From: Dave Airlie <airl...@redhat.com>
If we have an indirect index here we need to scale it by attribute slots
e.g. is this is vec2[256] then we get an indir_index in the 0.255 range
but the vec2 are aligned inside vec4 slots. So scale the indir index,
then extract the channels.
This appears to fix DOOM for me.
Signed-off-by: Dave Airlie <airl...@redhat.com>
---
src/amd/common/ac_nir_to_llvm.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 9be6e77..566516f 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -2237,6 +2237,9 @@ static LLVMValueRef visit_load_var(struct
nir_to_llvm_context *ctx,
LLVMValueRef ptr = get_shared_memory_ptr(ctx, idx, ctx->i32);
LLVMValueRef derived_ptr;
+ if (indir_index)
+ indir_index = LLVMBuildMul(ctx->builder, indir_index,
LLVMConstInt(ctx->i32, 4, false), "");
+
for (unsigned chan = 0; chan < ve; chan++) {
LLVMValueRef index = LLVMConstInt(ctx->i32, chan,
false);
if (indir_index)
@@ -2343,6 +2346,10 @@ visit_store_var(struct nir_to_llvm_context *ctx,
break;
case nir_var_shared: {
LLVMValueRef ptr = get_shared_memory_ptr(ctx, idx, ctx->i32);
+
+ if (indir_index)
+ indir_index = LLVMBuildMul(ctx->builder, indir_index,
LLVMConstInt(ctx->i32, 4, false), "");
+
for (unsigned chan = 0; chan < 8; chan++) {
if (!(writemask & (1 << chan)))
continue;
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
