radv: Use different intrinsic for ubo loads.

Not sure about the deprecation path, but this intrinsic
can be lowered to SMEM loads. This results in a significant
Talos performance improvement.

v2: Fix for LLVM attribute changes.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
This commit is contained in:
Bas Nieuwenhuizen 2016-11-29 00:18:43 +01:00
parent 0303201dfb
commit 05533ce418
1 changed files with 29 additions and 1 deletions

View File

@ -2035,6 +2035,34 @@ static LLVMValueRef visit_load_buffer(struct nir_to_llvm_context *ctx,
get_def_type(ctx, &instr->dest.ssa), "");
}
static LLVMValueRef visit_load_ubo_buffer(struct nir_to_llvm_context *ctx,
nir_intrinsic_instr *instr)
{
const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
const char *load_name;
LLVMTypeRef data_type = ctx->f32;
LLVMValueRef results[4], ret;
LLVMValueRef rsrc = get_src(ctx, instr->src[0]);
LLVMValueRef offset = get_src(ctx, instr->src[1]);
rsrc = LLVMBuildBitCast(ctx->builder, rsrc, LLVMVectorType(ctx->i8, 16), "");
for (unsigned i = 0; i < instr->num_components; ++i) {
LLVMValueRef params[] = {
rsrc,
LLVMBuildAdd(ctx->builder, LLVMConstInt(ctx->i32, 4 * i, 0),
offset, "")
};
results[i] = emit_llvm_intrinsic(ctx, "llvm.SI.load.const", ctx->f32,
params, 2, AC_FUNC_ATTR_READNONE);
}
ret = build_gather_values(ctx, results, instr->num_components);
return LLVMBuildBitCast(ctx->builder, ret,
get_def_type(ctx, &instr->dest.ssa), "");
}
static void
radv_get_deref_offset(struct nir_to_llvm_context *ctx, nir_deref *tail,
bool vs_in, unsigned *const_out, LLVMValueRef *indir_out)
@ -2956,7 +2984,7 @@ static void visit_intrinsic(struct nir_to_llvm_context *ctx,
result = visit_atomic_ssbo(ctx, instr);
break;
case nir_intrinsic_load_ubo:
result = visit_load_buffer(ctx, instr);
result = visit_load_ubo_buffer(ctx, instr);
break;
case nir_intrinsic_get_buffer_size:
result = visit_get_buffer_size(ctx, instr);