE d3d12_device_AddRef(ID3D12Device5 *iface)
     return refcount;
 }
 
+static HRESULT device_worker_stop(struct d3d12_device *device)
+{
+    HRESULT hr;
+
+    TRACE("device %p.\n", device);
+
+    vkd3d_mutex_lock(&device->worker_mutex);
+
+    device->worker_should_exit = true;
+    vkd3d_cond_signal(&device->worker_cond);
+
+    vkd3d_mutex_unlock(&device->worker_mutex);
+
+    if (FAILED(hr = vkd3d_join_thread(device->vkd3d_instance, &device->worker_thread)))
+        return hr;
+
+    vkd3d_mutex_destroy(&device->worker_mutex);
+    vkd3d_cond_destroy(&device->worker_cond);
+
+    return S_OK;
+}
+
 static ULONG STDMETHODCALLTYPE d3d12_device_Release(ID3D12Device5 *iface)
 {
     struct d3d12_device *device = impl_from_ID3D12Device5(iface);
@@ -2520,6 +2542,9 @@ static ULONG STDMETHODCALLTYPE d3d12_device_Release(ID3D12Device5 *iface)
         d3d12_device_destroy_vkd3d_queues(device);
         vkd3d_desc_object_cache_cleanup(&device->view_desc_cache);
         vkd3d_desc_object_cache_cleanup(&device->cbuffer_desc_cache);
+        if (device->use_vk_heaps)
+            device_worker_stop(device);
+        vkd3d_free(device->heaps);
         VK_CALL(vkDestroyDevice(device->vk_device, NULL));
         if (device->parent)
             IUnknown_Release(device->parent);
@@ -4251,6 +4276,40 @@ struct d3d12_device *unsafe_impl_from_ID3D12Device5(ID3D12Device5 *iface)
     return impl_from_ID3D12Device5(iface);
 }
 
+static void *device_worker_main(void *arg)
+{
+    struct d3d12_descriptor_heap *heap;
+    struct d3d12_device *device = arg;
+    size_t i;
+
+    vkd3d_set_thread_name("device_worker");
+
+    vkd3d_mutex_lock(&device->worker_mutex);
+
+    while (!device->worker_should_exit)
+    {
+        for (i = 0; i < device->heap_count; ++i)
+        {
+            /* Descriptor updates are not written to Vulkan descriptor sets until a command list
+             * is submitted to a queue, while the client is free to write d3d12 descriptors earlier,
+             * from any thread. This causes a delay right before command list execution, so
+             * handling these updates in a worker thread can speed up execution significantly. */
+            heap = device->heaps[i];
+            if (heap->dirty_list_head == UINT_MAX)
+                continue;
+            vkd3d_mutex_lock(&heap->vk_sets_mutex);
+            d3d12_desc_flush_vk_heap_updates_locked(heap, device);
+            vkd3d_mutex_unlock(&heap->vk_sets_mutex);
+        }
+
+        vkd3d_cond_wait(&device->worker_cond, &device->worker_mutex);
+    }
+
+    vkd3d_mutex_unlock(&device->worker_mutex);
+
+    return NULL;
+}
+
 static HRESULT d3d12_device_init(struct d3d12_device *device,
         struct vkd3d_instance *instance, const struct vkd3d_device_create_info *create_info)
 {
@@ -4270,6 +4329,14 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
 
     device->vk_device = VK_NULL_HANDLE;
 
+    device->heaps = NULL;
+    device->heap_capacity = 0;
+    device->heap_count = 0;
+    memset(&device->worker_thread, 0, sizeof(device->worker_thread));
+    device->worker_should_exit = false;
+    vkd3d_mutex_init(&device->worker_mutex);
+    vkd3d_cond_init(&device->worker_cond);
+
     if (FAILED(hr = vkd3d_create_vk_device(device, create_info)))
         goto out_free_instance;
 
@@ -4291,6 +4358,13 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
     if (FAILED(hr = vkd3d_vk_descriptor_heap_layouts_init(device)))
         goto out_cleanup_uav_clear_state;
 
+    if (device->use_vk_heaps && FAILED(hr = vkd3d_create_thread(device->vkd3d_instance,
+            device_worker_main, device, &device->worker_thread)))
+    {
+        WARN("Failed to create worker thread, hr %#x.\n", hr);
+        goto out_cleanup_descriptor_heap_layouts;
+    }
+
     vkd3d_render_pass_cache_init(&device->render_pass_cache);
     vkd3d_gpu_va_allocator_init(&device->gpu_va_allocator);
     vkd3d_time_domains_init(device);
@@ -4308,6 +4382,8 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
 
     return S_OK;
 
