From 8d907d30616c1267944145a51dd354118fed27aa Mon Sep 17 00:00:00 2001 From: Sandipan Das Date: Thu, 7 Jun 2018 01:18:55 +0530 Subject: [PATCH] arch-power: Reorder instruction decoding logic This reorders the decoding logic based on the category of instructions. The ordering applied here is roughly in line with the Power ISA manual which is as follows: * Branch facility instructions * Branch instructions * Condition Register instructions * System Call instructions * Fixed-point facility instructions * Load instructions * Store instructions * Arithmetic instructions * Compare instructions * Logical instructions * Rotate and Shift instructions * Move To/From System Register instructions * Floating-point facility instructions * Load instructions * Store instructions * Arithmetic instructions * Move instructions * Rounding and Conversion instructions * Compare instructions * Status and Control Register instructions Change-Id: Icfb57c5e442a959e502222222b84289d8e74ecbf Signed-off-by: Sandipan Das --- src/arch/power/isa/decoder.isa | 478 ++++++++++++++++++--------------- 1 file changed, 259 insertions(+), 219 deletions(-) diff --git a/src/arch/power/isa/decoder.isa b/src/arch/power/isa/decoder.isa index b7b9afffc..cd08bfdce 100644 --- a/src/arch/power/isa/decoder.isa +++ b/src/arch/power/isa/decoder.isa @@ -36,6 +36,164 @@ // decode OPCODE default Unknown::unknown() { + 18: decode AA { + + // Unconditionally branch relative to PC. + format BranchPCRel { + 0: b({{ NIA = (uint32_t)(CIA + disp); }}); + } + + // Unconditionally branch to fixed address. + format BranchNonPCRel { + 1: ba({{ NIA = targetAddr; }}); + } + } + + 16: decode AA { + + // Conditionally branch relative to PC based on CR and CTR. + format BranchPCRelCondCtr { + 0: bc({{ NIA = (uint32_t)(CIA + disp); }}); + } + + // Conditionally branch to fixed address based on CR and CTR. + format BranchNonPCRelCondCtr { + 1: bca({{ NIA = targetAddr; }}); + } + } + + 19: decode XO_XO { + + // Conditionally branch to address in LR based on CR and CTR. + format BranchLrCondCtr { + 16: bclr({{ NIA = LR & 0xfffffffc; }}); + } + + // Conditionally branch to address in CTR based on CR. + format BranchCtrCond { + 528: bcctr({{ NIA = CTR & 0xfffffffc; }}); + } + + // Condition register manipulation instructions. + format CondLogicOp { + 257: crand({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa & crBb); + }}); + + 449: cror({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa | crBb); + }}); + + 255: crnand({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, !(crBa & crBb)); + }}); + + 193: crxor({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa ^ crBb); + }}); + + 33: crnor({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, !