aco: use v_bfe_u32 for unsigned reductions sign-extension on GFX6-GFX7

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5327>
This commit is contained in:
Samuel Pitoiset
2020-06-05 08:54:52 +02:00
parent a874132cc4
commit ee4bc13de2
+8 -1
View File
@@ -591,7 +591,14 @@ void emit_reduction(lower_context *ctx, aco_opcode op, ReduceOp reduce_op, unsig
sdwa->dst_sel = sdwa_udword;
bld.insert(std::move(sdwa));
} else if (ctx->program->chip_class == GFX6 || ctx->program->chip_class == GFX7) {
bld.vop3(aco_opcode::v_bfe_i32, Definition(PhysReg{tmp}, v1),
aco_opcode opcode;
if (reduce_op == imin16 || reduce_op == imax16 || reduce_op == iadd16)
opcode = aco_opcode::v_bfe_i32;
else
opcode = aco_opcode::v_bfe_u32;
bld.vop3(opcode, Definition(PhysReg{tmp}, v1),
Operand(PhysReg{tmp}, v1), Operand(0u), Operand(16u));
}
}