+out_cleanup_descriptor_heap_layouts:
+    vkd3d_vk_descriptor_heap_layouts_cleanup(device);
 out_cleanup_uav_clear_state:
     vkd3d_uav_clear_state_cleanup(&device->uav_clear_state, device);
 out_destroy_null_resources:
@@ -4361,6 +4437,40 @@ void d3d12_device_mark_as_removed(struct d3d12_device *device, HRESULT reason,
     device->removed_reason = reason;
 }
 
+HRESULT d3d12_device_add_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap)
+{
+    vkd3d_mutex_lock(&device->worker_mutex);
+
+    if (!vkd3d_array_reserve((void **)&device->heaps, &device->heap_capacity, device->heap_count + 1,
+            sizeof(*device->heaps)))
+    {
+        vkd3d_mutex_unlock(&device->worker_mutex);
+        return E_OUTOFMEMORY;
+    }
+    device->heaps[device->heap_count++] = heap;
+
+    vkd3d_mutex_unlock(&device->worker_mutex);
+
+    return S_OK;
+}
+
+void d3d12_device_remove_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap)
+{
+    size_t i;
+
+    vkd3d_mutex_lock(&device->worker_mutex);
+
+    for (i = 0; i < device->heap_count; ++i)
+    {
+        if (device->heaps[i] == heap)
+        {
+            device->heaps[i] = device->heaps[--device->heap_count];
+            break;
+        }
+    }
+
+    vkd3d_mutex_unlock(&device->worker_mutex);
+}
 
 #ifdef _WIN32
 struct thread_data
diff --git a/libs/vkd3d/libs/vkd3d/resource.c b/libs/vkd3d/libs/vkd3d/resource.c
index abbdfbe2015..609c67102a6 100644
--- a/libs/vkd3d/libs/vkd3d/resource.c
+++ b/libs/vkd3d/libs/vkd3d/resource.c
@@ -2632,20 +2632,18 @@ void d3d12_desc_flush_vk_heap_updates_locked(struct d3d12_descriptor_heap *descr
     for (; i != UINT_MAX; i = next)
     {
         src = &descriptors[i];
-        next = (int)src->next >> 1;
+        next = vkd3d_atomic_exchange(&src->next, 0);
+        next = (int)next >> 1;
 
+        /* A race exists here between updating src->next and getting the current object. The best
+         * we can do is get the object last, which may result in a harmless rewrite later. */
         u.object = d3d12_desc_get_object_ref(src, device);
 
         if (!u.object)
-        {
-            vkd3d_atomic_exchange(&src->next, 0);
             continue;
-        }
 
         writes.held_refs[writes.held_ref_count++] = u.object;
         d3d12_desc_write_vk_heap(descriptor_heap, i, &writes, u.object, device);
-
-        vkd3d_atomic_exchange(&src->next, 0);
     }
 
     /* Avoid thunk calls wherever possible. */
@@ -3997,6 +3995,9 @@ static ULONG STDMETHODCALLTYPE d3d12_descriptor_heap_Release(ID3D12DescriptorHea
             {
                 struct d3d12_desc *descriptors = (struct d3d12_desc *)heap->descriptors;
 
+                if (heap->use_vk_heaps)
+                    d3d12_device_remove_descriptor_heap(device, heap);
+
                 for (i = 0; i < heap->desc.NumDescriptors; ++i)
                 {
                     d3d12_desc_destroy(&descriptors[i], device);
@@ -4320,6 +4321,12 @@ HRESULT d3d12_descriptor_heap_create(struct d3d12_device *device,
             dst[i].next = 0;
         }
         object->dirty_list_head = UINT_MAX;
+
+        if (object->use_vk_heaps && FAILED(hr = d3d12_device_add_descriptor_heap(device, object)))
+        {
+            vkd3d_free(object);
+            return hr;
+        }
     }
     else
     {
diff --git a/libs/vkd3d/libs/vkd3d/state.c b/libs/vkd3d/libs/vkd3d/state.c
index fc3187f4bea..1457ddf9c7f 100644
--- a/libs/vkd3d/libs/vkd3d/state.c
+++ b/libs/vkd3d/libs/vkd3d/state.c
@@ -515,7 +515,7 @@ static HRESULT d3d12_root_signature_init_push_constants(struct d3d12_root_signat
         assert(p->ShaderVisibility <= D3D12_SHADER_VISIBILITY_PIXEL);
         push_constants[p->ShaderVisibility].stageFlags = use_vk_heaps ? VK_SHADER_STAGE_ALL
                 : stage_flags_from_visibility(p->ShaderVisibility);
-        push_constants[p->ShaderVisibility].size += p->u.Constants.Num32BitValues * sizeof(uint32_t);
+        push_constants[p->ShaderVisibility].size += align(p->u.Constants.Num32BitValues, 4) * sizeof(uint32_t);
     }
     if (push_constants[D3D12_SHADER_VISIBILITY_ALL].size)
     {
@@ -564,7 +564,7 @@ static HRESULT d3d12_root_signature_init_push_constants(struct d3d12_root_signat
 