(crBa | crBb)); + }}); + + 289: creqv({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa == crBb); + }}); + + 129: crandc({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa & !crBb); + }}); + + 417: crorc({{ + uint32_t crBa = bits(CR, 31 - ba); + uint32_t crBb = bits(CR, 31 - bb); + CR = insertBits(CR, 31 - bt, crBa | !crBb); + }}); + } + + format CondMoveOp { + 0: mcrf({{ + uint32_t crBfa = bits(CR, 31 - bfa*4, 28 - bfa*4); + CR = insertBits(CR, 31 - bf*4, 28 - bf*4, crBfa); + }}); + } + + format MiscOp { + 150: isync({{ }}, [ IsSerializeAfter ]); + } + } + + 17: IntOp::sc({{ xc->syscall(R0, &fault); }}, + [ IsSyscall, IsNonSpeculative, IsSerializeAfter ]); + + format LoadDispOp { + 34: lbz({{ Rt = Mem_ub; }}); + 40: lhz({{ Rt = Mem_uh; }}); + 42: lha({{ Rt = Mem_sh; }}); + 32: lwz({{ Rt = Mem; }}); + 58: lwa({{ Rt = Mem_sw; }}, + {{ EA = Ra + (disp & 0xfffffffc); }}, + {{ EA = disp & 0xfffffffc; }}); + } + + format LoadDispUpdateOp { + 35: lbzu({{ Rt = Mem_ub; }}); + 41: lhzu({{ Rt = Mem_uh; }}); + 43: lhau({{ Rt = Mem_sh; }}); + 33: lwzu({{ Rt = Mem; }}); + } + + format StoreDispOp { + 38: stb({{ Mem_ub = Rs_ub; }}); + 44: sth({{ Mem_uh = Rs_uh; }}); + 36: stw({{ Mem = Rs; }}); + } + + format StoreDispUpdateOp { + 39: stbu({{ Mem_ub = Rs_ub; }}); + 45: sthu({{ Mem_uh = Rs_uh; }}); + 37: stwu({{ Mem = Rs; }}); + } + + format IntImmArithCheckRaOp { + 14: addi({{ Rt = Ra + imm; }}, + {{ Rt = imm }}); + + 15: addis({{ Rt = Ra + (imm << 16); }}, + {{ Rt = imm << 16; }}); + } + + format IntImmArithOp { + 12: addic({{ uint32_t src = Ra; Rt = src + imm; }}, + [computeCA]); + + 13: addic_({{ uint32_t src = Ra; Rt = src + imm; }}, + [computeCA, computeCR0]); + + 8: subfic({{ int32_t src = ~Ra; Rt = src + imm + 1; }}, + [computeCA]); + + 7: mulli({{ + int32_t src = Ra_sw; + int64_t prod = src * imm; + Rt = (uint32_t)prod; + }}); + } + format IntImmOp { 10: cmpli({{ Xer xer = XER; @@ -49,6 +207,24 @@ decode OPCODE default Unknown::unknown() { }}); } + format IntImmLogicOp { + 24: ori({{ Ra = Rs | uimm; }}); + 25: oris({{ Ra = Rs | (uimm << 16); }}); + 26: xori({{ Ra = Rs ^ uimm; }}); + 27: xoris({{ Ra = Rs ^ (uimm << 16); }}); + 28: andi_({{ Ra = Rs & uimm; }}, + true); + 29: andis_({{ Ra = Rs & (uimm << 16); }}, + true); + } + + format IntRotateOp { + 21: rlwinm({{ Ra = rotateValue(Rs, sh) & fullMask; }}); + 23: rlwnm({{ Ra = rotateValue(Rs, Rb) & fullMask; }}); + 20: rlwimi({{ Ra = (rotateValue(Rs, sh) & fullMask) | + (Ra & ~fullMask); }}); + } + // Some instructions use bits 21 - 30, others 22 - 30. We have to use // the larger size to account for all opcodes. For those that use the // smaller value, the OE bit is bit 21. Therefore, we have two versions @@ -56,6 +232,60 @@ decode OPCODE default Unknown::unknown() { // example see 'add' and 'addo'. 