From: Nicolai Hähnle <nicolai.haeh...@amd.com>
The spec actually says that calling EmitStreamVertex is undefined when
you exceed max_vertices. But we do need to avoid trampling over memory
outside the GSVS ring.
Cc: mesa-sta...@lists.freedesktop.org
--
One more thing I noticed on top of all the other GS changes...
---
src/gallium/drivers/radeonsi/si_shader.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index dee2a17..749823d 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5181,21 +5181,21 @@ static void si_llvm_emit_vertex(
"");
/* If this thread has already emitted the declared maximum number of
* vertices, skip the write: excessive vertex emissions are not
* supposed to have any effect.
*
* If the shader has no writes to memory, kill it instead. This skips
* further memory loads and may allow LLVM to skip to the end
* altogether.
*/
- can_emit = LLVMBuildICmp(gallivm->builder, LLVMIntULE, gs_next_vertex,
+ can_emit = LLVMBuildICmp(gallivm->builder, LLVMIntULT, gs_next_vertex,
lp_build_const_int32(gallivm,
shader->selector->gs_max_out_vertices), "");
bool use_kill = !info->writes_memory;
if (use_kill) {
kill = lp_build_select(&bld_base->base, can_emit,
lp_build_const_float(gallivm, 1.0f),
lp_build_const_float(gallivm, -1.0f));
lp_build_intrinsic(gallivm->builder, "llvm.AMDGPU.kill",
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
