diff options
Diffstat (limited to 'lib/Target/X86/X86InstrCompiler.td')
| -rw-r--r-- | lib/Target/X86/X86InstrCompiler.td | 220 |
1 files changed, 132 insertions, 88 deletions
diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td index 4c915d9..33534cd 100644 --- a/lib/Target/X86/X86InstrCompiler.td +++ b/lib/Target/X86/X86InstrCompiler.td @@ -214,6 +214,30 @@ def : Pat<(i64 (sext (i8 (X86setcc_c X86_COND_B, EFLAGS)))), def : Pat<(and (i8 (X86setcc_c X86_COND_B, EFLAGS)), 1), (SETBr)>; +// (add OP, SETB) -> (adc OP, 0) +def : Pat<(add (and (i8 (X86setcc_c X86_COND_B, EFLAGS)), 1), GR8:$op), + (ADC8ri GR8:$op, 0)>; +def : Pat<(add (and (i32 (X86setcc_c X86_COND_B, EFLAGS)), 1), GR32:$op), + (ADC32ri8 GR32:$op, 0)>; +def : Pat<(add (and (i64 (X86setcc_c X86_COND_B, EFLAGS)), 1), GR64:$op), + (ADC64ri8 GR64:$op, 0)>; + +// (sub OP, SETB) -> (sbb OP, 0) +def : Pat<(sub GR8:$op, (and (i8 (X86setcc_c X86_COND_B, EFLAGS)), 1)), + (SBB8ri GR8:$op, 0)>; +def : Pat<(sub GR32:$op, (and (i32 (X86setcc_c X86_COND_B, EFLAGS)), 1)), + (SBB32ri8 GR32:$op, 0)>; +def : Pat<(sub GR64:$op, (and (i64 (X86setcc_c X86_COND_B, EFLAGS)), 1)), + (SBB64ri8 GR64:$op, 0)>; + +// (sub OP, SETCC_CARRY) -> (adc OP, 0) +def : Pat<(sub GR8:$op, (i8 (X86setcc_c X86_COND_B, EFLAGS))), + (ADC8ri GR8:$op, 0)>; +def : Pat<(sub GR32:$op, (i32 (X86setcc_c X86_COND_B, EFLAGS))), + (ADC32ri8 GR32:$op, 0)>; +def : Pat<(sub GR64:$op, (i64 (X86setcc_c X86_COND_B, EFLAGS))), + (ADC64ri8 GR64:$op, 0)>; + //===----------------------------------------------------------------------===// // String Pseudo Instructions // @@ -519,85 +543,98 @@ def Int_MemBarrierNoSSE64 : RI<0x09, MRM1r, (outs), (ins GR64:$zero), Requires<[In64BitMode]>, LOCK; -// Optimized codegen when the non-memory output is not used. +// RegOpc corresponds to the mr version of the instruction +// ImmOpc corresponds to the mi version of the instruction +// ImmOpc8 corresponds to the mi8 version of the instruction +// ImmMod corresponds to the instruction format of the mi and mi8 versions +multiclass LOCK_ArithBinOp<bits<8> RegOpc, bits<8> ImmOpc, bits<8> ImmOpc8, + Format ImmMod, string mnemonic> { let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in { -def LOCK_ADD8mr : I<0x00, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src2), - "lock\n\t" - "add{b}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD16mr : I<0x01, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2), - "lock\n\t" - "add{w}\t{$src2, $dst|$dst, $src2}", []>, OpSize, LOCK; -def LOCK_ADD32mr : I<0x01, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2), - "lock\n\t" - "add{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD64mr : RI<0x01, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2), - "lock\n\t" - "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD8mi : Ii8<0x80, MRM0m, (outs), (ins i8mem :$dst, i8imm :$src2), - "lock\n\t" - "add{b}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD16mi : Ii16<0x81, MRM0m, (outs), (ins i16mem:$dst, i16imm:$src2), - "lock\n\t" - "add{w}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD32mi : Ii32<0x81, MRM0m, (outs), (ins i32mem:$dst, i32imm:$src2), - "lock\n\t" - "add{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD64mi32 : RIi32<0x81, MRM0m, (outs), - (ins i64mem:$dst, i64i32imm :$src2), - "lock\n\t" - "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; - -def LOCK_ADD16mi8 : Ii8<0x83, MRM0m, (outs), (ins i16mem:$dst, i16i8imm :$src2), - "lock\n\t" - "add{w}\t{$src2, $dst|$dst, $src2}", []>, OpSize, LOCK; -def LOCK_ADD32mi8 : Ii8<0x83, MRM0m, (outs), (ins i32mem:$dst, i32i8imm :$src2), - "lock\n\t" - "add{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_ADD64mi8 : RIi8<0x83, MRM0m, (outs), - (ins i64mem:$dst, i64i8imm :$src2), - "lock\n\t" - "add{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; - -def LOCK_SUB8mr : I<0x28, MRMDestMem, (outs), (ins i8mem :$dst, GR8 :$src2), - "lock\n\t" - "sub{b}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_SUB16mr : I<0x29, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2), - "lock\n\t" - "sub{w}\t{$src2, $dst|$dst, $src2}", []>, OpSize, LOCK; -def LOCK_SUB32mr : I<0x29, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2), - "lock\n\t" - "sub{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_SUB64mr : RI<0x29, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2), - "lock\n\t" - "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; +def #NAME#8mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4}, + RegOpc{3}, RegOpc{2}, RegOpc{1}, 0 }, + MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src2), + !strconcat("lock\n\t", mnemonic, "{b}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; +def #NAME#16mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4}, + RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 }, + MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2), + !strconcat("lock\n\t", mnemonic, "{w}\t", + "{$src2, $dst|$dst, $src2}"), + []>, OpSize, LOCK; +def #NAME#32mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4}, + RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 }, + MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2), + !strconcat("lock\n\t", mnemonic, "{l}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; +def #NAME#64mr : RI<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4}, + RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 }, + MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2), + !strconcat("lock\n\t", mnemonic, "{q}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; + +def #NAME#8mi : Ii8<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4}, + ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 0 }, + ImmMod, (outs), (ins i8mem :$dst, i8imm :$src2), + !strconcat("lock\n\t", mnemonic, "{b}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; + +def #NAME#16mi : Ii16<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4}, + ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 }, + ImmMod, (outs), (ins i16mem :$dst, i16imm :$src2), + !strconcat("lock\n\t", mnemonic, "{w}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; + +def #NAME#32mi : Ii32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4}, + ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 }, + ImmMod, (outs), (ins i32mem :$dst, i32imm :$src2), + !strconcat("lock\n\t", mnemonic, "{l}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; + +def #NAME#64mi32 : RIi32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4}, + ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 }, + ImmMod, (outs), (ins i64mem :$dst, i64i32imm :$src2), + !strconcat("lock\n\t", mnemonic, "{q}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; + +def #NAME#16mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4}, + ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 }, + ImmMod, (outs), (ins i16mem :$dst, i16i8imm :$src2), + !strconcat("lock\n\t", mnemonic, "{w}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; +def #NAME#32mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4}, + ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 }, + ImmMod, (outs), (ins i32mem :$dst, i32i8imm :$src2), + !strconcat("lock\n\t", mnemonic, "{l}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; +def #NAME#64mi8 : RIi8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4}, + ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 }, + ImmMod, (outs), (ins i64mem :$dst, i64i8imm :$src2), + !strconcat("lock\n\t", mnemonic, "{q}\t", + "{$src2, $dst|$dst, $src2}"), + []>, LOCK; +} -def LOCK_SUB8mi : Ii8<0x80, MRM5m, (outs), (ins i8mem :$dst, i8imm:$src2), - "lock\n\t" - "sub{b}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_SUB16mi : Ii16<0x81, MRM5m, (outs), (ins i16mem:$dst, i16imm:$src2), - "lock\n\t" - "sub{w}\t{$src2, $dst|$dst, $src2}", []>, OpSize, LOCK; -def LOCK_SUB32mi : Ii32<0x81, MRM5m, (outs), (ins i32mem:$dst, i32imm:$src2), - "lock\n\t" - "sub{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_SUB64mi32 : RIi32<0x81, MRM5m, (outs), - (ins i64mem:$dst, i64i32imm:$src2), - "lock\n\t" - "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; +} +defm LOCK_ADD : LOCK_ArithBinOp<0x00, 0x80, 0x83, MRM0m, "add">; +defm LOCK_SUB : LOCK_ArithBinOp<0x28, 0x80, 0x83, MRM5m, "sub">; +defm LOCK_OR : LOCK_ArithBinOp<0x08, 0x80, 0x83, MRM1m, "or">; +defm LOCK_AND : LOCK_ArithBinOp<0x08, 0x80, 0x83, MRM4m, "and">; +defm LOCK_XOR : LOCK_ArithBinOp<0x08, 0x80, 0x83, MRM6m, "xor">; -def LOCK_SUB16mi8 : Ii8<0x83, MRM5m, (outs), (ins i16mem:$dst, i16i8imm :$src2), - "lock\n\t" - "sub{w}\t{$src2, $dst|$dst, $src2}", []>, OpSize, LOCK; -def LOCK_SUB32mi8 : Ii8<0x83, MRM5m, (outs), (ins i32mem:$dst, i32i8imm :$src2), - "lock\n\t" - "sub{l}\t{$src2, $dst|$dst, $src2}", []>, LOCK; -def LOCK_SUB64mi8 : RIi8<0x83, MRM5m, (outs), - (ins i64mem:$dst, i64i8imm :$src2), - "lock\n\t" - "sub{q}\t{$src2, $dst|$dst, $src2}", []>, LOCK; +// Optimized codegen when the non-memory output is not used. +let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in { def LOCK_INC8m : I<0xFE, MRM0m, (outs), (ins i8mem :$dst), "lock\n\t" @@ -960,7 +997,8 @@ def : Pat<(extloadi64i32 addr:$src), // anyext. Define these to do an explicit zero-extend to // avoid partial-register updates. -def : Pat<(i16 (anyext GR8 :$src)), (MOVZX16rr8 GR8 :$src)>; +def : Pat<(i16 (anyext GR8 :$src)), (EXTRACT_SUBREG + (MOVZX32rr8 GR8 :$src), sub_16bit)>; def : Pat<(i32 (anyext GR8 :$src)), (MOVZX32rr8 GR8 :$src)>; // Except for i16 -> i32 since isel expect i16 ops to be promoted to i32. @@ -1127,9 +1165,9 @@ def : Pat<(and GR32:$src1, 0xff), Requires<[In32BitMode]>; // r & (2^8-1) ==> movz def : Pat<(and GR16:$src1, 0xff), - (MOVZX16rr8 (EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src1, - GR16_ABCD)), - sub_8bit))>, + (EXTRACT_SUBREG (MOVZX32rr8 (EXTRACT_SUBREG + (i16 (COPY_TO_REGCLASS GR16:$src1, GR16_ABCD)), sub_8bit)), + sub_16bit)>, Requires<[In32BitMode]>; // r & (2^32-1) ==> movz @@ -1147,7 +1185,8 @@ def : Pat<(and GR32:$src1, 0xff), Requires<[In64BitMode]>; // r & (2^8-1) ==> movz def : Pat<(and GR16:$src1, 0xff), - (MOVZX16rr8 (i8 (EXTRACT_SUBREG GR16:$src1, sub_8bit)))>, + (EXTRACT_SUBREG (MOVZX32rr8 (i8 + (EXTRACT_SUBREG GR16:$src1, sub_8bit))), sub_16bit)>, Requires<[In64BitMode]>; @@ -1159,10 +1198,11 @@ def : Pat<(sext_inreg GR32:$src, i8), GR32_ABCD)), sub_8bit))>, Requires<[In32BitMode]>; + def : Pat<(sext_inreg GR16:$src, i8), - (MOVSX16rr8 (EXTRACT_SUBREG (i16 (COPY_TO_REGCLASS GR16:$src, - GR16_ABCD)), - sub_8bit))>, + (EXTRACT_SUBREG (i32 (MOVSX32rr8 (EXTRACT_SUBREG + (i32 (COPY_TO_REGCLASS GR16:$src, GR16_ABCD)), sub_8bit))), + sub_16bit)>, Requires<[In32BitMode]>; def : Pat<(sext_inreg GR64:$src, i32), @@ -1175,9 +1215,19 @@ def : Pat<(sext_inreg GR32:$src, i8), (MOVSX32rr8 (EXTRACT_SUBREG GR32:$src, sub_8bit))>, Requires<[In64BitMode]>; def : Pat<(sext_inreg GR16:$src, i8), - (MOVSX16rr8 (i8 (EXTRACT_SUBREG GR16:$src, sub_8bit)))>, + (EXTRACT_SUBREG (MOVSX32rr8 + (EXTRACT_SUBREG GR16:$src, sub_8bit)), sub_16bit)>, Requires<[In64BitMode]>; +// sext, sext_load, zext, zext_load +def: Pat<(i16 (sext GR8:$src)), + (EXTRACT_SUBREG (MOVSX32rr8 GR8:$src), sub_16bit)>; +def: Pat<(sextloadi16i8 addr:$src), + (EXTRACT_SUBREG (MOVSX32rm8 addr:$src), sub_16bit)>; +def: Pat<(i16 (zext GR8:$src)), + (EXTRACT_SUBREG (MOVZX32rr8 GR8:$src), sub_16bit)>; +def: Pat<(zextloadi16i8 addr:$src), + (EXTRACT_SUBREG (MOVZX32rm8 addr:$src), sub_16bit)>; // trunc patterns def : Pat<(i16 (trunc GR32:$src)), @@ -1474,12 +1524,6 @@ def : Pat<(mul (loadi16 addr:$src1), i16immSExt8:$src2), def : Pat<(mul (loadi32 addr:$src1), i32immSExt8:$src2), (IMUL32rmi8 addr:$src1, i32immSExt8:$src2)>; -// Optimize multiply by 2 with EFLAGS result. -let AddedComplexity = 2 in { -def : Pat<(X86smul_flag GR16:$src1, 2), (ADD16rr GR16:$src1, GR16:$src1)>; -def : Pat<(X86smul_flag GR32:$src1, 2), (ADD32rr GR32:$src1, GR32:$src1)>; -} - // Patterns for nodes that do not produce flags, for instructions that do. // addition |
