diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 4eb96279..5a0cc9bb 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -9643,6 +9643,37 @@ static void generate_vsir_scan_global_flags(struct hlsl_ctx *ctx,
         program->global_flags |= VKD3DSGF_FORCE_EARLY_DEPTH_STENCIL;
 }
 
+static void sm4_generate_vsir_add_dcl_constant_buffer(struct hlsl_ctx *ctx,
+        struct vsir_program *program, const struct hlsl_buffer *cbuffer)
+{
+    unsigned int array_first = cbuffer->reg.index;
+    unsigned int array_last = cbuffer->reg.index; /* FIXME: array end. */
+    struct vkd3d_shader_src_param *src_param;
+    struct vkd3d_shader_instruction *ins;
+
+    if (!(ins = generate_vsir_add_program_instruction(ctx, program, &cbuffer->loc, VKD3DSIH_DCL_CONSTANT_BUFFER, 0, 0)))
+    {
+        ctx->result = VKD3D_ERROR_OUT_OF_MEMORY;
+        return;
+    }
+
+    ins->declaration.cb.size = cbuffer->size;
+
+    src_param = &ins->declaration.cb.src;
+    vsir_src_param_init(src_param, VKD3DSPR_CONSTBUFFER, VKD3D_DATA_FLOAT, 0);
+    src_param->reg.dimension = VSIR_DIMENSION_VEC4;
+    src_param->swizzle = VKD3D_SHADER_NO_SWIZZLE;
+
+    ins->declaration.cb.range.space = cbuffer->reg.space;
+    ins->declaration.cb.range.first = array_first;
+    ins->declaration.cb.range.last = array_last;
+
+    src_param->reg.idx[0].offset = cbuffer->reg.id;
+    src_param->reg.idx[1].offset = array_first;
+    src_param->reg.idx[2].offset = array_last;
+    src_param->reg.idx_count = 3;
+}
+
 static void sm4_generate_vsir_add_dcl_sampler(struct hlsl_ctx *ctx,
         struct vsir_program *program, const struct extern_resource *resource)
 {
@@ -9860,6 +9891,7 @@ static void sm4_generate_vsir(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl
     struct vkd3d_shader_version version = {0};
     struct extern_resource *extern_resources;
     unsigned int extern_resources_count;
+    const struct hlsl_buffer *cbuffer;
 
     version.major = ctx->profile->major_version;
     version.minor = ctx->profile->minor_version;
@@ -9882,6 +9914,12 @@ static void sm4_generate_vsir(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl
         program->thread_group_size.z = ctx->thread_count[2];
     }
 
+    LIST_FOR_EACH_ENTRY(cbuffer, &ctx->buffers, struct hlsl_buffer, entry)
+    {
+        if (cbuffer->reg.allocated)
+            sm4_generate_vsir_add_dcl_constant_buffer(ctx, program, cbuffer);
+    }
+
     extern_resources = sm4_get_extern_resources(ctx, &extern_resources_count);
     for (unsigned int i = 0; i < extern_resources_count; ++i)
     {
diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c
index c944ffbd..a26308a9 100644
--- a/libs/vkd3d-shader/tpf.c
+++ b/libs/vkd3d-shader/tpf.c
@@ -4312,38 +4312,40 @@ static void write_sm4_instruction(const struct tpf_compiler *tpf, const struct s
     sm4_update_stat_counters(tpf, instr);
 }
 
-static void write_sm4_dcl_constant_buffer(const struct tpf_compiler *tpf, const struct hlsl_buffer *cbuffer)
+static void tpf_dcl_constant_buffer(const struct tpf_compiler *tpf, const struct vkd3d_shader_instruction *ins)
 {
-    size_t size = (cbuffer->used_size + 3) / 4;
+    const struct vkd3d_shader_constant_buffer *cb = &ins->declaration.cb;
+    size_t size = (cb->size + 3) / 4;
 
     struct sm4_instruction instr =
     {
         .opcode = VKD3D_SM4_OP_DCL_CONSTANT_BUFFER,
 
-        .srcs[0].reg.dimension = VSIR_DIMENSION_VEC4,
-        .srcs[0].reg.type = VKD3DSPR_CONSTBUFFER,
-        .srcs[0].swizzle = VKD3D_SHADER_NO_SWIZZLE,
+        .srcs[0] = cb->src,
         .src_count = 1,
     };
 
-    if (hlsl_version_ge(tpf->ctx, 5, 1))
+    if (vkd3d_shader_ver_ge(&tpf->program->shader_version, 5, 1))
     {
-        instr.srcs[0].reg.idx[0].offset = cbuffer->reg.id;
-        instr.srcs[0].reg.idx[1].offset = cbuffer->reg.index;
-        instr.srcs[0].reg.idx[2].offset = cbuffer->reg.index; /* FIXME: array end */
+        instr.srcs[0].reg.idx[0].offset = cb->src.reg.idx[0].offset;
+        instr.srcs[0].reg.idx[1].offset = cb->range.first;
+        instr.srcs[0].reg.idx[2].offset = cb->range.last;
         instr.srcs[0].reg.idx_count = 3;
 
         instr.idx[0] = size;
-        instr.idx[1] = cbuffer->reg.space;
+        instr.idx[1] = cb->range.space;
         instr.idx_count = 2;
     }
     else
     {
-        instr.srcs[0].reg.idx[0].offset = cbuffer->reg.index;
+        instr.srcs[0].reg.idx[0].offset = cb->range.first;
         instr.srcs[0].reg.idx[1].offset = size;
         instr.srcs[0].reg.idx_count = 2;
     }
 
+    if (ins->flags & VKD3DSI_INDEXED_DYNAMIC)
+        instr.extra_bits |= VKD3D_SM4_INDEX_TYPE_MASK;
+
     write_sm4_instruction(tpf, &instr);
 }
 
@@ -4664,6 +4666,10 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
 {
     switch (ins->opcode)
     {
+        case VKD3DSIH_DCL_CONSTANT_BUFFER:
+            tpf_dcl_constant_buffer(tpf, ins);
+            break;
+
         case VKD3DSIH_DCL_TEMPS:
             tpf_dcl_temps(tpf, ins->declaration.count);
             break;
@@ -4831,7 +4837,6 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
 {
     const struct vkd3d_shader_version *version = &tpf->program->shader_version;
     struct vkd3d_bytecode_buffer buffer = {0};
-    const struct hlsl_buffer *cbuffer;
     struct hlsl_ctx *ctx = tpf->ctx;
     size_t token_count_position;
 
@@ -4872,12 +4877,6 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
         tpf_write_dcl_tessellator_domain(tpf, ctx->domain);
     }
 
-    LIST_FOR_EACH_ENTRY(cbuffer, &ctx->buffers, struct hlsl_buffer, entry)
-    {
-        if (cbuffer->reg.allocated)
-            write_sm4_dcl_constant_buffer(tpf, cbuffer);
-    }
-
     tpf_write_program(tpf, tpf->program);
 
     set_u32(&buffer, token_count_position, bytecode_get_size(&buffer) / sizeof(uint32_t));