llvm-6502/lib/Target/PowerPC/PPCInstr64Bit.td

1071 lines
48 KiB
TableGen
Raw Normal View History

//===-- PPCInstr64Bit.td - The PowerPC 64-bit Support ------*- tablegen -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file describes the PowerPC 64-bit instructions. These patterns are used
// both when in ppc64 mode and when in "use 64-bit extensions in 32-bit" mode.
//
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// 64-bit operands.
//
def s16imm64 : Operand<i64> {
let PrintMethod = "printS16ImmOperand";
let EncoderMethod = "getImm16Encoding";
let ParserMatchClass = PPCS16ImmAsmOperand;
}
def u16imm64 : Operand<i64> {
let PrintMethod = "printU16ImmOperand";
let EncoderMethod = "getImm16Encoding";
let ParserMatchClass = PPCU16ImmAsmOperand;
}
def s17imm64 : Operand<i64> {
// This operand type is used for addis/lis to allow the assembler parser
// to accept immediates in the range -65536..65535 for compatibility with
// the GNU assembler. The operand is treated as 16-bit otherwise.
let PrintMethod = "printS16ImmOperand";
let EncoderMethod = "getImm16Encoding";
let ParserMatchClass = PPCS17ImmAsmOperand;
}
def tocentry : Operand<iPTR> {
let MIOperandInfo = (ops i64imm:$imm);
}
def PPCTLSRegOperand : AsmOperandClass {
let Name = "TLSReg"; let PredicateMethod = "isTLSReg";
let RenderMethod = "addTLSRegOperands";
}
def tlsreg : Operand<i64> {
let EncoderMethod = "getTLSRegEncoding";
let ParserMatchClass = PPCTLSRegOperand;
}
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
def tlsgd : Operand<i64> {}
def tlscall : Operand<i64> {
let PrintMethod = "printTLSCall";
let MIOperandInfo = (ops calltarget:$func, tlsgd:$sym);
let EncoderMethod = "getTLSCallEncoding";
}
//===----------------------------------------------------------------------===//
// 64-bit transformation functions.
//
def SHL64 : SDNodeXForm<imm, [{
// Transformation function: 63 - imm
return getI32Imm(63 - N->getZExtValue());
}]>;
def SRL64 : SDNodeXForm<imm, [{
// Transformation function: 64 - imm
return N->getZExtValue() ? getI32Imm(64 - N->getZExtValue()) : getI32Imm(0);
}]>;
def HI32_48 : SDNodeXForm<imm, [{
// Transformation function: shift the immediate value down into the low bits.
return getI32Imm((unsigned short)(N->getZExtValue() >> 32));
}]>;
def HI48_64 : SDNodeXForm<imm, [{
// Transformation function: shift the immediate value down into the low bits.
return getI32Imm((unsigned short)(N->getZExtValue() >> 48));
}]>;
//===----------------------------------------------------------------------===//
// Calls.
//
let Interpretation64Bit = 1 in {
let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7 in {
let isBranch = 1, isIndirectBranch = 1, Uses = [CTR8] in {
def BCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
Requires<[In64BitMode]>;
let isCodeGenOnly = 1 in
def BCCTR8 : XLForm_2_br<19, 528, 0, (outs), (ins pred:$cond),
"b${cond:cc}ctr${cond:pm} ${cond:reg}", BrB, []>,
Requires<[In64BitMode]>;
}
}
let Defs = [LR8] in
def MovePCtoLR8 : Pseudo<(outs), (ins), "#MovePCtoLR8", []>,
PPC970_Unit_BRU;
let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7 in {
let Defs = [CTR8], Uses = [CTR8] in {
def BDZ8 : BForm_1<16, 18, 0, 0, (outs), (ins condbrtarget:$dst),
"bdz $dst">;
def BDNZ8 : BForm_1<16, 16, 0, 0, (outs), (ins condbrtarget:$dst),
"bdnz $dst">;
}
let isReturn = 1, Defs = [CTR8], Uses = [CTR8, LR8, RM] in {
def BDZLR8 : XLForm_2_ext<19, 16, 18, 0, 0, (outs), (ins),
"bdzlr", BrB, []>;
def BDNZLR8 : XLForm_2_ext<19, 16, 16, 0, 0, (outs), (ins),
"bdnzlr", BrB, []>;
}
}
let isCall = 1, PPC970_Unit = 7, Defs = [LR8] in {
// Convenient aliases for call instructions
let Uses = [RM] in {
def BL8 : IForm<18, 0, 1, (outs), (ins calltarget:$func),
"bl $func", BrB, []>; // See Pat patterns below.
def BL8_TLS : IForm<18, 0, 1, (outs), (ins tlscall:$func),
"bl $func", BrB, []>;
def BLA8 : IForm<18, 1, 1, (outs), (ins abscalltarget:$func),
"bla $func", BrB, [(PPCcall (i64 imm:$func))]>;
}
let Uses = [RM], isCodeGenOnly = 1 in {
def BL8_NOP : IForm_and_DForm_4_zero<18, 0, 1, 24,
(outs), (ins calltarget:$func),
"bl $func\n\tnop", BrB, []>;
def BL8_NOP_TLS : IForm_and_DForm_4_zero<18, 0, 1, 24,
(outs), (ins tlscall:$func),
"bl $func\n\tnop", BrB, []>;
def BLA8_NOP : IForm_and_DForm_4_zero<18, 1, 1, 24,
(outs), (ins abscalltarget:$func),
"bla $func\n\tnop", BrB,
[(PPCcall_nop (i64 imm:$func))]>;
}
let Uses = [CTR8, RM] in {
def BCTRL8 : XLForm_2_ext<19, 528, 20, 0, 1, (outs), (ins),
"bctrl", BrB, [(PPCbctrl)]>,
Requires<[In64BitMode]>;
let isCodeGenOnly = 1 in
def BCCTRL8 : XLForm_2_br<19, 528, 1, (outs), (ins pred:$cond),
"b${cond:cc}ctrl${cond:pm} ${cond:reg}", BrB, []>,
Requires<[In64BitMode]>;
}
}
} // Interpretation64Bit
// Calls
def : Pat<(PPCcall (i64 tglobaladdr:$dst)),
(BL8 tglobaladdr:$dst)>;
def : Pat<(PPCcall_nop (i64 tglobaladdr:$dst)),
(BL8_NOP tglobaladdr:$dst)>;
def : Pat<(PPCcall (i64 texternalsym:$dst)),
(BL8 texternalsym:$dst)>;
def : Pat<(PPCcall_nop (i64 texternalsym:$dst)),
(BL8_NOP texternalsym:$dst)>;
// Atomic operations
let usesCustomInserter = 1 in {
let Defs = [CR0] in {
def ATOMIC_LOAD_ADD_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_ADD_I64",
[(set i64:$dst, (atomic_load_add_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_LOAD_SUB_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_SUB_I64",
[(set i64:$dst, (atomic_load_sub_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_LOAD_OR_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_OR_I64",
[(set i64:$dst, (atomic_load_or_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_LOAD_XOR_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_XOR_I64",
[(set i64:$dst, (atomic_load_xor_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_LOAD_AND_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_AND_i64",
[(set i64:$dst, (atomic_load_and_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_LOAD_NAND_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$incr), "#ATOMIC_LOAD_NAND_I64",
[(set i64:$dst, (atomic_load_nand_64 xoaddr:$ptr, i64:$incr))]>;
def ATOMIC_CMP_SWAP_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$old, g8rc:$new), "#ATOMIC_CMP_SWAP_I64",
[(set i64:$dst, (atomic_cmp_swap_64 xoaddr:$ptr, i64:$old, i64:$new))]>;
def ATOMIC_SWAP_I64 : Pseudo<
(outs g8rc:$dst), (ins memrr:$ptr, g8rc:$new), "#ATOMIC_SWAP_I64",
[(set i64:$dst, (atomic_swap_64 xoaddr:$ptr, i64:$new))]>;
}
}
// Instructions to support atomic operations
def LDARX : XForm_1<31, 84, (outs g8rc:$rD), (ins memrr:$ptr),
"ldarx $rD, $ptr", LdStLDARX,
[(set i64:$rD, (PPClarx xoaddr:$ptr))]>;
let Defs = [CR0] in
def STDCX : XForm_1<31, 214, (outs), (ins g8rc:$rS, memrr:$dst),
"stdcx. $rS, $dst", LdStSTDCX,
[(PPCstcx i64:$rS, xoaddr:$dst)]>,
isDOT;
let Interpretation64Bit = 1 in {
let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [RM] in
def TCRETURNdi8 :Pseudo< (outs),
(ins calltarget:$dst, i32imm:$offset),
"#TC_RETURNd8 $dst $offset",
[]>;
let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [RM] in
def TCRETURNai8 :Pseudo<(outs), (ins abscalltarget:$func, i32imm:$offset),
"#TC_RETURNa8 $func $offset",
[(PPCtc_return (i64 imm:$func), imm:$offset)]>;
let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [RM] in
def TCRETURNri8 : Pseudo<(outs), (ins CTRRC8:$dst, i32imm:$offset),
"#TC_RETURNr8 $dst $offset",
