aco: fix GS with no outputs

With NGG, ngg_gs_known_vtxcnt[0] would be false and ngg_gs_finale() would
assert.

With legacy GS, I don't know why the assertion was there.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7576>

src/amd/compiler/aco_instruction_selection.cpp

@@ -7020,7 +7020,7 @@ void ngg_gs_write_shader_query(isel_context *ctx, nir_intrinsic_instr *instr);
 void ngg_visit_set_vertex_and_primitive_count(isel_context *ctx, nir_intrinsic_instr *instr)
 {
    unsigned stream = nir_intrinsic_stream_id(instr);
-   if (!ctx->args->shader_info->gs.num_stream_output_components[stream])
+   if (stream > 0 && !ctx->args->shader_info->gs.num_stream_output_components[stream])
       return;
 
    ctx->ngg_gs_known_vtxcnt[stream] = true;
@@ -7048,7 +7048,6 @@ void visit_emit_vertex_with_counter(isel_context *ctx, nir_intrinsic_instr *inst
 
    unsigned num_components =
       ctx->program->info->gs.num_stream_output_components[stream];
-   assert(num_components);
 
    unsigned stride = 4u * num_components * ctx->shader->info.gs.vertices_out;
    unsigned stream_offset = 0;

src/amd/compiler/tests/helpers.cpp

@@ -479,6 +479,13 @@ void PipelineBuilder::add_stage(VkShaderStageFlagBits stage, VkShaderModule modu
    owned_stages |= stage;
 }
 
+void PipelineBuilder::add_stage(VkShaderStageFlagBits stage, QoShaderModuleCreateInfo module, const char *name)
+{
+   add_stage(stage, __qoCreateShaderModule(device, &module), name);
+   add_resource_decls(&module);
+   add_io_decls(&module);
+}
+
 void PipelineBuilder::add_vsfs(VkShaderModule vs, VkShaderModule fs)
 {
    add_stage(VK_SHADER_STAGE_VERTEX_BIT, vs);
@@ -487,11 +494,8 @@ void PipelineBuilder::add_vsfs(VkShaderModule vs, VkShaderModule fs)
 
 void PipelineBuilder::add_vsfs(QoShaderModuleCreateInfo vs, QoShaderModuleCreateInfo fs)
 {
-   add_vsfs(__qoCreateShaderModule(device, &vs), __qoCreateShaderModule(device, &fs));
-   add_resource_decls(&vs);
-   add_io_decls(&vs);
-   add_resource_decls(&fs);
-   add_io_decls(&fs);
+   add_stage(VK_SHADER_STAGE_VERTEX_BIT, vs);
+   add_stage(VK_SHADER_STAGE_FRAGMENT_BIT, fs);
 }
 
 void PipelineBuilder::add_cs(VkShaderModule cs)
@@ -501,8 +505,7 @@ void PipelineBuilder::add_cs(VkShaderModule cs)
 
 void PipelineBuilder::add_cs(QoShaderModuleCreateInfo cs)
 {
-   add_cs(__qoCreateShaderModule(device, &cs));
-   add_resource_decls(&cs);
+   add_stage(VK_SHADER_STAGE_COMPUTE_BIT, cs);
 }
 
 bool PipelineBuilder::is_compute() {

src/amd/compiler/tests/helpers.h

@@ -135,6 +135,7 @@ public:
    void add_io_decls(QoShaderModuleCreateInfo *module);
 
    void add_stage(VkShaderStageFlagBits stage, VkShaderModule module, const char *name="main");
+   void add_stage(VkShaderStageFlagBits stage, QoShaderModuleCreateInfo module, const char *name="main");
    void add_vsfs(VkShaderModule vs, VkShaderModule fs);
    void add_vsfs(QoShaderModuleCreateInfo vs, QoShaderModuleCreateInfo fs);
    void add_cs(VkShaderModule cs);

src/amd/compiler/tests/test_isel.cpp

@@ -78,3 +78,32 @@ BEGIN_TEST(isel.compute.simple)
       pbld.print_ir(VK_SHADER_STAGE_COMPUTE_BIT, "ACO IR", true);
    }
 END_TEST
+
+BEGIN_TEST(isel.gs.no_outputs)
+   for (unsigned i = GFX8; i <= GFX10; i++) {
+      if (!set_variant((chip_class)i))
+         continue;
+
+      QoShaderModuleCreateInfo vs = qoShaderModuleCreateInfoGLSL(VERTEX,
+         void main() {}
+      );
+
+      QoShaderModuleCreateInfo gs = qoShaderModuleCreateInfoGLSL(GEOMETRY,
+         layout(points) in;
+         layout(points, max_vertices = 1) out;
+
+         void main() {
+            EmitVertex();
+            EndPrimitive();
+         }
+      );
+
+      PipelineBuilder pbld(get_vk_device((chip_class)i));
+      pbld.add_stage(VK_SHADER_STAGE_VERTEX_BIT, vs);
+      pbld.add_stage(VK_SHADER_STAGE_GEOMETRY_BIT, gs);
+      pbld.create_pipeline();
+
+      //! success
+      fprintf(output, "success\n");
+   }
+END_TEST