From bcf94bb933e8ccc0b91305ed8189a35e8938abbf Mon Sep 17 00:00:00 2001 From: Rhys Perry Date: Wed, 15 Jul 2020 17:34:21 +0100 Subject: [PATCH] aco: properly recognize that s_waitcnt mitigates VMEMtoScalarWriteHazard MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit fossil-db (Navi): Totals from 555 (0.41% of 135946) affected shaders: CodeSize: 1005716 -> 1003400 (-0.23%) Instrs: 195326 -> 194744 (-0.30%) Signed-off-by: Rhys Perry Reviewed-by: Daniel Schürmann Part-of: --- src/amd/compiler/aco_insert_NOPs.cpp | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/src/amd/compiler/aco_insert_NOPs.cpp b/src/amd/compiler/aco_insert_NOPs.cpp index bb703d7481e..a877172b67f 100644 --- a/src/amd/compiler/aco_insert_NOPs.cpp +++ b/src/amd/compiler/aco_insert_NOPs.cpp @@ -602,6 +602,14 @@ void handle_instruction_gfx10(Program *program, Block *cur_block, NOP_ctx_gfx10 if (program->wave_size == 64) ctx.sgprs_read_by_VMEM.set(exec_hi); } else if (instr->isSALU() || instr->format == Format::SMEM) { + if (instr->opcode == aco_opcode::s_waitcnt) { + /* Hazard is mitigated by "s_waitcnt vmcnt(0)" */ + uint16_t imm = static_cast(instr.get())->imm; + unsigned vmcnt = (imm & 0xF) | ((imm & (0x3 << 14)) >> 10); + if (vmcnt == 0) + ctx.sgprs_read_by_VMEM.reset(); + } + /* Check if SALU writes an SGPR that was previously read by the VALU */ if (check_written_regs(instr, ctx.sgprs_read_by_VMEM)) { ctx.sgprs_read_by_VMEM.reset(); @@ -610,12 +618,6 @@ void handle_instruction_gfx10(Program *program, Block *cur_block, NOP_ctx_gfx10 aco_ptr nop{create_instruction(aco_opcode::v_nop, Format::VOP1, 0, 0)}; new_instructions.emplace_back(std::move(nop)); } - } else if (instr->opcode == aco_opcode::s_waitcnt) { - /* Hazard is mitigated by "s_waitcnt vmcnt(0)" */ - uint16_t imm = static_cast(instr.get())->imm; - unsigned vmcnt = (imm & 0xF) | ((imm & (0x3 << 14)) >> 10); - if (vmcnt == 0) - ctx.sgprs_read_by_VMEM.reset(); } else if (instr->isVALU()) { /* Hazard is mitigated by any VALU instruction */ ctx.sgprs_read_by_VMEM.reset(); -- 2.30.2