[]>;
let isCodeGenOnly = 1 in {
let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7, isBranch = 1,
isIndirectBranch = 1, isCall = 1, isReturn = 1, Uses = [CTR8, RM] in
def TAILBCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
Requires<[In64BitMode]>;
let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7,
isBarrier = 1, isCall = 1, isReturn = 1, Uses = [RM] in
def TAILB8 : IForm<18, 0, 0, (outs), (ins calltarget:$dst),
"b $dst", BrB,
[]>;
let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7,
isBarrier = 1, isCall = 1, isReturn = 1, Uses = [RM] in
def TAILBA8 : IForm<18, 0, 0, (outs), (ins abscalltarget:$dst),
"ba $dst", BrB,
[]>;
}
} // Interpretation64Bit
def : Pat<(PPCtc_return (i64 tglobaladdr:$dst), imm:$imm),
(TCRETURNdi8 tglobaladdr:$dst, imm:$imm)>;
def : Pat<(PPCtc_return (i64 texternalsym:$dst), imm:$imm),
(TCRETURNdi8 texternalsym:$dst, imm:$imm)>;
def : Pat<(PPCtc_return CTRRC8:$dst, imm:$imm),
(TCRETURNri8 CTRRC8:$dst, imm:$imm)>;
// 64-bit CR instructions
let Interpretation64Bit = 1 in {
let neverHasSideEffects = 1 in {
def MTOCRF8: XFXForm_5a<31, 144, (outs crbitm:$FXM), (ins g8rc:$ST),
"mtocrf $FXM, $ST", BrMCRX>,
PPC970_DGroup_First, PPC970_Unit_CRU;
def MTCRF8 : XFXForm_5<31, 144, (outs), (ins i32imm:$FXM, g8rc:$rS),
"mtcrf $FXM, $rS", BrMCRX>,
PPC970_MicroCode, PPC970_Unit_CRU;
def MFOCRF8: XFXForm_5a<31, 19, (outs g8rc:$rT), (ins crbitm:$FXM),
"mfocrf $rT, $FXM", SprMFCR>,
PPC970_DGroup_First, PPC970_Unit_CRU;
def MFCR8 : XFXForm_3<31, 19, (outs g8rc:$rT), (ins),
"mfcr $rT", SprMFCR>,
PPC970_MicroCode, PPC970_Unit_CRU;
} // neverHasSideEffects = 1
let hasSideEffects = 1, isBarrier = 1, usesCustomInserter = 1 in {
let Defs = [CTR8] in
def EH_SjLj_SetJmp64 : Pseudo<(outs gprc:$dst), (ins memr:$buf),
"#EH_SJLJ_SETJMP64",
[(set i32:$dst, (PPCeh_sjlj_setjmp addr:$buf))]>,
Requires<[In64BitMode]>;
let isTerminator = 1 in
def EH_SjLj_LongJmp64 : Pseudo<(outs), (ins memr:$buf),
"#EH_SJLJ_LONGJMP64",
[(PPCeh_sjlj_longjmp addr:$buf)]>,
Requires<[In64BitMode]>;
}
//===----------------------------------------------------------------------===//
// 64-bit SPR manipulation instrs.
let Uses = [CTR8] in {
def MFCTR8 : XFXForm_1_ext<31, 339, 9, (outs g8rc:$rT), (ins),
"mfctr $rT", SprMFSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
}
let Pattern = [(PPCmtctr i64:$rS)], Defs = [CTR8] in {
def MTCTR8 : XFXForm_7_ext<31, 467, 9, (outs), (ins g8rc:$rS),
"mtctr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
}
Implement PPC counter loops as a late IR-level pass The old PPCCTRLoops pass, like the Hexagon pass version from which it was derived, could only handle some simple loops in canonical form. We cannot directly adapt the new Hexagon hardware loops pass, however, because the Hexagon pass contains a fundamental assumption that non-constant-trip-count loops will contain a guard, and this is not always true (the result being that incorrect negative counts can be generated). With this commit, we replace the pass with a late IR-level pass which makes use of SE to calculate the backedge-taken counts and safely generate the loop-count expressions (including any necessary max() parts). This IR level pass inserts custom intrinsics that are lowered into the desired decrement-and-branch instructions. The most fragile part of this new implementation is that interfering uses of the counter register must be detected on the IR level (and, on PPC, this also includes any indirect branches in addition to function calls). Also, to make all of this work, we need a variant of the mtctr instruction that is marked as having side effects. Without this, machine-code level CSE, DCE, etc. illegally transform the resulting code. Hopefully, this can be improved in the future. This new pass is smaller than the original (and much smaller than the new Hexagon hardware loops pass), and can handle many additional cases correctly. In addition, the preheader-creation code has been copied from LoopSimplify, and after we decide on where it belongs, this code will be refactored so that it can be explicitly shared (making this implementation even smaller). The new test-case files ctrloop-{le,lt,ne}.ll have been adapted from tests for the new Hexagon pass. There are a few classes of loops that this pass does not transform (noted by FIXMEs in the files), but these deficiencies can be addressed within the SE infrastructure (thus helping many other passes as well). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181927 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-15 21:37:41 +00:00
let hasSideEffects = 1, isCodeGenOnly = 1, Defs = [CTR8] in {
let Pattern = [(int_ppc_mtctr i64:$rS)] in
def MTCTR8loop : XFXForm_7_ext<31, 467, 9, (outs), (ins g8rc:$rS),
"mtctr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
Implement PPC counter loops as a late IR-level pass The old PPCCTRLoops pass, like the Hexagon pass version from which it was derived, could only handle some simple loops in canonical form. We cannot directly adapt the new Hexagon hardware loops pass, however, because the Hexagon pass contains a fundamental assumption that non-constant-trip-count loops will contain a guard, and this is not always true (the result being that incorrect negative counts can be generated). With this commit, we replace the pass with a late IR-level pass which makes use of SE to calculate the backedge-taken counts and safely generate the loop-count expressions (including any necessary max() parts). This IR level pass inserts custom intrinsics that are lowered into the desired decrement-and-branch instructions. The most fragile part of this new implementation is that interfering uses of the counter register must be detected on the IR level (and, on PPC, this also includes any indirect branches in addition to function calls). Also, to make all of this work, we need a variant of the mtctr instruction that is marked as having side effects. Without this, machine-code level CSE, DCE, etc. illegally transform the resulting code. Hopefully, this can be improved in the future. This new pass is smaller than the original (and much smaller than the new Hexagon hardware loops pass), and can handle many additional cases correctly. In addition, the preheader-creation code has been copied from LoopSimplify, and after we decide on where it belongs, this code will be refactored so that it can be explicitly shared (making this implementation even smaller). The new test-case files ctrloop-{le,lt,ne}.ll have been adapted from tests for the new Hexagon pass. There are a few classes of loops that this pass does not transform (noted by FIXMEs in the files), but these deficiencies can be addressed within the SE infrastructure (thus helping many other passes as well). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181927 91177308-0d34-0410-b5e6-96231b3b80d8
2013-05-15 21:37:41 +00:00
}
let isCodeGenOnly = 1, Pattern = [(set i64:$rT, readcyclecounter)] in
def MFTB8 : XFXForm_1_ext<31, 339, 268, (outs g8rc:$rT), (ins),
"mfspr $rT, 268", SprMFTB>,
PPC970_DGroup_First, PPC970_Unit_FXU;
// Note that encoding mftb using mfspr is now the preferred form,
// and has been since at least ISA v2.03. The mftb instruction has
// now been phased out. Using mfspr, however, is known not to work on
// the POWER3.