31: decode XO_XO { + // All loads with an index register. The non-update versions + // all use the value 0 if Ra == R0, not the value contained in + // R0. Others update Ra with the effective address. In all cases, + // Ra and Rb are source registers, Rt is the destintation. + format LoadIndexOp { + 87: lbzx({{ Rt = Mem_ub; }}); + 279: lhzx({{ Rt = Mem_uh; }}); + 343: lhax({{ Rt = Mem_sh; }}); + 23: lwzx({{ Rt = Mem; }}); + 341: lwax({{ Rt = Mem_sw; }}); + 20: lwarx({{ Rt = Mem_sw; Rsv = 1; RsvLen = 4; RsvAddr = EA; }}); + 535: lfsx({{ Ft_sf = Mem_sf; }}); + 599: lfdx({{ Ft = Mem_df; }}); + 855: lfiwax({{ Ft_uw = Mem; }}); + } + + format LoadIndexUpdateOp { + 119: lbzux({{ Rt = Mem_ub; }}); + 311: lhzux({{ Rt = Mem_uh; }}); + 375: lhaux({{ Rt = Mem_sh; }}); + 55: lwzux({{ Rt = Mem; }}); + 373: lwaux({{ Rt = Mem_sw; }}); + 567: lfsux({{ Ft_sf = Mem_sf; }}); + 631: lfdux({{ Ft = Mem_df; }}); + } + + format StoreIndexOp { + 215: stbx({{ Mem_ub = Rs_ub; }}); + 407: sthx({{ Mem_uh = Rs_uh; }}); + 151: stwx({{ Mem = Rs; }}); + 150: stwcx({{ + bool store_performed = false; + Mem = Rs; + if (Rsv) { + if (RsvLen == 4) { + if (RsvAddr == EA) { + store_performed = true; + } + } + } + Xer xer = XER; + Cr cr = CR; + cr.cr0 = ((store_performed ? 0x2 : 0x0) | xer.so); + CR = cr; + Rsv = 0; + }}); + } + + format StoreIndexUpdateOp { + 247: stbux({{ Mem_ub = Rs_ub; }}); + 439: sthux({{ Mem_uh = Rs_uh; }}); + 183: stwux({{ Mem = Rs; }}); + } + // These instructions can all be reduced to the form // Rt = src1 + src2 [+ CA], therefore we just give src1 and src2 // (and, if necessary, CA) definitions and let the python script @@ -144,6 +374,20 @@ decode OPCODE default Unknown::unknown() { true); } + format IntOp { + 0: cmp({{ + Xer xer = XER; + uint32_t cr = makeCRField(Ra_sw, Rb_sw, xer.so); + CR = insertCRField(CR, BF, cr); + }}); + + 32: cmpl({{ + Xer xer = XER; + uint32_t cr = makeCRField(Ra, Rb, xer.so); + CR = insertCRField(CR, BF, cr); + }}); + } + // Integer logic instructions use source registers Rs and Rb, // with destination register Ra. format IntLogicOp { @@ -248,16 +492,6 @@ decode OPCODE default Unknown::unknown() { // Generic integer format instructions. format IntOp { - 0: cmp({{ - Xer xer = XER; - uint32_t cr = makeCRField(Ra_sw, Rb_sw, xer.so); - CR = insertCRField(CR, BF, cr); - }}); - 32: cmpl({{ - Xer xer = XER; - uint32_t cr = makeCRField(Ra, Rb, xer.so); - CR = insertCRField(CR, BF, cr); - }}); 144: mtcrf({{ uint32_t mask = 0; for (int i = 0; i < 8; ++i) { @@ -280,61 +514,13 @@ decode OPCODE default Unknown::unknown() { } } - // All loads with an index register. The non-update versions - // all use the value 0 if Ra == R0, not the value contained in - // R0. Others update Ra with the effective address. In all cases, - // Ra and Rb are source registers, Rt is the destintation. - format LoadIndexOp { - 87: lbzx({{ Rt = Mem_ub; }}); - 279: lhzx({{ Rt = Mem_uh; }}); - 343: lhax({{ Rt = Mem_sh; }}); - 23: lwzx({{ Rt = Mem; }}); - 341: lwax({{ Rt = Mem_sw; }}); - 20: lwarx({{ Rt = Mem_sw; Rsv = 1; RsvLen = 4; RsvAddr = EA; }}); - 535: lfsx({{ Ft_sf = Mem_sf; }}); - 599: lfdx({{ Ft = Mem_df; }}); - 855: lfiwax({{ Ft_uw = Mem; }}); - } - - format LoadIndexUpdateOp { - 119: lbzux({{ Rt = Mem_ub; }}); - 311: lhzux({{ Rt = Mem_uh; }}); - 375: