From d6af7571c22816a0724737711941589e4cbe248a Mon Sep 17 00:00:00 2001 From: Samuel Pitoiset Date: Fri, 17 Jan 2020 13:22:16 +0100 Subject: [PATCH] aco: implement nir_intrinsic_store_global on GFX6 MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit GFX6 doesn't have FLAT instructions, use MUBUF instructions instead. Signed-off-by: Samuel Pitoiset Reviewed-by: Daniel Schürmann Part-of: --- .../compiler/aco_instruction_selection.cpp | 141 ++++++++++++------ 1 file changed, 98 insertions(+), 43 deletions(-) diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 47e1737b5d9..9d29fae316c 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -4716,6 +4716,16 @@ void visit_get_buffer_size(isel_context *ctx, nir_intrinsic_instr *instr) { get_buffer_size(ctx, desc, get_ssa_temp(ctx, &instr->dest.ssa), false); } +Temp get_gfx6_global_rsrc(Builder& bld, Temp addr) +{ + uint32_t rsrc_conf = S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT) | + S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32); + + if (addr.type() == RegType::vgpr) + return bld.pseudo(aco_opcode::p_create_vector, bld.def(s4), Operand(0u), Operand(0u), Operand(-1u), Operand(rsrc_conf)); + return bld.pseudo(aco_opcode::p_create_vector, bld.def(s4), addr, Operand(-1u), Operand(rsrc_conf)); +} + void visit_load_global(isel_context *ctx, nir_intrinsic_instr *instr) { Builder bld(ctx->program, ctx->block); @@ -4811,12 +4821,20 @@ void visit_store_global(isel_context *ctx, nir_intrinsic_instr *instr) unsigned elem_size_bytes = instr->src[0].ssa->bit_size / 8; Temp data = as_vgpr(ctx, get_ssa_temp(ctx, instr->src[0].ssa)); - Temp addr = as_vgpr(ctx, get_ssa_temp(ctx, instr->src[1].ssa)); + Temp addr = get_ssa_temp(ctx, instr->src[1].ssa); + + if (ctx->options->chip_class >= GFX7) + addr = as_vgpr(ctx, addr); unsigned writemask = nir_intrinsic_write_mask(instr); while (writemask) { int start, count; u_bit_scan_consecutive_range(&writemask, &start, &count); + if (count == 3 && ctx->options->chip_class == GFX6) { + /* GFX6 doesn't support storing vec3, split it. */ + writemask |= 1u << (start + 2); + count = 2; + } unsigned num_bytes = count * elem_size_bytes; Temp write_data = data; @@ -4829,54 +4847,91 @@ void visit_store_global(isel_context *ctx, nir_intrinsic_instr *instr) ctx->block->instructions.emplace_back(std::move(vec)); } + bool glc = nir_intrinsic_access(instr) & (ACCESS_VOLATILE | ACCESS_COHERENT | ACCESS_NON_READABLE); unsigned offset = start * elem_size_bytes; - if (offset > 0 && ctx->options->chip_class < GFX9) { - Temp addr0 = bld.tmp(v1), addr1 = bld.tmp(v1); - Temp new_addr0 = bld.tmp(v1), new_addr1 = bld.tmp(v1); - Temp carry = bld.tmp(bld.lm); - bld.pseudo(aco_opcode::p_split_vector, Definition(addr0), Definition(addr1), addr); - bld.vop2(aco_opcode::v_add_co_u32, Definition(new_addr0), bld.hint_vcc(Definition(carry)), - Operand(offset), addr0); - bld.vop2(aco_opcode::v_addc_co_u32, Definition(new_addr1), bld.def(bld.lm), - Operand(0u), addr1, - carry).def(1).setHint(vcc); + if (ctx->options->chip_class >= GFX7) { + if (offset > 0 && ctx->options->chip_class < GFX9) { + Temp addr0 = bld.tmp(v1), addr1 = bld.tmp(v1); + Temp new_addr0 = bld.tmp(v1), new_addr1 = bld.tmp(v1); + Temp carry = bld.tmp(bld.lm); + bld.pseudo(aco_opcode::p_split_vector, Definition(addr0), Definition(addr1), addr); - addr = bld.pseudo(aco_opcode::p_create_vector, bld.def(v2), new_addr0, new_addr1); + bld.vop2(aco_opcode::v_add_co_u32, Definition(new_addr0), bld.hint_vcc(Definition(carry)), + Operand(offset), addr0); + bld.vop2(aco_opcode::v_addc_co_u32, Definition(new_addr1), bld.def(bld.lm), + Operand(0u), addr1, + carry).def(1).setHint(vcc); - offset = 0; - } + addr = bld.pseudo(aco_opcode::p_create_vector, bld.def(v2), new_addr0, new_addr1); - bool glc = nir_intrinsic_access(instr) & (ACCESS_VOLATILE | ACCESS_COHERENT | ACCESS_NON_READABLE); - bool global = ctx->options->chip_class >= GFX9; - aco_opcode op; - switch (num_bytes) { - case 4: - op = global ? aco_opcode::global_store_dword : aco_opcode::flat_store_dword; - break; - case 8: - op = global ? aco_opcode::global_store_dwordx2 : aco_opcode::flat_store_dwordx2; - break; - case 12: - op = global ? aco_opcode::global_store_dwordx3 : aco_opcode::flat_store_dwordx3; - break; - case 16: - op = global ? aco_opcode::global_store_dwordx4 : aco_opcode::flat_store_dwordx4; - break; - default: - unreachable("store_global not implemented for this size."); + offset = 0; + } + + bool global = ctx->options->chip_class >= GFX9; + aco_opcode op; + switch (num_bytes) { + case 4: + op = global ? aco_opcode::global_store_dword : aco_opcode::flat_store_dword; + break; + case 8: + op = global ? aco_opcode::global_store_dwordx2 : aco_opcode::flat_store_dwordx2; + break; + case 12: + op = global ? aco_opcode::global_store_dwordx3 : aco_opcode::flat_store_dwordx3; + break; + case 16: + op = global ? aco_opcode::global_store_dwordx4 : aco_opcode::flat_store_dwordx4; + break; + default: + unreachable("store_global not implemented for this size."); + } + + aco_ptr flat{create_instruction(op, global ? Format::GLOBAL : Format::FLAT, 3, 0)}; + flat->operands[0] = Operand(addr); + flat->operands[1] = Operand(s1); + flat->operands[2] = Operand(data); + flat->glc = glc; + flat->dlc = false; + flat->offset = offset; + flat->disable_wqm = true; + flat->barrier = barrier_buffer; + ctx->program->needs_exact = true; + ctx->block->instructions.emplace_back(std::move(flat)); + } else { + assert(ctx->options->chip_class == GFX6); + + aco_opcode op; + switch (num_bytes) { + case 4: + op = aco_opcode::buffer_store_dword; + break; + case 8: + op = aco_opcode::buffer_store_dwordx2; + break; + case 16: + op = aco_opcode::buffer_store_dwordx4; + break; + default: + unreachable("store_global not implemented for this size."); + } + + Temp rsrc = get_gfx6_global_rsrc(bld, addr); + + aco_ptr mubuf{create_instruction(op, Format::MUBUF, 4, 0)}; + mubuf->operands[0] = addr.type() == RegType::vgpr ? Operand(addr) : Operand(v1); + mubuf->operands[1] = Operand(rsrc); + mubuf->operands[2] = Operand(0u); + mubuf->operands[3] = Operand(write_data); + mubuf->glc = glc; + mubuf->dlc = false; + mubuf->offset = offset; + mubuf->addr64 = addr.type() == RegType::vgpr; + mubuf->disable_wqm = true; + mubuf->barrier = barrier_buffer; + ctx->program->needs_exact = true; + ctx->block->instructions.emplace_back(std::move(mubuf)); } - aco_ptr flat{create_instruction(op, global ? Format::GLOBAL : Format::FLAT, 3, 0)}; - flat->operands[0] = Operand(addr); - flat->operands[1] = Operand(s1); - flat->operands[2] = Operand(data); - flat->glc = glc; - flat->dlc = false; - flat->offset = offset; - flat->disable_wqm = true; - flat->barrier = barrier_buffer; - ctx->program->needs_exact = true; - ctx->block->instructions.emplace_back(std::move(flat)); } } -- 2.30.2