let Defs = [X1], Uses = [X1] in
def DYNALLOC8 : Pseudo<(outs g8rc:$result), (ins g8rc:$negsize, memri:$fpsi),"#DYNALLOC8",
[(set i64:$result,
(PPCdynalloc i64:$negsize, iaddr:$fpsi))]>;
let Defs = [LR8] in {
def MTLR8 : XFXForm_7_ext<31, 467, 8, (outs), (ins g8rc:$rS),
"mtlr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
}
let Uses = [LR8] in {
def MFLR8 : XFXForm_1_ext<31, 339, 8, (outs g8rc:$rT), (ins),
"mflr $rT", SprMFSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
}
} // Interpretation64Bit
//===----------------------------------------------------------------------===//
// Fixed point instructions.
//
let PPC970_Unit = 1 in { // FXU Operations.
let Interpretation64Bit = 1 in {
let neverHasSideEffects = 1 in {
let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in {
def LI8 : DForm_2_r0<14, (outs g8rc:$rD), (ins s16imm64:$imm),
"li $rD, $imm", IntSimple,
[(set i64:$rD, imm64SExt16:$imm)]>;
def LIS8 : DForm_2_r0<15, (outs g8rc:$rD), (ins s17imm64:$imm),
"lis $rD, $imm", IntSimple,
[(set i64:$rD, imm16ShiftedSExt:$imm)]>;
}
// Logical ops.
defm NAND8: XForm_6r<31, 476, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"nand", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (not (and i64:$rS, i64:$rB)))]>;
defm AND8 : XForm_6r<31, 28, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"and", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (and i64:$rS, i64:$rB))]>;
defm ANDC8: XForm_6r<31, 60, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"andc", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (and i64:$rS, (not i64:$rB)))]>;
defm OR8 : XForm_6r<31, 444, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"or", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (or i64:$rS, i64:$rB))]>;
defm NOR8 : XForm_6r<31, 124, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"nor", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (not (or i64:$rS, i64:$rB)))]>;
defm ORC8 : XForm_6r<31, 412, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"orc", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (or i64:$rS, (not i64:$rB)))]>;
defm EQV8 : XForm_6r<31, 284, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"eqv", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (not (xor i64:$rS, i64:$rB)))]>;
defm XOR8 : XForm_6r<31, 316, (outs g8rc:$rA), (ins g8rc:$rS, g8rc:$rB),
"xor", "$rA, $rS, $rB", IntSimple,
[(set i64:$rA, (xor i64:$rS, i64:$rB))]>;
// Logical ops with immediate.
let Defs = [CR0] in {
def ANDIo8 : DForm_4<28, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"andi. $dst, $src1, $src2", IntGeneral,
[(set i64:$dst, (and i64:$src1, immZExt16:$src2))]>,
isDOT;
def ANDISo8 : DForm_4<29, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"andis. $dst, $src1, $src2", IntGeneral,
[(set i64:$dst, (and i64:$src1, imm16ShiftedZExt:$src2))]>,
isDOT;
}
def ORI8 : DForm_4<24, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"ori $dst, $src1, $src2", IntSimple,
[(set i64:$dst, (or i64:$src1, immZExt16:$src2))]>;
def ORIS8 : DForm_4<25, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"oris $dst, $src1, $src2", IntSimple,
[(set i64:$dst, (or i64:$src1, imm16ShiftedZExt:$src2))]>;
def XORI8 : DForm_4<26, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"xori $dst, $src1, $src2", IntSimple,
[(set i64:$dst, (xor i64:$src1, immZExt16:$src2))]>;
def XORIS8 : DForm_4<27, (outs g8rc:$dst), (ins g8rc:$src1, u16imm:$src2),
"xoris $dst, $src1, $src2", IntSimple,
[(set i64:$dst, (xor i64:$src1, imm16ShiftedZExt:$src2))]>;
defm ADD8 : XOForm_1r<31, 266, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"add", "$rT, $rA, $rB", IntSimple,
[(set i64:$rT, (add i64:$rA, i64:$rB))]>;
// ADD8 has a special form: reg = ADD8(reg, sym@tls) for use by the
// initial-exec thread-local storage model.
def ADD8TLS : XOForm_1<31, 266, 0, (outs g8rc:$rT), (ins g8rc:$rA, tlsreg:$rB),
"add $rT, $rA, $rB", IntSimple,
[(set i64:$rT, (add i64:$rA, tglobaltlsaddr:$rB))]>;
defm ADDC8 : XOForm_1rc<31, 10, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"addc", "$rT, $rA, $rB", IntGeneral,
[(set i64:$rT, (addc i64:$rA, i64:$rB))]>,
PPC970_DGroup_Cracked;
let Defs = [CARRY] in
def ADDIC8 : DForm_2<12, (outs g8rc:$rD), (ins g8rc:$rA, s16imm64:$imm),
"addic $rD, $rA, $imm", IntGeneral,
[(set i64:$rD, (addc i64:$rA, imm64SExt16:$imm))]>;
def ADDI8 : DForm_2<14, (outs g8rc:$rD), (ins g8rc_nox0:$rA, s16imm64:$imm),
"addi $rD, $rA, $imm", IntSimple,
[(set i64:$rD, (add i64:$rA, imm64SExt16:$imm))]>;
def ADDIS8 : DForm_2<15, (outs g8rc:$rD), (ins g8rc_nox0:$rA, s17imm64:$imm),
"addis $rD, $rA, $imm", IntSimple,
[(set i64:$rD, (add i64:$rA, imm16ShiftedSExt:$imm))]>;
let Defs = [CARRY] in {
def SUBFIC8: DForm_2< 8, (outs g8rc:$rD), (ins g8rc:$rA, s16imm64:$imm),
"subfic $rD, $rA, $imm", IntGeneral,
[(set i64:$rD, (subc imm64SExt16:$imm, i64:$rA))]>;
defm SUBFC8 : XOForm_1r<31, 8, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"subfc", "$rT, $rA, $rB", IntGeneral,
[(set i64:$rT, (subc i64:$rB, i64:$rA))]>,
PPC970_DGroup_Cracked;
}
defm SUBF8 : XOForm_1r<31, 40, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"subf", "$rT, $rA, $rB", IntGeneral,
[(set i64:$rT, (sub i64:$rB, i64:$rA))]>;
defm NEG8 : XOForm_3r<31, 104, 0, (outs g8rc:$rT), (ins g8rc:$rA),