lhaux({{ Rt = Mem_sh; }}); - 55: lwzux({{ Rt = Mem; }}); - 373: lwaux({{ Rt = Mem_sw; }}); - 567: lfsux({{ Ft_sf = Mem_sf; }}); - 631: lfdux({{ Ft = Mem_df; }}); - } - format StoreIndexOp { - 215: stbx({{ Mem_ub = Rs_ub; }}); - 407: sthx({{ Mem_uh = Rs_uh; }}); - 151: stwx({{ Mem = Rs; }}); - 150: stwcx({{ - bool store_performed = false; - Mem = Rs; - if (Rsv) { - if (RsvLen == 4) { - if (RsvAddr == EA) { - store_performed = true; - } - } - } - Xer xer = XER; - Cr cr = CR; - cr.cr0 = ((store_performed ? 0x2 : 0x0) | xer.so); - CR = cr; - Rsv = 0; - }}); 663: stfsx({{ Mem_sf = Fs_sf; }}); 727: stfdx({{ Mem_df = Fs; }}); 983: stfiwx({{ Mem = Fs_uw; }}); } format StoreIndexUpdateOp { - 247: stbux({{ Mem_ub = Rs_ub; }}); - 439: sthux({{ Mem_uh = Rs_uh; }}); - 183: stwux({{ Mem = Rs; }}); 695: stfsux({{ Mem_sf = Fs_sf; }}); 759: stfdux({{ Mem_df = Fs; }}); } @@ -348,175 +534,26 @@ decode OPCODE default Unknown::unknown() { } } - format IntImmArithCheckRaOp { - 14: addi({{ Rt = Ra + imm; }}, - {{ Rt = imm }}); - 15: addis({{ Rt = Ra + (imm << 16); }}, - {{ Rt = imm << 16; }}); - } - - format IntImmArithOp { - 12: addic({{ uint32_t src = Ra; Rt = src + imm; }}, - [computeCA]); - 13: addic_({{ uint32_t src = Ra; Rt = src + imm; }}, - [computeCA, computeCR0]); - 8: subfic({{ int32_t src = ~Ra; Rt = src + imm + 1; }}, - [computeCA]); - 7: mulli({{ - int32_t src = Ra_sw; - int64_t prod = src * imm; - Rt = (uint32_t)prod; - }}); - } - - format IntImmLogicOp { - 24: ori({{ Ra = Rs | uimm; }}); - 25: oris({{ Ra = Rs | (uimm << 16); }}); - 26: xori({{ Ra = Rs ^ uimm; }}); - 27: xoris({{ Ra = Rs ^ (uimm << 16); }}); - 28: andi_({{ Ra = Rs & uimm; }}, - true); - 29: andis_({{ Ra = Rs & (uimm << 16); }}, - true); - } - - 16: decode AA { - - // Conditionally branch relative to PC based on CR and CTR. - format BranchPCRelCondCtr { - 0: bc({{ NIA = (uint32_t)(CIA + disp); }}); - } - - // Conditionally branch to fixed address based on CR and CTR. - format BranchNonPCRelCondCtr { - 1: bca({{ NIA = targetAddr; }}); - } - } - - 18: decode AA { - - // Unconditionally branch relative to PC. - format BranchPCRel { - 0: b({{ NIA = (uint32_t)(CIA + disp); }}); - } - - // Unconditionally branch to fixed address. - format BranchNonPCRel { - 1: ba({{ NIA = targetAddr; }}); - } - } - - 19: decode XO_XO { - - // Conditionally branch to address in LR based on CR and CTR. - format BranchLrCondCtr { - 16: bclr({{ NIA = LR & 0xfffffffc; }}); - } - - // Conditionally branch to address in CTR based on CR. - format BranchCtrCond { - 528: bcctr({{ NIA = CTR & 0xfffffffc; }}); - } - - // Condition register manipulation instructions. - format CondLogicOp { - 257: crand({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa & crBb); - }}); - 449: cror({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa | crBb); - }}); - 255: crnand({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, !