[PATCH vkd3d 6/8] vkd3d-shader/hlsl: Write the SM4 dcl_temps instruction.

[Zebediah Figura]

Signed-off-by: Zebediah Figura 
---
 libs/vkd3d-shader/hlsl.h         |  1 +
 libs/vkd3d-shader/hlsl_codegen.c |  4 ++++
 libs/vkd3d-shader/hlsl_sm4.c     | 16 ++++++++++++++++
 3 files changed, 21 insertions(+)

diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h
index 80cea512..9b579b72 100644
--- a/libs/vkd3d-shader/hlsl.h
+++ b/libs/vkd3d-shader/hlsl.h
@@ -469,6 +469,7 @@ struct hlsl_ctx
         struct hlsl_vec4 *values;
         size_t count, size;
     } constant_defs;
+    uint32_t temp_count;
 };

enum hlsl_error_level
diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index a1d6e8e9..c2aeb7b0 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -685,6 +685,7 @@ static void compute_liveness(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl
 struct liveness
 {
     size_t size;
+    uint32_t reg_count;
     struct
     {
         /* 0 if not live yet. */
@@ -747,6 +748,7 @@ static struct hlsl_reg allocate_register(struct hlsl_ctx *ctx, struct liveness *
     ret.id = component_idx / 4;
     ret.writemask = writemask;
     ret.allocated = true;
+    liveness->reg_count = max(liveness->reg_count, ret.id + 1);
     return ret;
 }

@@ -782,6 +784,7 @@ static struct hlsl_reg allocate_range(struct hlsl_ctx *ctx, struct liveness *liv
         liveness->regs[component_idx + i].last_read = last_read;
     ret.id = component_idx / 4;
     ret.allocated = true;
+    liveness->reg_count = max(liveness->reg_count, ret.id + align(component_count, 4));
     return ret;
 }

@@ -1008,6 +1011,7 @@ static void allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_ir_functio
 {
     struct liveness liveness = {0};
     allocate_temp_registers_recurse(ctx, entry_func->body, &liveness);
+    ctx->temp_count = liveness.reg_count;
 }

static void allocate_semantic_register(struct hlsl_ctx *ctx, struct hlsl_ir_var *var, unsigned int *counter, bool output)
diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c
index 92f854c0..a66bcaae 100644
--- a/libs/vkd3d-shader/hlsl_sm4.c
+++ b/libs/vkd3d-shader/hlsl_sm4.c
@@ -762,6 +762,19 @@ static void write_sm4_dcl_semantic(struct hlsl_ctx *ctx, struct vkd3d_bytecode_b
     write_sm4_instruction(buffer, &instr);
 }

+static void write_sm4_dcl_temps(struct vkd3d_bytecode_buffer *buffer, uint32_t temp_count)
+{
+    struct sm4_instruction instr =
+    {
+        .opcode = VKD3D_SM4_OP_DCL_TEMPS,
+
+        .idx = {temp_count},
+        .idx_count = 1,
+    };
+
+    write_sm4_instruction(buffer, &instr);
+}
+
 static void write_sm4_shdr(struct hlsl_ctx *ctx, struct dxbc_writer *dxbc)
 {
     const struct hlsl_profile_info *profile = ctx->profile;
@@ -797,6 +810,9 @@ static void write_sm4_shdr(struct hlsl_ctx *ctx, struct dxbc_writer *dxbc)
             write_sm4_dcl_semantic(ctx, &buffer, var);
     }

+    if (ctx->temp_count)
+        write_sm4_dcl_temps(&buffer, ctx->temp_count);
+
     dxbc_writer_add_section(dxbc, TAG_SHDR, buffer.data, buffer.size);
 }