"neg", "$rT, $rA", IntSimple,
[(set i64:$rT, (ineg i64:$rA))]>;
let Uses = [CARRY] in {
defm ADDE8 : XOForm_1rc<31, 138, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"adde", "$rT, $rA, $rB", IntGeneral,
[(set i64:$rT, (adde i64:$rA, i64:$rB))]>;
defm ADDME8 : XOForm_3rc<31, 234, 0, (outs g8rc:$rT), (ins g8rc:$rA),
"addme", "$rT, $rA", IntGeneral,
[(set i64:$rT, (adde i64:$rA, -1))]>;
defm ADDZE8 : XOForm_3rc<31, 202, 0, (outs g8rc:$rT), (ins g8rc:$rA),
"addze", "$rT, $rA", IntGeneral,
[(set i64:$rT, (adde i64:$rA, 0))]>;
defm SUBFE8 : XOForm_1rc<31, 136, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"subfe", "$rT, $rA, $rB", IntGeneral,
[(set i64:$rT, (sube i64:$rB, i64:$rA))]>;
defm SUBFME8 : XOForm_3rc<31, 232, 0, (outs g8rc:$rT), (ins g8rc:$rA),
"subfme", "$rT, $rA", IntGeneral,
[(set i64:$rT, (sube -1, i64:$rA))]>;
defm SUBFZE8 : XOForm_3rc<31, 200, 0, (outs g8rc:$rT), (ins g8rc:$rA),
"subfze", "$rT, $rA", IntGeneral,
[(set i64:$rT, (sube 0, i64:$rA))]>;
}
defm MULHD : XOForm_1r<31, 73, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"mulhd", "$rT, $rA, $rB", IntMulHW,
[(set i64:$rT, (mulhs i64:$rA, i64:$rB))]>;
defm MULHDU : XOForm_1r<31, 9, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"mulhdu", "$rT, $rA, $rB", IntMulHWU,
[(set i64:$rT, (mulhu i64:$rA, i64:$rB))]>;
}
} // Interpretation64Bit
let isCompare = 1, neverHasSideEffects = 1 in {
def CMPD : XForm_16_ext<31, 0, (outs crrc:$crD), (ins g8rc:$rA, g8rc:$rB),
"cmpd $crD, $rA, $rB", IntCompare>, isPPC64;
def CMPLD : XForm_16_ext<31, 32, (outs crrc:$crD), (ins g8rc:$rA, g8rc:$rB),
"cmpld $crD, $rA, $rB", IntCompare>, isPPC64;
def CMPDI : DForm_5_ext<11, (outs crrc:$crD), (ins g8rc:$rA, s16imm:$imm),
"cmpdi $crD, $rA, $imm", IntCompare>, isPPC64;
def CMPLDI : DForm_6_ext<10, (outs crrc:$dst), (ins g8rc:$src1, u16imm:$src2),
"cmpldi $dst, $src1, $src2", IntCompare>, isPPC64;
}
let neverHasSideEffects = 1 in {
defm SLD : XForm_6r<31, 27, (outs g8rc:$rA), (ins g8rc:$rS, gprc:$rB),
"sld", "$rA, $rS, $rB", IntRotateD,
[(set i64:$rA, (PPCshl i64:$rS, i32:$rB))]>, isPPC64;
defm SRD : XForm_6r<31, 539, (outs g8rc:$rA), (ins g8rc:$rS, gprc:$rB),
"srd", "$rA, $rS, $rB", IntRotateD,
[(set i64:$rA, (PPCsrl i64:$rS, i32:$rB))]>, isPPC64;
defm SRAD : XForm_6rc<31, 794, (outs g8rc:$rA), (ins g8rc:$rS, gprc:$rB),
"srad", "$rA, $rS, $rB", IntRotateD,
[(set i64:$rA, (PPCsra i64:$rS, i32:$rB))]>, isPPC64;
let Interpretation64Bit = 1 in {
defm EXTSB8 : XForm_11r<31, 954, (outs g8rc:$rA), (ins g8rc:$rS),
"extsb", "$rA, $rS", IntSimple,
[(set i64:$rA, (sext_inreg i64:$rS, i8))]>;
defm EXTSH8 : XForm_11r<31, 922, (outs g8rc:$rA), (ins g8rc:$rS),
"extsh", "$rA, $rS", IntSimple,
[(set i64:$rA, (sext_inreg i64:$rS, i16))]>;
} // Interpretation64Bit
// For fast-isel:
let isCodeGenOnly = 1 in {
def EXTSB8_32_64 : XForm_11<31, 954, (outs g8rc:$rA), (ins gprc:$rS),
"extsb $rA, $rS", IntSimple, []>, isPPC64;
def EXTSH8_32_64 : XForm_11<31, 922, (outs g8rc:$rA), (ins gprc:$rS),
"extsh $rA, $rS", IntSimple, []>, isPPC64;
} // isCodeGenOnly for fast-isel
defm EXTSW : XForm_11r<31, 986, (outs g8rc:$rA), (ins g8rc:$rS),
"extsw", "$rA, $rS", IntSimple,
[(set i64:$rA, (sext_inreg i64:$rS, i32))]>, isPPC64;
let Interpretation64Bit = 1 in
defm EXTSW_32_64 : XForm_11r<31, 986, (outs g8rc:$rA), (ins gprc:$rS),
"extsw", "$rA, $rS", IntSimple,
[(set i64:$rA, (sext i32:$rS))]>, isPPC64;
defm SRADI : XSForm_1rc<31, 413, (outs g8rc:$rA), (ins g8rc:$rS, u6imm:$SH),
"sradi", "$rA, $rS, $SH", IntRotateDI,
[(set i64:$rA, (sra i64:$rS, (i32 imm:$SH)))]>, isPPC64;
defm CNTLZD : XForm_11r<31, 58, (outs g8rc:$rA), (ins g8rc:$rS),
"cntlzd", "$rA, $rS", IntGeneral,
[(set i64:$rA, (ctlz i64:$rS))]>;
defm POPCNTD : XForm_11r<31, 506, (outs g8rc:$rA), (ins g8rc:$rS),
"popcntd", "$rA, $rS", IntGeneral,
[(set i64:$rA, (ctpop i64:$rS))]>;
// popcntw also does a population count on the high 32 bits (storing the
// results in the high 32-bits of the output). We'll ignore that here (which is
// safe because we never separately use the high part of the 64-bit registers).
defm POPCNTW : XForm_11r<31, 378, (outs gprc:$rA), (ins gprc:$rS),
"popcntw", "$rA, $rS", IntGeneral,
[(set i32:$rA, (ctpop i32:$rS))]>;
defm DIVD : XOForm_1r<31, 489, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"divd", "$rT, $rA, $rB", IntDivD,
[(set i64:$rT, (sdiv i64:$rA, i64:$rB))]>, isPPC64,
PPC970_DGroup_First, PPC970_DGroup_Cracked;
defm DIVDU : XOForm_1r<31, 457, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"divdu", "$rT, $rA, $rB", IntDivD,
[(set i64:$rT, (udiv i64:$rA, i64:$rB))]>, isPPC64,
PPC970_DGroup_First, PPC970_DGroup_Cracked;
defm MULLD : XOForm_1r<31, 233, 0, (outs g8rc:$rT), (ins g8rc:$rA, g8rc:$rB),
"mulld", "$rT, $rA, $rB", IntMulHD,
[(set i64:$rT, (mul i64:$rA, i64:$rB))]>, isPPC64;
def MULLI8 : DForm_2<7, (outs g8rc:$rD), (ins g8rc:$rA, s16imm64:$imm),
"mulli $rD, $rA, $imm", IntMulLI,
[(set i64:$rD, (mul i64:$rA, imm64SExt16:$imm))]>;
}
let neverHasSideEffects = 1 in {
let isCommutable = 1 in {
defm RLDIMI : MDForm_1r<30, 3, (outs g8rc:$rA),
(ins g8rc:$rSi, g8rc:$rS, u6imm:$SH, u6imm:$MBE),
"rldimi", "$rA, $rS, $SH, $MBE", IntRotateDI,
[]>, isPPC64, RegConstraint<"$rSi = $rA">,
NoEncode<"$rSi">;
}
// Rotate instructions.
