aco: add missing SCC clobber in get_buffer_size
authorRhys Perry <pendingchaos02@gmail.com>
Thu, 15 Oct 2020 13:10:07 +0000 (14:10 +0100)
committerMarge Bot <eric+marge@anholt.net>
Thu, 15 Oct 2020 21:11:45 +0000 (21:11 +0000)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: fcd6d832456 ("aco: fix imageSize()/textureSize() with large buffers on GFX8")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7162>

src/amd/compiler/aco_instruction_selection.cpp

index d5ba2bd..1da9ea4 100644 (file)
@@ -5975,7 +5975,7 @@ void get_buffer_size(isel_context *ctx, Temp desc, Temp dst, bool in_elements)
       Temp size = emit_extract_vector(ctx, desc, 2, s1);
 
       Temp size_div3 = bld.vop3(aco_opcode::v_mul_hi_u32, bld.def(v1), bld.copy(bld.def(v1), Operand(0xaaaaaaabu)), size);
-      size_div3 = bld.sop2(aco_opcode::s_lshr_b32, bld.def(s1), bld.as_uniform(size_div3), Operand(1u));
+      size_div3 = bld.sop2(aco_opcode::s_lshr_b32, bld.def(s1), bld.def(s1, scc), bld.as_uniform(size_div3), Operand(1u));
 
       Temp stride = emit_extract_vector(ctx, desc, 1, s1);
       stride = bld.sop2(aco_opcode::s_bfe_u32, bld.def(s1), bld.def(s1, scc), stride, Operand((5u << 16) | 16u));