(crBa & crBb)); - }}); - 193: crxor({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa ^ crBb); - }}); - 33: crnor({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, !(crBa | crBb)); - }}); - 289: creqv({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa == crBb); - }}); - 129: crandc({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa & !crBb); - }}); - 417: crorc({{ - uint32_t crBa = bits(CR, 31 - ba); - uint32_t crBb = bits(CR, 31 - bb); - CR = insertBits(CR, 31 - bt, crBa | !crBb); - }}); - } - format CondMoveOp { - 0: mcrf({{ - uint32_t crBfa = bits(CR, 31 - bfa*4, 28 - bfa*4); - CR = insertBits(CR, 31 - bf*4, 28 - bf*4, crBfa); - }}); - } - format MiscOp { - 150: isync({{ }}, [ IsSerializeAfter ]); - } - } - - format IntRotateOp { - 21: rlwinm({{ Ra = rotateValue(Rs, sh) & fullMask; }}); - 23: rlwnm({{ Ra = rotateValue(Rs, Rb) & fullMask; }}); - 20: rlwimi({{ Ra = (rotateValue(Rs, sh) & fullMask) | (Ra & ~fullMask); }}); - } - format LoadDispOp { - 34: lbz({{ Rt = Mem_ub; }}); - 40: lhz({{ Rt = Mem_uh; }}); - 42: lha({{ Rt = Mem_sh; }}); - 32: lwz({{ Rt = Mem; }}); - 58: lwa({{ Rt = Mem_sw; }}, - {{ EA = Ra + (disp & 0xfffffffc); }}, - {{ EA = disp & 0xfffffffc; }}); 48: lfs({{ Ft_sf = Mem_sf; }}); 50: lfd({{ Ft = Mem_df; }}); } format LoadDispUpdateOp { - 35: lbzu({{ Rt = Mem_ub; }}); - 41: lhzu({{ Rt = Mem_uh; }}); - 43: lhau({{ Rt = Mem_sh; }}); - 33: lwzu({{ Rt = Mem; }}); 49: lfsu({{ Ft_sf = Mem_sf; }}); 51: lfdu({{ Ft = Mem_df; }}); } format StoreDispOp { - 38: stb({{ Mem_ub = Rs_ub; }}); - 44: sth({{ Mem_uh = Rs_uh; }}); - 36: stw({{ Mem = Rs; }}); 52: stfs({{ Mem_sf = Fs_sf; }}); 54: stfd({{ Mem_df = Fs; }}); } format StoreDispUpdateOp { - 39: stbu({{ Mem_ub = Rs_ub; }}); - 45: sthu({{ Mem_uh = Rs_uh; }}); - 37: stwu({{ Mem = Rs; }}); 53: stfsu({{ Mem_sf = Fs_sf; }}); 55: stfdu({{ Mem_df = Fs; }}); } - 17: IntOp::sc({{ return std::make_shared(); }}); - format FloatArithOp { 59: decode A_XO { 21: fadds({{ Ft = Fa + Fb; }}); @@ -543,6 +580,21 @@ decode OPCODE default Unknown::unknown() { } default: decode XO_XO { + format FloatRCCheckOp { + 72: fmr({{ Ft = Fb; }}); + 264: fabs({{ + Ft_ud = Fb_ud; + Ft_ud = insertBits(Ft_ud, 63, 0); }}); + 136: fnabs({{ + Ft_ud = Fb_ud; + Ft_ud = insertBits(Ft_ud, 63, 1); }}); + 40: fneg({{ Ft = -Fb; }}); + 8: fcpsgn({{ + Ft_ud = Fb_ud; + Ft_ud = insertBits(Ft_ud, 63, Fa_ud<63:63>); + }}); + } + format FloatConvertOp { 12: frsp({{ Ft_sf = Fb; }}); 15: fctiwz({{ Ft_sw = (int32_t)trunc(Fb); }}); @@ -559,18 +611,6 @@ decode OPCODE default Unknown::unknown() { } format FloatRCCheckOp { - 72: fmr({{ Ft = Fb; }}); - 264: fabs({{ - Ft_ud = Fb_ud; - Ft_ud = insertBits(Ft_ud, 63, 0); }}); - 136: fnabs({{ - Ft_ud = Fb_ud; - Ft_ud = insertBits(Ft_ud, 63, 1); }}); - 40: fneg({{ Ft = -Fb; }}); - 8: fcpsgn({{ - Ft_ud = Fb_ud; - Ft_ud = insertBits(Ft_ud, 63, Fa_ud<63:63>); - }}); 583: mffs({{ Ft_ud = FPSCR; }}); 134: mtfsfi({{ FPSCR = insertCRField(FPSCR, BF + (8 * (1 - W_FIELD)), -- 2.30.2