def A4_ext_g : T_Immext<globaladdress>;
}
+def BITPOS32 : SDNodeXForm<imm, [{
+ // Return the bit position we will set [0-31].
+ // As an SDNode.
+ int32_t imm = N->getSExtValue();
+ return XformMskToBitPosU5Imm(imm);
+}]>;
+
// Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
let Inst{4-0} = Rd;
}
-// Add and accumulate.
-// Rd=add(Rs,add(Ru,#s6))
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
-validSubTargets = HasV4SubT in
-def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
- "$dst = add($src1, add($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
- s6_16ExtPred:$src3)))]>,
- Requires<[HasV4T]>;
-
-// Rd=add(Rs,sub(#s6,Ru))
-let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
-validSubTargets = HasV4SubT in
-def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
- (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
- "$dst = add($src1, sub(#$src2, $src3))",
- [(set (i32 IntRegs:$dst),
- (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
- (i32 IntRegs:$src3))))]>,
- Requires<[HasV4T]>;
-
-// Generates the same instruction as ADDr_SUBri_V4 but matches different
-// pattern.
-// Rd=add(Rs,sub(#s6,Ru))
-let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
-validSubTargets = HasV4SubT in
-def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
- (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
- "$dst = add($src1, sub(#$src2, $src3))",
- [(set (i32 IntRegs:$dst),
- (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
- (i32 IntRegs:$src3)))]>,
- Requires<[HasV4T]>;
-
-
-// Add or subtract doublewords with carry.
-//TODO:
-// Rdd=add(Rss,Rtt,Px):carry
-//TODO:
-// Rdd=sub(Rss,Rtt,Px):carry
-
-
-// Logical doublewords.
-// Rdd=and(Rtt,~Rss)
-let validSubTargets = HasV4SubT in
-def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2),
- "$dst = and($src1, ~$src2)",
- [(set (i64 DoubleRegs:$dst), (and (i64 DoubleRegs:$src1),
- (not (i64 DoubleRegs:$src2))))]>,
- Requires<[HasV4T]>;
-
-// Rdd=or(Rtt,~Rss)
-let validSubTargets = HasV4SubT in
-def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2),
- "$dst = or($src1, ~$src2)",
- [(set (i64 DoubleRegs:$dst),
- (or (i64 DoubleRegs:$src1), (not (i64 DoubleRegs:$src2))))]>,
- Requires<[HasV4T]>;
-
-
-// Logical-logical doublewords.
-// Rxx^=xor(Rss,Rtt)
-let validSubTargets = HasV4SubT in
-def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
- "$dst ^= xor($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (xor (i64 DoubleRegs:$src1), (xor (i64 DoubleRegs:$src2),
- (i64 DoubleRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+let isCodeGenOnly = 0 in {
+// Rx[&|]=xor(Rs,Rt)
+def M4_or_xor : T_MType_acc_rr < "|= xor", 0b110, 0b001, 0>;
+def M4_and_xor : T_MType_acc_rr < "&= xor", 0b010, 0b010, 0>;
+// Rx[&|^]=or(Rs,Rt)
+def M4_xor_or : T_MType_acc_rr < "^= or", 0b110, 0b011, 0>;
-// Logical-logical words.
-// Rx=or(Ru,and(Rx,#s10))
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
-validSubTargets = HasV4SubT in
-def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
- "$dst = or($src1, and($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- s10ExtPred:$src3)))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+let CextOpcode = "ORr_ORr" in
+def M4_or_or : T_MType_acc_rr < "|= or", 0b110, 0b000, 0>;
+def M4_and_or : T_MType_acc_rr < "&= or", 0b010, 0b001, 0>;
// Rx[&|^]=and(Rs,Rt)
-// Rx&=and(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst &= and($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+def M4_xor_and : T_MType_acc_rr < "^= and", 0b110, 0b010, 0>;
-// Rx|=and(Rs,Rt)
-let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
-def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst |= and($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>, ImmRegRel;
-
-// Rx^=and(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst ^= and($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+let CextOpcode = "ORr_ANDr" in
+def M4_or_and : T_MType_acc_rr < "|= and", 0b010, 0b011, 0>;
+def M4_and_and : T_MType_acc_rr < "&= and", 0b010, 0b000, 0>;
// Rx[&|^]=and(Rs,~Rt)
-// Rx&=and(Rs,~Rt)
-let validSubTargets = HasV4SubT in
-def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst &= and($src2, ~$src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (not (i32 IntRegs:$src3)))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-
-// Rx|=and(Rs,~Rt)
-let validSubTargets = HasV4SubT in
-def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst |= and($src2, ~$src3)",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (not (i32 IntRegs:$src3)))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-
-// Rx^=and(Rs,~Rt)
-let validSubTargets = HasV4SubT in
-def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst ^= and($src2, ~$src3)",
- [(set (i32 IntRegs:$dst),