defm RLDCL : MDSForm_1r<30, 8,
(outs g8rc:$rA), (ins g8rc:$rS, gprc:$rB, u6imm:$MBE),
"rldcl", "$rA, $rS, $rB, $MBE", IntRotateD,
[]>, isPPC64;
defm RLDCR : MDSForm_1r<30, 9,
(outs g8rc:$rA), (ins g8rc:$rS, gprc:$rB, u6imm:$MBE),
"rldcr", "$rA, $rS, $rB, $MBE", IntRotateD,
[]>, isPPC64;
defm RLDICL : MDForm_1r<30, 0,
(outs g8rc:$rA), (ins g8rc:$rS, u6imm:$SH, u6imm:$MBE),
"rldicl", "$rA, $rS, $SH, $MBE", IntRotateDI,
[]>, isPPC64;
// For fast-isel:
let isCodeGenOnly = 1 in
def RLDICL_32_64 : MDForm_1<30, 0,
(outs g8rc:$rA),
(ins gprc:$rS, u6imm:$SH, u6imm:$MBE),
"rldicl $rA, $rS, $SH, $MBE", IntRotateDI,
[]>, isPPC64;
// End fast-isel.
defm RLDICR : MDForm_1r<30, 1,
(outs g8rc:$rA), (ins g8rc:$rS, u6imm:$SH, u6imm:$MBE),
"rldicr", "$rA, $rS, $SH, $MBE", IntRotateDI,
[]>, isPPC64;
defm RLDIC : MDForm_1r<30, 2,
(outs g8rc:$rA), (ins g8rc:$rS, u6imm:$SH, u6imm:$MBE),
"rldic", "$rA, $rS, $SH, $MBE", IntRotateDI,
[]>, isPPC64;
let Interpretation64Bit = 1 in {
defm RLWINM8 : MForm_2r<21, (outs g8rc:$rA),
(ins g8rc:$rS, u5imm:$SH, u5imm:$MB, u5imm:$ME),
"rlwinm", "$rA, $rS, $SH, $MB, $ME", IntGeneral,
[]>;
let isSelect = 1 in
def ISEL8 : AForm_4<31, 15,
(outs g8rc:$rT), (ins g8rc_nox0:$rA, g8rc:$rB, crbitrc:$cond),
"isel $rT, $rA, $rB, $cond", IntGeneral,
[]>;
} // Interpretation64Bit
} // neverHasSideEffects = 1
} // End FXU Operations.
//===----------------------------------------------------------------------===//
// Load/Store instructions.
//
// Sign extending loads.
let canFoldAsLoad = 1, PPC970_Unit = 2 in {
let Interpretation64Bit = 1 in
def LHA8: DForm_1<42, (outs g8rc:$rD), (ins memri:$src),
"lha $rD, $src", LdStLHA,
[(set i64:$rD, (sextloadi16 iaddr:$src))]>,
PPC970_DGroup_Cracked;
def LWA : DSForm_1<58, 2, (outs g8rc:$rD), (ins memrix:$src),
"lwa $rD, $src", LdStLWA,
[(set i64:$rD,
(aligned4sextloadi32 ixaddr:$src))]>, isPPC64,
PPC970_DGroup_Cracked;
let Interpretation64Bit = 1 in
def LHAX8: XForm_1<31, 343, (outs g8rc:$rD), (ins memrr:$src),
"lhax $rD, $src", LdStLHA,
[(set i64:$rD, (sextloadi16 xaddr:$src))]>,
PPC970_DGroup_Cracked;
def LWAX : XForm_1<31, 341, (outs g8rc:$rD), (ins memrr:$src),
"lwax $rD, $src", LdStLHA,
[(set i64:$rD, (sextloadi32 xaddr:$src))]>, isPPC64,
PPC970_DGroup_Cracked;
// Update forms.
let mayLoad = 1, neverHasSideEffects = 1 in {
let Interpretation64Bit = 1 in
def LHAU8 : DForm_1<43, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memri:$addr),
"lhau $rD, $addr", LdStLHAU,
[]>, RegConstraint<"$addr.reg = $ea_result">,
NoEncode<"$ea_result">;
// NO LWAU!
let Interpretation64Bit = 1 in
def LHAUX8 : XForm_1<31, 375, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"lhaux $rD, $addr", LdStLHAU,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">;
def LWAUX : XForm_1<31, 373, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"lwaux $rD, $addr", LdStLHAU,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">, isPPC64;
}
}
let Interpretation64Bit = 1 in {
// Zero extending loads.
let canFoldAsLoad = 1, PPC970_Unit = 2 in {
def LBZ8 : DForm_1<34, (outs g8rc:$rD), (ins memri:$src),
"lbz $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi8 iaddr:$src))]>;
def LHZ8 : DForm_1<40, (outs g8rc:$rD), (ins memri:$src),
"lhz $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi16 iaddr:$src))]>;
def LWZ8 : DForm_1<32, (outs g8rc:$rD), (ins memri:$src),
"lwz $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi32 iaddr:$src))]>, isPPC64;
def LBZX8 : XForm_1<31, 87, (outs g8rc:$rD), (ins memrr:$src),
"lbzx $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi8 xaddr:$src))]>;
def LHZX8 : XForm_1<31, 279, (outs g8rc:$rD), (ins memrr:$src),
"lhzx $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi16 xaddr:$src))]>;
def LWZX8 : XForm_1<31, 23, (outs g8rc:$rD), (ins memrr:$src),
"lwzx $rD, $src", LdStLoad,
[(set i64:$rD, (zextloadi32 xaddr:$src))]>;
// Update forms.
let mayLoad = 1, neverHasSideEffects = 1 in {
def LBZU8 : DForm_1<35, (outs g8rc:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
"lbzu $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.reg = $ea_result">,
NoEncode<"$ea_result">;
def LHZU8 : DForm_1<41, (outs g8rc:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
"lhzu $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.reg = $ea_result">,
NoEncode<"$ea_result">;
def LWZU8 : DForm_1<33, (outs g8rc:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
"lwzu $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.reg = $ea_result">,
NoEncode<"$ea_result">;
def LBZUX8 : XForm_1<31, 119, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"lbzux $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">;
def LHZUX8 : XForm_1<31, 311, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"lhzux $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">;
def LWZUX8 : XForm_1<31, 55, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"lwzux $rD, $addr", LdStLoadUpd,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">;
}
}
} // Interpretation64Bit
// Full 8-byte loads.
let canFoldAsLoad = 1, PPC970_Unit = 2 in {
def LD : DSForm_1<58, 0, (outs g8rc:$rD), (ins memrix:$src),
"ld $rD, $src", LdStLD,
[(set i64:$rD, (aligned4load ixaddr:$src))]>, isPPC64;
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
// The following three definitions are selected for small code model only.
