aco: allow to load/store 16-bit values in VMEM for tess and geom

[mesa.git] / src / amd / compiler / aco_instruction_selection.cpp

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index c3b12fedeb904fbe3ef345dc7cf07de3561564b2..5346376a48de9c423b6ac08d152a5e71644f624f 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -3940,7 +3940,7 @@ void store_vmem_mubuf(isel_context *ctx, Temp src, Temp descriptor, Temp voffset
                                    bool allow_combining = true, bool reorder = true, bool slc = false)
 {
    Builder bld(ctx->program, ctx->block);
-   assert(elem_size_bytes == 4 || elem_size_bytes == 8);
+   assert(elem_size_bytes == 2 || elem_size_bytes == 4 || elem_size_bytes == 8);
    assert(write_mask);
    write_mask = widen_mask(write_mask, elem_size_bytes);
 
@@ -3960,8 +3960,8 @@ void load_vmem_mubuf(isel_context *ctx, Temp dst, Temp descriptor, Temp voffset,
                      unsigned base_const_offset, unsigned elem_size_bytes, unsigned num_components,
                      unsigned stride = 0u, bool allow_combining = true, bool allow_reorder = true)
 {
-   assert(elem_size_bytes == 4 || elem_size_bytes == 8);
-   assert((num_components * elem_size_bytes / 4) == dst.size());
+   assert(elem_size_bytes == 2 || elem_size_bytes == 4 || elem_size_bytes == 8);
+   assert((num_components * elem_size_bytes) == dst.bytes());
    assert(!!stride != allow_combining);
 
    Builder bld(ctx->program, ctx->block);