- (xor (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- (not (i32 IntRegs:$src3)))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-
-// Rx[&|^]=or(Rs,Rt)
-// Rx&=or(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst &= or($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-
-// Rx|=or(Rs,Rt)
-let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
-def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst |= or($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>, ImmRegRel;
-
-// Rx^=or(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst ^= or($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (xor (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-
-// Rx[&|^]=xor(Rs,Rt)
-// Rx&=xor(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst &= xor($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+def M4_xor_andn : T_MType_acc_rr < "^= and", 0b001, 0b010, 0, [], 1>;
+def M4_or_andn : T_MType_acc_rr < "|= and", 0b001, 0b000, 0, [], 1>;
+def M4_and_andn : T_MType_acc_rr < "&= and", 0b001, 0b001, 0, [], 1>;
+}
-// Rx|=xor(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst |= xor($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+// Compound or-or and or-and
+let isExtentSigned = 1, InputType = "imm", hasNewValue = 1, isExtendable = 1,
+ opExtentBits = 10, opExtendable = 3 in
+class T_CompOR <string mnemonic, bits<2> MajOp, SDNode OpNode>
+ : MInst_acc <(outs IntRegs:$Rx),
+ (ins IntRegs:$src1, IntRegs:$Rs, s10Ext:$s10),
+ "$Rx |= "#mnemonic#"($Rs, #$s10)",
+ [(set (i32 IntRegs:$Rx), (or (i32 IntRegs:$src1),
+ (OpNode (i32 IntRegs:$Rs), s10ExtPred:$s10)))],
+ "$src1 = $Rx", ALU64_tc_2_SLOT23>, ImmRegRel {
+ bits<5> Rx;
+ bits<5> Rs;
+ bits<10> s10;
-// Rx^=xor(Rs,Rt)
-let validSubTargets = HasV4SubT in
-def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
- "$dst ^= xor($src2, $src3)",
- [(set (i32 IntRegs:$dst),
- (and (i32 IntRegs:$src1), (xor (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+ let IClass = 0b1101;
-// Rx|=and(Rs,#s10)
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
-validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
-def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
- "$dst |= and($src2, #$src3)",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- s10ExtPred:$src3)))],
- "$src1 = $dst">,
- Requires<[HasV4T]>, ImmRegRel;
+ let Inst{27-24} = 0b1010;
+ let Inst{23-22} = MajOp;
+ let Inst{20-16} = Rs;
+ let Inst{21} = s10{9};
+ let Inst{13-5} = s10{8-0};
+ let Inst{4-0} = Rx;
+ }
-// Rx|=or(Rs,#s10)
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
-validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
-def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
- "$dst |= or($src2, #$src3)",
- [(set (i32 IntRegs:$dst),
- (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
- s10ExtPred:$src3)))],
- "$src1 = $dst">,
- Requires<[HasV4T]>, ImmRegRel;
+let CextOpcode = "ORr_ANDr", isCodeGenOnly = 0 in
+def S4_or_andi : T_CompOR <"and", 0b00, and>;
+let CextOpcode = "ORr_ORr", isCodeGenOnly = 0 in
+def S4_or_ori : T_CompOR <"or", 0b10, or>;
// Modulo wrap
// Rd=modwrap(Rs,Rt)
// XTYPE/ALU -
//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// XTYPE/BIT +
+//===----------------------------------------------------------------------===//
+
+// Bit reverse
+let isCodeGenOnly = 0 in
+def S2_brevp : T_S2op_3 <"brev", 0b11, 0b110>;
+
+// Bit count
+let isCodeGenOnly = 0 in {
+def S2_ct0p : T_COUNT_LEADING_64<"ct0", 0b111, 0b010>;
+def S2_ct1p : T_COUNT_LEADING_64<"ct1", 0b111, 0b100>;
+def S4_clbpnorm : T_COUNT_LEADING_64<"normamt", 0b011, 0b000>;
+}
+
+def: Pat<(i32 (trunc (cttz (i64 DoubleRegs:$Rss)))),
+ (S2_ct0p (i64 DoubleRegs:$Rss))>;
+def: Pat<(i32 (trunc (cttz (not (i64 DoubleRegs:$Rss))))),
+ (S2_ct1p (i64 DoubleRegs:$Rss))>;
+
+let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
+def S4_clbaddi : SInst<(outs IntRegs:$Rd), (ins IntRegs:$Rs, s6Imm:$s6),
+ "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> {
+ bits<5> Rs;
+ bits<5> Rd;
+ bits<6> s6;
+ let IClass = 0b1000;
+ let Inst{27-24} = 0b1100;
+ let Inst{23-21} = 0b001;
+ let Inst{20-16} = Rs;
+ let Inst{13-8} = s6;
+ let Inst{7-5} = 0b000;
+ let Inst{4-0} = Rd;
+}
+
+let hasSideEffects = 0, hasNewValue = 1, isCodeGenOnly = 0 in
+def S4_clbpaddi : SInst<(outs IntRegs:$Rd), (ins DoubleRegs:$Rs, s6Imm:$s6),
+ "$Rd = add(clb($Rs), #$s6)", [], "", S_2op_tc_2_SLOT23> {
+ bits<5> Rs;
+ bits<5> Rd;
+ bits<6> s6;
+ let IClass = 0b1000;
+ let Inst{27-24} = 0b1000;
+ let Inst{23-21} = 0b011;
+ let Inst{20-16} = Rs;
+ let Inst{13-8} = s6;
+ let Inst{7-5} = 0b010;
+ let Inst{4-0} = Rd;
+}
+
+
+// Bit test/set/clear
+let isCodeGenOnly = 0 in {
+def S4_ntstbit_i : T_TEST_BIT_IMM<"!tstbit", 0b001>;
+def S4_ntstbit_r : T_TEST_BIT_REG<"!tstbit", 1>;
+}
+
+let AddedComplexity = 20 in { // Complexity greater than cmp reg-imm.