// Otherwise, we need to create two instructions to form a 32-bit offset,
// so we have a custom matcher for TOC_ENTRY in PPCDAGToDAGIsel::Select().
def LDtoc: Pseudo<(outs g8rc:$rD), (ins tocentry:$disp, g8rc:$reg),
"#LDtoc",
[(set i64:$rD,
(PPCtoc_entry tglobaladdr:$disp, i64:$reg))]>, isPPC64;
def LDtocJTI: Pseudo<(outs g8rc:$rD), (ins tocentry:$disp, g8rc:$reg),
"#LDtocJTI",
[(set i64:$rD,
(PPCtoc_entry tjumptable:$disp, i64:$reg))]>, isPPC64;
def LDtocCPT: Pseudo<(outs g8rc:$rD), (ins tocentry:$disp, g8rc:$reg),
"#LDtocCPT",
[(set i64:$rD,
(PPCtoc_entry tconstpool:$disp, i64:$reg))]>, isPPC64;
let hasSideEffects = 1, isCodeGenOnly = 1 in {
let RST = 2, DS = 2 in
def LDinto_toc: DSForm_1a<58, 0, (outs), (ins g8rc:$reg),
"ld 2, 8($reg)", LdStLD,
[(PPCload_toc i64:$reg)]>, isPPC64;
let RST = 2, DS = 10, RA = 1 in
def LDtoc_restore : DSForm_1a<58, 0, (outs), (ins),
"ld 2, 40(1)", LdStLD,
[(PPCtoc_restore)]>, isPPC64;
}
def LDX : XForm_1<31, 21, (outs g8rc:$rD), (ins memrr:$src),
"ldx $rD, $src", LdStLD,
[(set i64:$rD, (load xaddr:$src))]>, isPPC64;
def LDBRX : XForm_1<31, 532, (outs g8rc:$rD), (ins memrr:$src),
"ldbrx $rD, $src", LdStLoad,
[(set i64:$rD, (PPClbrx xoaddr:$src, i64))]>, isPPC64;
let mayLoad = 1, neverHasSideEffects = 1 in {
def LDU : DSForm_1<58, 1, (outs g8rc:$rD, ptr_rc_nor0:$ea_result), (ins memrix:$addr),
"ldu $rD, $addr", LdStLDU,
[]>, RegConstraint<"$addr.reg = $ea_result">, isPPC64,
NoEncode<"$ea_result">;
def LDUX : XForm_1<31, 53, (outs g8rc:$rD, ptr_rc_nor0:$ea_result),
(ins memrr:$addr),
"ldux $rD, $addr", LdStLDU,
[]>, RegConstraint<"$addr.ptrreg = $ea_result">,
NoEncode<"$ea_result">, isPPC64;
}
}
def : Pat<(PPCload ixaddr:$src),
(LD ixaddr:$src)>;
def : Pat<(PPCload xaddr:$src),
(LDX xaddr:$src)>;
// Support for medium and large code model.
def ADDIStocHA: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, tocentry:$disp),
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
"#ADDIStocHA",
[(set i64:$rD,
(PPCaddisTocHA i64:$reg, tglobaladdr:$disp))]>,
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
isPPC64;
def LDtocL: Pseudo<(outs g8rc:$rD), (ins tocentry:$disp, g8rc_nox0:$reg),
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
"#LDtocL",
[(set i64:$rD,
(PPCldTocL tglobaladdr:$disp, i64:$reg))]>, isPPC64;
def ADDItocL: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, tocentry:$disp),
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
"#ADDItocL",
[(set i64:$rD,
(PPCaddiTocL i64:$reg, tglobaladdr:$disp))]>, isPPC64;
This patch implements medium code model support for 64-bit PowerPC. The default for 64-bit PowerPC is small code model, in which TOC entries must be addressable using a 16-bit offset from the TOC pointer. Additionally, only TOC entries are addressed via the TOC pointer. With medium code model, TOC entries and data sections can all be addressed via the TOC pointer using a 32-bit offset. Cooperation with the linker allows 16-bit offsets to be used when these are sufficient, reducing the number of extra instructions that need to be executed. Medium code model also does not generate explicit TOC entries in ".section toc" for variables that are wholly internal to the compilation unit. Consider a load of an external 4-byte integer. With small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei With medium model, it instead generates: addis 3, 2, .LC1@toc@ha ld 3, .LC1@toc@l(3) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc ei[TC],ei Here .LC1@toc@ha is a relocation requesting the upper 16 bits of the 32-bit offset of ei's TOC entry from the TOC base pointer. Similarly, .LC1@toc@l is a relocation requesting the lower 16 bits. Note that if the linker determines that ei's TOC entry is within a 16-bit offset of the TOC base pointer, it will replace the "addis" with a "nop", and replace the "ld" with the identical "ld" instruction from the small code model example. Consider next a load of a function-scope static integer. For small code model, the compiler generates: ld 3, .LC1@toc(2) lwz 4, 0(3) .section .toc,"aw",@progbits .LC1: .tc test_fn_static.si[TC],test_fn_static.si .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 For medium code model, the compiler generates: addis 3, 2, test_fn_static.si@toc@ha addi 3, 3, test_fn_static.si@toc@l lwz 4, 0(3) .type test_fn_static.si,@object .local test_fn_static.si .comm test_fn_static.si,4,4 Again, the linker may replace the "addis" with a "nop", calculating only a 16-bit offset when this is sufficient. Note that it would be more efficient for the compiler to generate: addis 3, 2, test_fn_static.si@toc@ha lwz 4, test_fn_static.si@toc@l(3) The current patch does not perform this optimization yet. This will be addressed as a peephole optimization in a later patch. For the moment, the default code model for 64-bit PowerPC will remain the small code model. We plan to eventually change the default to medium code model, which matches current upstream GCC behavior. Note that the different code models are ABI-compatible, so code compiled with different models will be linked and execute correctly. I've tested the regression suite and the application/benchmark test suite in two ways: Once with the patch as submitted here, and once with additional logic to force medium code model as the default. The tests all compile cleanly, with one exception. The mandel-2 application test fails due to an unrelated ABI compatibility with passing complex numbers. It just so happens that small code model was incredibly lucky, in that temporary values in floating-point registers held the expected values needed by the external library routine that was called incorrectly. My current thought is to correct the ABI problems with _Complex before making medium code model the default, to avoid introducing this "regression." Here are a few comments on how the patch works, since the selection code can be difficult to follow: The existing logic for small code model defines three pseudo-instructions: LDtoc for most uses, LDtocJTI for jump table addresses, and LDtocCPT for constant pool addresses. These are expanded by SelectCodeCommon(). The pseudo-instruction approach doesn't work for medium code model, because we need to generate two instructions when we match the same pattern. Instead, new logic in PPCDAGToDAGISel::Select() intercepts the TOC_ENTRY node for medium code model, and generates an ADDIStocHA followed by either a LDtocL or an ADDItocL. These new node types correspond naturally to the sequences described above. The addis/ld sequence is generated for the following cases: * Jump table addresses * Function addresses * External global variables * Tentative definitions of global variables (common linkage) The addis/addi sequence is generated for the following cases: * Constant pool entries * File-scope static global variables * Function-scope static variables Expanding to the two-instruction sequences at select time exposes the instructions to subsequent optimization, particularly scheduling. The rest of the processing occurs at assembly time, in PPCAsmPrinter::EmitInstruction. Each of the instructions is converted to a "real" PowerPC instruction. When a TOC entry needs to be created, this is done here in the same manner as for the existing LDtoc, LDtocJTI, and LDtocCPT pseudo-instructions (I factored out a new routine to handle this). I had originally thought that if a TOC entry was needed for LDtocL or ADDItocL, it would already have been generated for the previous ADDIStocHA. However, at higher optimization levels, the ADDIStocHA may appear in a different block, which may be assembled textually following the block containing the LDtocL or ADDItocL. So it is necessary to include the possibility of creating a new TOC entry for those two instructions. Note that for LDtocL, we generate a new form of LD called LDrs. This allows specifying the @toc@l relocation for the offset field of the LD instruction (i.e., the offset is replaced by a SymbolLo relocation). When the peephole optimization described above is added, we will need to do similar things for all immediate-form load and store operations. The seven "mcm-n.ll" test cases are kept separate because otherwise the intermingling of various TOC entries and so forth makes the tests fragile and hard to understand. The above assumes use of an external assembler. For use of the integrated assembler, new relocations are added and used by PPCELFObjectWriter. Testing is done with "mcm-obj.ll", which tests for proper generation of the various relocations for the same sequences tested with the external assembler. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168708 91177308-0d34-0410-b5e6-96231b3b80d8
2012-11-27 17:35:46 +00:00
// Support for thread-local storage.
