From 77f08982af8f0807ab489d36a5c137661e10624d Mon Sep 17 00:00:00 2001 From: Samuel Pitoiset Date: Tue, 26 May 2020 16:21:44 +0200 Subject: [PATCH] aco: sign-extend input/identity for 16-bit subgroup ops on GFX6-GFX7 16-bit subgroup ops are implemented with 32-bit instructions on GFX6-GFX7. Signed-off-by: Samuel Pitoiset Reviewed-by: Rhys Perry Part-of: --- src/amd/compiler/aco_lower_to_hw_instr.cpp | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/amd/compiler/aco_lower_to_hw_instr.cpp b/src/amd/compiler/aco_lower_to_hw_instr.cpp index df7b571..40d4669 100644 --- a/src/amd/compiler/aco_lower_to_hw_instr.cpp +++ b/src/amd/compiler/aco_lower_to_hw_instr.cpp @@ -590,6 +590,9 @@ void emit_reduction(lower_context *ctx, aco_opcode op, ReduceOp reduce_op, unsig sdwa->sel[0] = sdwa_uword; 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), + Operand(PhysReg{tmp}, v1), Operand(0u), Operand(16u)); } } -- 2.7.4