         idx = push_constant_count == 1 ? 0 : p->ShaderVisibility;
         offset = push_constants_offset[idx];
-        push_constants_offset[idx] += p->u.Constants.Num32BitValues * sizeof(uint32_t);
+        push_constants_offset[idx] += align(p->u.Constants.Num32BitValues, 4) * sizeof(uint32_t);
 
         root_signature->parameters[i].parameter_type = p->ParameterType;
         root_constant->stage_flags = push_constant_count == 1
@@ -1691,7 +1691,7 @@ HRESULT vkd3d_render_pass_cache_find(struct vkd3d_render_pass_cache *cache,
     HRESULT hr = S_OK;
     unsigned int i;
 
-    vkd3d_mutex_lock(&device->mutex);
+    vkd3d_mutex_lock(&device->pipeline_cache_mutex);
 
     for (i = 0; i < cache->render_pass_count; ++i)
     {
@@ -1708,7 +1708,7 @@ HRESULT vkd3d_render_pass_cache_find(struct vkd3d_render_pass_cache *cache,
     if (!found)
         hr = vkd3d_render_pass_cache_create_pass_locked(cache, device, key, vk_render_pass);
 
-    vkd3d_mutex_unlock(&device->mutex);
+    vkd3d_mutex_unlock(&device->pipeline_cache_mutex);
 
     return hr;
 }
@@ -2129,6 +2129,16 @@ static inline unsigned int typed_uav_compile_option(const struct d3d12_device *d
             : VKD3D_SHADER_COMPILE_OPTION_TYPED_UAV_READ_FORMAT_R32;
 }
 
+static unsigned int feature_flags_compile_option(const struct d3d12_device *device)
+{
+    unsigned int flags = 0;
+
+    if (device->feature_options1.Int64ShaderOps)
+        flags |= VKD3D_SHADER_COMPILE_OPTION_FEATURE_INT64;
+
+    return flags;
+}
+
 static HRESULT create_shader_stage(struct d3d12_device *device,
         struct VkPipelineShaderStageCreateInfo *stage_desc, enum VkShaderStageFlagBits stage,
         const D3D12_SHADER_BYTECODE *code, const struct vkd3d_shader_interface_info *shader_interface)
@@ -2145,6 +2155,7 @@ static HRESULT create_shader_stage(struct d3d12_device *device,
         {VKD3D_SHADER_COMPILE_OPTION_API_VERSION, VKD3D_SHADER_API_VERSION_1_10},
         {VKD3D_SHADER_COMPILE_OPTION_TYPED_UAV, typed_uav_compile_option(device)},
         {VKD3D_SHADER_COMPILE_OPTION_WRITE_TESS_GEOM_POINT_SIZE, 0},
+        {VKD3D_SHADER_COMPILE_OPTION_FEATURE, feature_flags_compile_option(device)},
     };
 
     stage_desc->sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
@@ -3615,7 +3626,7 @@ static VkPipeline d3d12_pipeline_state_find_compiled_pipeline(const struct d3d12
 
     *vk_render_pass = VK_NULL_HANDLE;
 
-    vkd3d_mutex_lock(&device->mutex);
+    vkd3d_mutex_lock(&device->pipeline_cache_mutex);
 
     LIST_FOR_EACH_ENTRY(current, &graphics->compiled_pipelines, struct vkd3d_compiled_pipeline, entry)
     {
@@ -3627,7 +3638,7 @@ static VkPipeline d3d12_pipeline_state_find_compiled_pipeline(const struct d3d12
         }
     }
 
-    vkd3d_mutex_unlock(&device->mutex);
+    vkd3d_mutex_unlock(&device->pipeline_cache_mutex);
 
     return vk_pipeline;
 }
@@ -3646,7 +3657,7 @@ static bool d3d12_pipeline_state_put_pipeline_to_cache(struct d3d12_pipeline_sta
     compiled_pipeline->vk_pipeline = vk_pipeline;
     compiled_pipeline->vk_render_pass = vk_render_pass;
 
-    vkd3d_mutex_lock(&device->mutex);
+    vkd3d_mutex_lock(&device->pipeline_cache_mutex);
 