def ADDISgotTprelHA: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
"#ADDISgotTprelHA",
[(set i64:$rD,
(PPCaddisGotTprelHA i64:$reg,
tglobaltlsaddr:$disp))]>,
isPPC64;
def LDgotTprelL: Pseudo<(outs g8rc:$rD), (ins s16imm64:$disp, g8rc_nox0:$reg),
"#LDgotTprelL",
[(set i64:$rD,
(PPCldGotTprelL tglobaltlsaddr:$disp, i64:$reg))]>,
isPPC64;
def : Pat<(PPCaddTls i64:$in, tglobaltlsaddr:$g),
(ADD8TLS $in, tglobaltlsaddr:$g)>;
def ADDIStlsgdHA: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
"#ADDIStlsgdHA",
[(set i64:$rD,
(PPCaddisTlsgdHA i64:$reg, tglobaltlsaddr:$disp))]>,
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
isPPC64;
def ADDItlsgdL : Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
"#ADDItlsgdL",
[(set i64:$rD,
(PPCaddiTlsgdL i64:$reg, tglobaltlsaddr:$disp))]>,
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
isPPC64;
def GETtlsADDR : Pseudo<(outs g8rc:$rD), (ins g8rc:$reg, tlsgd:$sym),
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
"#GETtlsADDR",
[(set i64:$rD,
(PPCgetTlsAddr i64:$reg, tglobaltlsaddr:$sym))]>,
This patch implements the general dynamic TLS model for 64-bit PowerPC. Given a thread-local symbol x with global-dynamic access, the generated code to obtain x's address is: Instruction Relocation Symbol addis ra,r2,x@got@tlsgd@ha R_PPC64_GOT_TLSGD16_HA x addi r3,ra,x@got@tlsgd@l R_PPC64_GOT_TLSGD16_L x bl __tls_get_addr(x@tlsgd) R_PPC64_TLSGD x R_PPC64_REL24 __tls_get_addr nop <use address in r3> The implementation borrows from the medium code model work for introducing special forms of ADDIS and ADDI into the DAG representation. This is made slightly more complicated by having to introduce a call to the external function __tls_get_addr. Using the full call machinery is overkill and, more importantly, makes it difficult to add a special relocation. So I've introduced another opcode GET_TLS_ADDR to represent the function call, and surrounded it with register copies to set up the parameter and return value. Most of the code is pretty straightforward. I ran into one peculiarity when I introduced a new PPC opcode BL8_NOP_ELF_TLSGD, which is just like BL8_NOP_ELF except that it takes another parameter to represent the symbol ("x" above) that requires a relocation on the call. Something in the TblGen machinery causes BL8_NOP_ELF and BL8_NOP_ELF_TLSGD to be treated identically during the emit phase, so this second operand was never visited to generate relocations. This is the reason for the slightly messy workaround in PPCMCCodeEmitter.cpp:getDirectBrEncoding(). Two new tests are included to demonstrate correct external assembly and correct generation of relocations using the integrated assembler. Comments welcome! Thanks, Bill git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169910 91177308-0d34-0410-b5e6-96231b3b80d8
2012-12-11 20:30:11 +00:00
isPPC64;
def ADDIStlsldHA: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
"#ADDIStlsldHA",
[(set i64:$rD,
(PPCaddisTlsldHA i64:$reg, tglobaltlsaddr:$disp))]>,
isPPC64;
def ADDItlsldL : Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
"#ADDItlsldL",
[(set i64:$rD,
(PPCaddiTlsldL i64:$reg, tglobaltlsaddr:$disp))]>,
isPPC64;
def GETtlsldADDR : Pseudo<(outs g8rc:$rD), (ins g8rc:$reg, tlsgd:$sym),
"#GETtlsldADDR",
[(set i64:$rD,
(PPCgetTlsldAddr i64:$reg, tglobaltlsaddr:$sym))]>,
isPPC64;
def ADDISdtprelHA: Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
"#ADDISdtprelHA",
[(set i64:$rD,
(PPCaddisDtprelHA i64:$reg,
tglobaltlsaddr:$disp))]>,
isPPC64;
def ADDIdtprelL : Pseudo<(outs g8rc:$rD), (ins g8rc_nox0:$reg, s16imm64:$disp),
"#ADDIdtprelL",
[(set i64:$rD,
(PPCaddiDtprelL i64:$reg, tglobaltlsaddr:$disp))]>,
isPPC64;
let PPC970_Unit = 2 in {
let Interpretation64Bit = 1 in {
// Truncating stores.
def STB8 : DForm_1<38, (outs), (ins g8rc:$rS, memri:$src),
"stb $rS, $src", LdStStore,
[(truncstorei8 i64:$rS, iaddr:$src)]>;
def STH8 : DForm_1<44, (outs), (ins g8rc:$rS, memri:$src),
"sth $rS, $src", LdStStore,
[(truncstorei16 i64:$rS, iaddr:$src)]>;
def STW8 : DForm_1<36, (outs), (ins g8rc:$rS, memri:$src),
"stw $rS, $src", LdStStore,
[(truncstorei32 i64:$rS, iaddr:$src)]>;
def STBX8 : XForm_8<31, 215, (outs), (ins g8rc:$rS, memrr:$dst),
"stbx $rS, $dst", LdStStore,
[(truncstorei8 i64:$rS, xaddr:$dst)]>,
PPC970_DGroup_Cracked;
def STHX8 : XForm_8<31, 407, (outs), (ins g8rc:$rS, memrr:$dst),
"sthx $rS, $dst", LdStStore,
[(truncstorei16 i64:$rS, xaddr:$dst)]>,
PPC970_DGroup_Cracked;
def STWX8 : XForm_8<31, 151, (outs), (ins g8rc:$rS, memrr:$dst),
"stwx $rS, $dst", LdStStore,
[(truncstorei32 i64:$rS, xaddr:$dst)]>,
PPC970_DGroup_Cracked;
} // Interpretation64Bit
// Normal 8-byte stores.
def STD : DSForm_1<62, 0, (outs), (ins g8rc:$rS, memrix:$dst),
"std $rS, $dst", LdStSTD,
[(aligned4store i64:$rS, ixaddr:$dst)]>, isPPC64;
def STDX : XForm_8<31, 149, (outs), (ins g8rc:$rS, memrr:$dst),
"stdx $rS, $dst", LdStSTD,
[(store i64:$rS, xaddr:$dst)]>, isPPC64,
PPC970_DGroup_Cracked;
def STDBRX: XForm_8<31, 660, (outs), (ins g8rc:$rS, memrr:$dst),
"stdbrx $rS, $dst", LdStStore,
[(PPCstbrx i64:$rS, xoaddr:$dst, i64)]>, isPPC64,
PPC970_DGroup_Cracked;
}
// Stores with Update (pre-inc).
let PPC970_Unit = 2, mayStore = 1 in {
let Interpretation64Bit = 1 in {
def STBU8 : DForm_1<39, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memri:$dst),
"stbu $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
def STHU8 : DForm_1<45, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memri:$dst),
"sthu $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
def STWU8 : DForm_1<37, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memri:$dst),
"stwu $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
def STDU : DSForm_1<62, 1, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memrix:$dst),
"stdu $rS, $dst", LdStSTDU, []>,
RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">,
isPPC64;
def STBUX8: XForm_8<31, 247, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memrr:$dst),
"stbux $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
PPC970_DGroup_Cracked;
def STHUX8: XForm_8<31, 439, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memrr:$dst),
"sthux $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
PPC970_DGroup_Cracked;
def STWUX8: XForm_8<31, 183, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memrr:$dst),
"stwux $rS, $dst", LdStStoreUpd, []>,
RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
PPC970_DGroup_Cracked;
} // Interpretation64Bit
def STDUX : XForm_8<31, 181, (outs ptr_rc_nor0:$ea_res), (ins g8rc:$rS, memrr:$dst),
"stdux $rS, $dst", LdStSTDU, []>,
RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
PPC970_DGroup_Cracked, isPPC64;
}
// Patterns to match the pre-inc stores. We can't put the patterns on
// the instruction definitions directly as ISel wants the address base
// and offset to be separate operands, not a single complex operand.
def : Pat<(pre_truncsti8 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
(STBU8 $rS, iaddroff:$ptroff, $ptrreg)>;
def : Pat<(pre_truncsti16 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
(STHU8 $rS, iaddroff:$ptroff, $ptrreg)>;
def : Pat<(pre_truncsti32 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
(STWU8 $rS, iaddroff:$ptroff, $ptrreg)>;
def : Pat<(aligned4pre_store i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
(STDU $rS, iaddroff:$ptroff, $ptrreg)>;
def : Pat<(pre_truncsti8 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
(STBUX8 $rS, $ptrreg, $ptroff)>;
def : Pat<(pre_truncsti16 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
(STHUX8 $rS, $ptrreg, $ptroff)>;
def : Pat<(pre_truncsti32 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
(STWUX8 $rS, $ptrreg, $ptroff)>;
def : Pat<(pre_store i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
(STDUX $rS, $ptrreg, $ptroff)>;
//===----------------------------------------------------------------------===//
// Floating point instructions.
