160 lines · plain
1//=== ---- PPCMacroFusion.def - PowerPC MacroFuson Candidates -v-*- C++ -*-===//2//3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.4// See https)//llvm.org/LICENSE.txt for license information.5// SPDX-License-Identifier) Apache-2.0 WITH LLVM-exception6//7//===----------------------------------------------------------------------===//8//9// This file contains descriptions of the macro-fusion pair for PowerPC.10//11//===----------------------------------------------------------------------===//12 13// NOTE: NO INCLUDE GUARD DESIRED!14 15#ifndef FUSION_FEATURE16 17// Each FUSION_FEATURE is assigned with one TYPE, and can be enabled/disabled18// by HAS_FEATURE. The instructions pair is fusable only when the opcode19// of the first instruction is in OPSET1, and the second instruction opcode is20// in OPSET2. And if DEP_OP_IDX >=0, we will check the result of first OP is 21// the operand of the second op with DEP_OP_IDX as its operand index. We assume22// that the result of the first op is its operand zero. 23#define FUSION_FEATURE(TYPE, HAS_FEATURE, DEP_OP_IDX, OPSET1, OPSET2)24 25#endif26 27#ifndef FUSION_OP_SET28#define FUSION_OP_SET(...) __VA_ARGS__ 29#endif30 31// Power8 User Manual Section 10.1.12, Instruction Fusion 32// {addi} followed by one of these {lxvd2x, lxvw4x, lxvdsx, lvebx, lvehx,33// lvewx, lvx, lxsdx}34FUSION_FEATURE(AddiLoad, hasAddiLoadFusion, 2, \35 FUSION_OP_SET(ADDI, ADDI8, ADDItocL, ADDItocL8), \36 FUSION_OP_SET(LXVD2X, LXVW4X, LXVDSX, LVEBX, LVEHX, LVEWX, \37 LVX, LXSDX))38 39// {addis) followed by one of these {ld, lbz, lhz, lwz}40FUSION_FEATURE(AddisLoad, hasAddisLoadFusion, 2, \41 FUSION_OP_SET(ADDIS, ADDIS8, ADDIStocHA8), \42 FUSION_OP_SET(LD, LBZ, LBZ8, LHZ, LHZ8, LWZ, LWZ8))43 44// Power10 User Manual Section 19.1.5.4, Fusion45// {add, mulld} - add46FUSION_FEATURE(ArithAdd, hasArithAddFusion, -1,47 FUSION_OP_SET(ADD4, ADD8, MULLD), FUSION_OP_SET(ADD4, ADD8))48 49// {add, subf} - {and, nand, nor, or}50FUSION_FEATURE(ArithLogical, hasAddLogicalFusion, -1,51 FUSION_OP_SET(ADD4, ADD8, SUBF, SUBF8),52 FUSION_OP_SET(AND, AND8, OR, OR8, NAND, NAND8, NOR, NOR8))53 54// {and, andc, eqv, nand, nor, or, orc, xor} - {add, subf}55FUSION_FEATURE(LogicalArith, hasLogicalAddFusion, -1,56 FUSION_OP_SET(AND, ANDC, EQV, NAND, NOR, OR, ORC, XOR, AND8,57 ANDC8, EQV8, NAND8, NOR8, OR8, ORC8, XOR8),58 FUSION_OP_SET(ADD4, ADD8, SUBF, SUBF8))59 60// Either of {and, andc, eqv, nand, nor, or, orc, xor}61FUSION_FEATURE(Logical, hasLogicalFusion, -1,62 FUSION_OP_SET(AND, ANDC, EQV, NAND, NOR, OR, ORC, XOR, AND8,63 ANDC8, EQV8, NAND8, NOR8, OR8, ORC8, XOR8),64 FUSION_OP_SET(AND, ANDC, EQV, NAND, NOR, OR, ORC, XOR, AND8,65 ANDC8, EQV8, NAND8, NOR8, OR8, ORC8, XOR8))66 67// vaddudm - vaddudm68FUSION_FEATURE(VecAdd, hasArithAddFusion, -1, FUSION_OP_SET(VADDUDM),69 FUSION_OP_SET(VADDUDM))70 71// Either of {vand, vandc, veqv, vnand, vnor, vor, vorc, vxor}72FUSION_FEATURE(VecLogical, hasLogicalFusion, -1,73 FUSION_OP_SET(VAND, VANDC, VEQV, VNAND, VNOR, VOR, VORC, VXOR),74 FUSION_OP_SET(VAND, VANDC, VEQV, VNAND, VNOR, VOR, VORC, VXOR))75 76// sldi rx, ra, {3, 