From d368d18527c7d44c2aab4b5443cb21980466da15 Mon Sep 17 00:00:00 2001
From: Shaun Ren <sren@codeweavers.com>
Date: Mon, 26 May 2025 21:34:06 -0400
Subject: [PATCH] vkd3d-shader/hlsl: Generate vsir instructions for indexed
 output streams in geometry shaders.

---
 libs/vkd3d-shader/hlsl_codegen.c              | 29 ++++++++++++++-----
 libs/vkd3d-shader/tpf.c                       |  2 ++
 tests/hlsl/geometry-shader-syntax.shader_test |  6 ++--
 3 files changed, 27 insertions(+), 10 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index c9c5da22d..ef412c9d8 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -8485,6 +8485,14 @@ static bool sm4_generate_vsir_reg_from_deref(struct hlsl_ctx *ctx, struct vsir_p
             VKD3D_ASSERT(regset == HLSL_REGSET_SAMPLERS);
             *writemask = VKD3DSP_WRITEMASK_ALL;
         }
+        else if (regset == HLSL_REGSET_STREAM_OUTPUTS)
+        {
+            reg->type = VKD3DSPR_STREAM;
+            reg->dimension = VSIR_DIMENSION_NONE;
+            reg->idx[0].offset = var->regs[HLSL_REGSET_STREAM_OUTPUTS].index;
+            reg->idx_count = 1;
+            *writemask = VKD3DSP_WRITEMASK_ALL;
+        }
         else
         {
             unsigned int offset = deref->const_offset + var->buffer_offset;
@@ -10835,20 +10843,27 @@ static bool sm4_generate_vsir_instr_resource_store(struct hlsl_ctx *ctx,
 
     if (store->store_type != HLSL_RESOURCE_STORE)
     {
-        enum vkd3d_shader_opcode opcode = store->store_type == HLSL_RESOURCE_STREAM_APPEND
-                ? VKD3DSIH_EMIT : VKD3DSIH_CUT;
+        enum vkd3d_shader_opcode opcode;
 
         VKD3D_ASSERT(!store->value.node && !store->coords.node);
         VKD3D_ASSERT(store->resource.var->regs[HLSL_REGSET_STREAM_OUTPUTS].allocated);
 
-        if (store->resource.var->regs[HLSL_REGSET_STREAM_OUTPUTS].index)
+        if (hlsl_version_lt(ctx, 5, 0))
         {
-            hlsl_fixme(ctx, &instr->loc, "Stream output operation with a nonzero stream index.");
-            return false;
+            opcode = store->store_type == HLSL_RESOURCE_STREAM_APPEND ? VKD3DSIH_EMIT : VKD3DSIH_CUT;
+            ins = generate_vsir_add_program_instruction(ctx, program, &store->node.loc, opcode, 0, 0);
+            return !!ins;
         }
 
-        ins = generate_vsir_add_program_instruction(ctx, program, &store->node.loc, opcode, 0, 0);
-        return !!ins;
+        opcode = store->store_type == HLSL_RESOURCE_STREAM_APPEND ? VKD3DSIH_EMIT_STREAM : VKD3DSIH_CUT_STREAM;
+        if (!(ins = generate_vsir_add_program_instruction(ctx, program, &store->node.loc, opcode, 0, 1)))
+            return false;
+
+        if (!sm4_generate_vsir_init_src_param_from_deref(ctx, program, &ins->src[0],
+                &store->resource, VKD3DSP_WRITEMASK_ALL, &instr->loc))
+            return false;
+
+        return true;
     }
 
     if (!store->resource.var->is_uniform)
diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c
index 8ed9fa031..19ca97cb4 100644
--- a/libs/vkd3d-shader/tpf.c
+++ b/libs/vkd3d-shader/tpf.c
@@ -4176,6 +4176,7 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
         case VKD3DSIH_CASE:
         case VKD3DSIH_CONTINUE:
         case VKD3DSIH_CUT:
+        case VKD3DSIH_CUT_STREAM:
         case VKD3DSIH_DCL_STREAM:
         case VKD3DSIH_DEFAULT:
         case VKD3DSIH_DISCARD:
@@ -4191,6 +4192,7 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
         case VKD3DSIH_DSY_FINE:
         case VKD3DSIH_ELSE:
         case VKD3DSIH_EMIT:
+        case VKD3DSIH_EMIT_STREAM:
         case VKD3DSIH_ENDIF:
         case VKD3DSIH_ENDLOOP:
         case VKD3DSIH_ENDSWITCH:
diff --git a/tests/hlsl/geometry-shader-syntax.shader_test b/tests/hlsl/geometry-shader-syntax.shader_test
index be481a534..6fa833011 100644
--- a/tests/hlsl/geometry-shader-syntax.shader_test
+++ b/tests/hlsl/geometry-shader-syntax.shader_test
@@ -541,7 +541,7 @@ gs_data main(point gs_data vin[1], inout TriangleStream<gs_data> vout)
 }
 
 % Output streams with nonzero stream indices are only allowed with point streams, and with SM >= 5.
-[geometry shader fail(sm<5) todo(sm>=5)]
+[geometry shader fail(sm<5)]
 struct gs_data
 {
     float4 pos : SV_POSITION;
@@ -610,7 +610,7 @@ void main(point gs_data vin[1], inout PointStream<gs_data> vout1, inout Triangle
     vout1.RestartStrip();
 }
 
-[geometry shader fail(sm<5) todo(sm>=5)]
+[geometry shader fail(sm<5)]
 struct gs_data
 {
     float4 pos : SV_POSITION;
@@ -623,7 +623,7 @@ void main(point gs_data vin[1], inout PointStream<gs_data> vout1, inout Triangle
 }
 
 % A maximum of 4 streams are allowed.
-[geometry shader fail(sm<5) todo(sm>=5)]
+[geometry shader fail(sm<5)]
 struct gs_data
 {
     float4 pos : SV_POSITION;