//
let PPC970_Unit = 3, neverHasSideEffects = 1,
Uses = [RM] in { // FPU Operations.
defm FCFID : XForm_26r<63, 846, (outs f8rc:$frD), (ins f8rc:$frB),
"fcfid", "$frD, $frB", FPGeneral,
[(set f64:$frD, (PPCfcfid f64:$frB))]>, isPPC64;
defm FCTIDZ : XForm_26r<63, 815, (outs f8rc:$frD), (ins f8rc:$frB),
"fctidz", "$frD, $frB", FPGeneral,
[(set f64:$frD, (PPCfctidz f64:$frB))]>, isPPC64;
defm FCFIDU : XForm_26r<63, 974, (outs f8rc:$frD), (ins f8rc:$frB),
"fcfidu", "$frD, $frB", FPGeneral,
[(set f64:$frD, (PPCfcfidu f64:$frB))]>, isPPC64;
defm FCFIDS : XForm_26r<59, 846, (outs f4rc:$frD), (ins f8rc:$frB),
"fcfids", "$frD, $frB", FPGeneral,
[(set f32:$frD, (PPCfcfids f64:$frB))]>, isPPC64;
defm FCFIDUS : XForm_26r<59, 974, (outs f4rc:$frD), (ins f8rc:$frB),
"fcfidus", "$frD, $frB", FPGeneral,
[(set f32:$frD, (PPCfcfidus f64:$frB))]>, isPPC64;
defm FCTIDUZ : XForm_26r<63, 943, (outs f8rc:$frD), (ins f8rc:$frB),
"fctiduz", "$frD, $frB", FPGeneral,
[(set f64:$frD, (PPCfctiduz f64:$frB))]>, isPPC64;
defm FCTIWUZ : XForm_26r<63, 143, (outs f8rc:$frD), (ins f8rc:$frB),
"fctiwuz", "$frD, $frB", FPGeneral,
[(set f64:$frD, (PPCfctiwuz f64:$frB))]>, isPPC64;
}
//===----------------------------------------------------------------------===//
// Instruction Patterns
//
// Extensions and truncates to/from 32-bit regs.
def : Pat<(i64 (zext i32:$in)),
(RLDICL (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $in, sub_32),
0, 32)>;
def : Pat<(i64 (anyext i32:$in)),
(INSERT_SUBREG (i64 (IMPLICIT_DEF)), $in, sub_32)>;
def : Pat<(i32 (trunc i64:$in)),
(EXTRACT_SUBREG $in, sub_32)>;
// Extending loads with i64 targets.
def : Pat<(zextloadi1 iaddr:$src),
(LBZ8 iaddr:$src)>;
def : Pat<(zextloadi1 xaddr:$src),
(LBZX8 xaddr:$src)>;
def : Pat<(extloadi1 iaddr:$src),
(LBZ8 iaddr:$src)>;
def : Pat<(extloadi1 xaddr:$src),
(LBZX8 xaddr:$src)>;
def : Pat<(extloadi8 iaddr:$src),
(LBZ8 iaddr:$src)>;
def : Pat<(extloadi8 xaddr:$src),
(LBZX8 xaddr:$src)>;
def : Pat<(extloadi16 iaddr:$src),
(LHZ8 iaddr:$src)>;
def : Pat<(extloadi16 xaddr:$src),
(LHZX8 xaddr:$src)>;
def : Pat<(extloadi32 iaddr:$src),
(LWZ8 iaddr:$src)>;
def : Pat<(extloadi32 xaddr:$src),
(LWZX8 xaddr:$src)>;
// Standard shifts. These are represented separately from the real shifts above
// so that we can distinguish between shifts that allow 6-bit and 7-bit shift
// amounts.
def : Pat<(sra i64:$rS, i32:$rB),
(SRAD $rS, $rB)>;
def : Pat<(srl i64:$rS, i32:$rB),
(SRD $rS, $rB)>;
def : Pat<(shl i64:$rS, i32:$rB),
(SLD $rS, $rB)>;
// SHL/SRL
def : Pat<(shl i64:$in, (i32 imm:$imm)),
(RLDICR $in, imm:$imm, (SHL64 imm:$imm))>;
def : Pat<(srl i64:$in, (i32 imm:$imm)),
(RLDICL $in, (SRL64 imm:$imm), imm:$imm)>;
// ROTL
def : Pat<(rotl i64:$in, i32:$sh),
(RLDCL $in, $sh, 0)>;
def : Pat<(rotl i64:$in, (i32 imm:$imm)),
(RLDICL $in, imm:$imm, 0)>;
// Hi and Lo for Darwin Global Addresses.
def : Pat<(PPChi tglobaladdr:$in, 0), (LIS8 tglobaladdr:$in)>;
def : Pat<(PPClo tglobaladdr:$in, 0), (LI8 tglobaladdr:$in)>;
def : Pat<(PPChi tconstpool:$in , 0), (LIS8 tconstpool:$in)>;
def : Pat<(PPClo tconstpool:$in , 0), (LI8 tconstpool:$in)>;
def : Pat<(PPChi tjumptable:$in , 0), (LIS8 tjumptable:$in)>;
def : Pat<(PPClo tjumptable:$in , 0), (LI8 tjumptable:$in)>;
def : Pat<(PPChi tblockaddress:$in, 0), (LIS8 tblockaddress:$in)>;
def : Pat<(PPClo tblockaddress:$in, 0), (LI8 tblockaddress:$in)>;
def : Pat<(PPChi tglobaltlsaddr:$g, i64:$in),
(ADDIS8 $in, tglobaltlsaddr:$g)>;
def : Pat<(PPClo tglobaltlsaddr:$g, i64:$in),
(ADDI8 $in, tglobaltlsaddr:$g)>;
def : Pat<(add i64:$in, (PPChi tglobaladdr:$g, 0)),
(ADDIS8 $in, tglobaladdr:$g)>;
def : Pat<(add i64:$in, (PPChi tconstpool:$g, 0)),
(ADDIS8 $in, tconstpool:$g)>;
def : Pat<(add i64:$in, (PPChi tjumptable:$g, 0)),
(ADDIS8 $in, tjumptable:$g)>;
def : Pat<(add i64:$in, (PPChi tblockaddress:$g, 0)),
(ADDIS8 $in, tblockaddress:$g)>;
// Patterns to match r+r indexed loads and stores for
// addresses without at least 4-byte alignment.
def : Pat<(i64 (unaligned4sextloadi32 xoaddr:$src)),
(LWAX xoaddr:$src)>;
def : Pat<(i64 (unaligned4load xoaddr:$src)),
(LDX xoaddr:$src)>;
def : Pat<(unaligned4store i64:$rS, xoaddr:$dst),
(STDX $rS, xoaddr:$dst)>;