6} - {add, subf}77// sldi rx, ra n is alias of rldicr rx, ra, n, 63-n78FUSION_FEATURE(SldiAdd, hasArithAddFusion, -1, FUSION_OP_SET(RLDICR, RLDICR_32),79 FUSION_OP_SET(ADD4, ADD8, SUBF, SUBF8))80 81// rldicl rx, ra, 1, 0 - xor82FUSION_FEATURE(RotateLeftXor, hasSha3Fusion, 1,83 FUSION_OP_SET(RLDICL, RLDICL_32, RLDICL_32_64),84 FUSION_OP_SET(XOR, XOR8))85 86// rldicr rx, ra, 1, 63 - xor87FUSION_FEATURE(RotateRightXor, hasSha3Fusion, 1,88 FUSION_OP_SET(RLDICR, RLDICR_32), FUSION_OP_SET(XOR, XOR8))89 90// There're two special cases in 'load-compare' series, so we have to split91// them into several pattern groups to fit into current framework. This can92// be clearer once we switched to a more expressive approach.93 94// { lbz,lbzx,lhz,lhzx,lwz,lwzx } - cmpi 0,1,rx,{ 0,1,-1 }95// { lbz,lbzx,lhz,lhzx,lwz,lwzx } - cmpli 0,L,rx,{ 0,1 }96FUSION_FEATURE(LoadCmp1, hasCompareFusion, 1,97 FUSION_OP_SET(LBZ, LBZ8, LBZX, LBZX8, LBZXTLS, LBZXTLS_,98 LBZXTLS_32, LHZ, LHZ8, LHZX, LHZX8, LHZXTLS,99 LHZXTLS_, LHZXTLS_32, LWZ, LWZ8, LWZX, LWZX8,100 LWZXTLS, LWZXTLS_, LWZXTLS_32),101 FUSION_OP_SET(CMPDI, CMPLDI, CMPLWI))102 103// { ld,ldx } - cmpi 0,1,rx,{ 0,1,-1 }104// { ld,ldx } - cmpli 0,1,rx,{ 0,1 }105FUSION_FEATURE(LoadCmp2, hasCompareFusion, 1,106 FUSION_OP_SET(LD, LDX, LDXTLS, LDXTLS_),107 FUSION_OP_SET(CMPDI, CMPLDI))108 109// { lha,lhax,lwa,lwax } - cmpi 0,L,rx,{ 0,1,-1 }110FUSION_FEATURE(LoadCmp3, hasCompareFusion, 1,111 FUSION_OP_SET(LHA, LHA8, LHAX, LHAX8, LWA, LWA_32, LWAX,112 LWAX_32),113 FUSION_OP_SET(CMPLDI, CMPLWI))114 115// ori - oris116FUSION_FEATURE(OriOris, hasWideImmFusion, 1, FUSION_OP_SET(ORI, ORI8),117 FUSION_OP_SET(ORIS, ORIS8))118 119// lis - ori120FUSION_FEATURE(LisOri, hasWideImmFusion, 1, FUSION_OP_SET(LIS, LIS8),121 FUSION_OP_SET(ORI, ORI8))122 123// oris - ori124FUSION_FEATURE(OrisOri, hasWideImmFusion, 1, FUSION_OP_SET(ORIS, ORIS8),125 FUSION_OP_SET(ORI, ORI8))126 127// xori - xoris128FUSION_FEATURE(XoriXoris, hasWideImmFusion, 1, FUSION_OP_SET(XORI, XORI8),129 FUSION_OP_SET(XORIS, XORIS8))130 131// xoris - xori132FUSION_FEATURE(XorisXori, hasWideImmFusion, 1, FUSION_OP_SET(XORIS, XORIS8),133 FUSION_OP_SET(XORI, XORI8))134 135// addis rx,ra,si - addi rt,rx,SI, SI >= 0136FUSION_FEATURE(AddisAddi, hasWideImmFusion, 1,137 FUSION_OP_SET(ADDIS, ADDIS8, ADDIStocHA8, ADDIStocHA),138 FUSION_OP_SET(ADDI, ADDI8, ADDItocL8, ADDItocL))139 140// addi rx,ra,si - addis rt,rx,SI, ra > 0, SI >= 2141FUSION_FEATURE(AddiAddis, hasWideImmFusion, 1,142 FUSION_OP_SET(ADDI, ADDI8, ADDItocL8, ADDItocL),143 FUSION_OP_SET(ADDIS, ADDIS8, ADDIStocHA8, ADDIStocHA))144 145// mtctr - { bcctr,bcctrl }146FUSION_FEATURE(ZeroMoveCTR, hasZeroMoveFusion, -1,147 FUSION_OP_SET(MTCTR, MTCTRloop, MTSPR8, MTSPR),148 FUSION_OP_SET(BCCTR, BCCTRn, BCCTR8, BCCTR8n, BCCTRL, BCCTRLn,149 BCCTRL8, BCCTRL8n, gBCCTR, gBCCTRL))150 151// mtlr - { bclr,bclrl }152FUSION_FEATURE(ZeroMoveLR, hasZeroMoveFusion, -1,153 FUSION_OP_SET(MTLR8, MTLR, MTSPR8, MTSPR),154 FUSION_OP_SET(BCLR, BCLRn, gBCLR, BCLRL, BCLRLn, gBCLRL))155 156#include "PPCBack2BackFusion.def"157 158#undef FUSION_FEATURE159#undef FUSION_OP_SET160