     LIST_FOR_EACH_ENTRY(current, &graphics->compiled_pipelines, struct vkd3d_compiled_pipeline, entry)
     {
@@ -3661,7 +3672,7 @@ static bool d3d12_pipeline_state_put_pipeline_to_cache(struct d3d12_pipeline_sta
     if (compiled_pipeline)
         list_add_tail(&graphics->compiled_pipelines, &compiled_pipeline->entry);
 
-    vkd3d_mutex_unlock(&device->mutex);
+    vkd3d_mutex_unlock(&device->pipeline_cache_mutex);
     return compiled_pipeline;
 }
 
diff --git a/libs/vkd3d/libs/vkd3d/utils.c b/libs/vkd3d/libs/vkd3d/utils.c
index 9b28068be51..5ebe1b63e99 100644
--- a/libs/vkd3d/libs/vkd3d/utils.c
+++ b/libs/vkd3d/libs/vkd3d/utils.c
@@ -539,6 +539,7 @@ bool is_valid_feature_level(D3D_FEATURE_LEVEL feature_level)
 {
     static const D3D_FEATURE_LEVEL valid_feature_levels[] =
     {
+        D3D_FEATURE_LEVEL_12_2,
         D3D_FEATURE_LEVEL_12_1,
         D3D_FEATURE_LEVEL_12_0,
         D3D_FEATURE_LEVEL_11_1,
@@ -548,6 +549,7 @@ bool is_valid_feature_level(D3D_FEATURE_LEVEL feature_level)
         D3D_FEATURE_LEVEL_9_3,
         D3D_FEATURE_LEVEL_9_2,
         D3D_FEATURE_LEVEL_9_1,
+        D3D_FEATURE_LEVEL_1_0_CORE,
     };
     unsigned int i;
 
diff --git a/libs/vkd3d/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/libs/vkd3d/vkd3d_private.h
index f06f564d6ea..bd7e7290d9e 100644
--- a/libs/vkd3d/libs/vkd3d/vkd3d_private.h
+++ b/libs/vkd3d/libs/vkd3d/vkd3d_private.h
@@ -1757,9 +1757,23 @@ struct d3d12_device
 
     struct vkd3d_gpu_va_allocator gpu_va_allocator;
 
-    struct vkd3d_mutex mutex;
     struct vkd3d_desc_object_cache view_desc_cache;
     struct vkd3d_desc_object_cache cbuffer_desc_cache;
+
+    VkDescriptorPoolSize vk_pool_sizes[VKD3D_DESCRIPTOR_POOL_COUNT];
+    unsigned int vk_pool_count;
+    struct vkd3d_vk_descriptor_heap_layout vk_descriptor_heap_layouts[VKD3D_SET_INDEX_COUNT];
+    bool use_vk_heaps;
+
+    struct d3d12_descriptor_heap **heaps;
+    size_t heap_capacity;
+    size_t heap_count;
+    union vkd3d_thread_handle worker_thread;
+    struct vkd3d_mutex worker_mutex;
+    struct vkd3d_cond worker_cond;
+    bool worker_should_exit;
+
+    struct vkd3d_mutex pipeline_cache_mutex;
     struct vkd3d_render_pass_cache render_pass_cache;
     VkPipelineCache vk_pipeline_cache;
 
@@ -1799,11 +1813,6 @@ struct d3d12_device
     const struct vkd3d_format_compatibility_list *format_compatibility_lists;
     struct vkd3d_null_resources null_resources;
     struct vkd3d_uav_clear_state uav_clear_state;
-
-    VkDescriptorPoolSize vk_pool_sizes[VKD3D_DESCRIPTOR_POOL_COUNT];
-    unsigned int vk_pool_count;
-    struct vkd3d_vk_descriptor_heap_layout vk_descriptor_heap_layouts[VKD3D_SET_INDEX_COUNT];
-    bool use_vk_heaps;
 };
 
 HRESULT d3d12_device_create(struct vkd3d_instance *instance,
@@ -1813,6 +1822,8 @@ bool d3d12_device_is_uma(struct d3d12_device *device, bool *coherent);
 void d3d12_device_mark_as_removed(struct d3d12_device *device, HRESULT reason,
         const char *message, ...) VKD3D_PRINTF_FUNC(3, 4);
 struct d3d12_device *unsafe_impl_from_ID3D12Device5(ID3D12Device5 *iface);
+HRESULT d3d12_device_add_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap);
+void d3d12_device_remove_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap);
 
 static inline HRESULT d3d12_device_query_interface(struct d3d12_device *device, REFIID iid, void **object)
 {
-- 
2.43.0