+ def: Pat<(i1 (seteq (and (shl 1, u5ImmPred:$u5), (i32 IntRegs:$Rs)), 0)),
+ (S4_ntstbit_i (i32 IntRegs:$Rs), u5ImmPred:$u5)>;
+ def: Pat<(i1 (seteq (and (shl 1, (i32 IntRegs:$Rt)), (i32 IntRegs:$Rs)), 0)),
+ (S4_ntstbit_r (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))>;
+}
+
+// Add extra complexity to prefer these instructions over bitsset/bitsclr.
+// The reason is that tstbit/ntstbit can be folded into a compound instruction:
+// if ([!]tstbit(...)) jump ...
+let AddedComplexity = 100 in
+def: Pat<(i1 (setne (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))),
+ (S2_tstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>;
+
+let AddedComplexity = 100 in
+def: Pat<(i1 (seteq (and (i32 IntRegs:$Rs), (i32 Set5ImmPred:$u5)), (i32 0))),
+ (S4_ntstbit_i (i32 IntRegs:$Rs), (BITPOS32 Set5ImmPred:$u5))>;
+
+let isCodeGenOnly = 0 in {
+def C4_nbitsset : T_TEST_BITS_REG<"!bitsset", 0b01, 1>;
+def C4_nbitsclr : T_TEST_BITS_REG<"!bitsclr", 0b10, 1>;
+def C4_nbitsclri : T_TEST_BITS_IMM<"!bitsclr", 0b10, 1>;
+}
+
+// Do not increase complexity of these patterns. In the DAG, "cmp i8" may be
+// represented as a compare against "value & 0xFF", which is an exact match
+// for cmpb (same for cmph). The patterns below do not contain any additional
+// complexity that would make them preferable, and if they were actually used
+// instead of cmpb/cmph, they would result in a compare against register that
+// is loaded with the byte/half mask (i.e. 0xFF or 0xFFFF).
+def: Pat<(i1 (setne (and I32:$Rs, u6ImmPred:$u6), 0)),
+ (C4_nbitsclri I32:$Rs, u6ImmPred:$u6)>;
+def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), 0)),
+ (C4_nbitsclr I32:$Rs, I32:$Rt)>;
+def: Pat<(i1 (setne (and I32:$Rs, I32:$Rt), I32:$Rt)),
+ (C4_nbitsset I32:$Rs, I32:$Rt)>;
+
+//===----------------------------------------------------------------------===//
+// XTYPE/BIT -
+//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// XTYPE/MPY +
//===----------------------------------------------------------------------===//
-// Multiply and user lower result.
-// Rd=add(#u6,mpyi(Rs,#U6))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
-validSubTargets = HasV4SubT in
-def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
- (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
- "$dst = add(#$src1, mpyi($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
- u6ExtPred:$src1))]>,
- Requires<[HasV4T]>;
+// Rd=add(#u6,mpyi(Rs,#U6)) -- Multiply by immed and add immed.
+
+let hasNewValue = 1, isExtendable = 1, opExtentBits = 6, opExtendable = 1,
+ isCodeGenOnly = 0 in
+def M4_mpyri_addi : MInst<(outs IntRegs:$Rd),
+ (ins u6Ext:$u6, IntRegs:$Rs, u6Imm:$U6),
+ "$Rd = add(#$u6, mpyi($Rs, #$U6))" ,
+ [(set (i32 IntRegs:$Rd),
+ (add (mul (i32 IntRegs:$Rs), u6ImmPred:$U6),
+ u6ExtPred:$u6))] ,"",ALU64_tc_3x_SLOT23> {
+ bits<5> Rd;
+ bits<6> u6;
+ bits<5> Rs;
+ bits<6> U6;
+
+ let IClass = 0b1101;
+
+ let Inst{27-24} = 0b1000;
+ let Inst{23} = U6{5};
+ let Inst{22-21} = u6{5-4};
+ let Inst{20-16} = Rs;
+ let Inst{13} = u6{3};
+ let Inst{12-8} = Rd;
+ let Inst{7-5} = u6{2-0};
+ let Inst{4-0} = U6{4-0};
+ }
+
+// Rd=add(#u6,mpyi(Rs,Rt))
+let CextOpcode = "ADD_MPY", InputType = "imm", hasNewValue = 1,
+ isExtendable = 1, opExtentBits = 6, opExtendable = 1, isCodeGenOnly = 0 in
+def M4_mpyrr_addi : MInst <(outs IntRegs:$Rd),
+ (ins u6Ext:$u6, IntRegs:$Rs, IntRegs:$Rt),
+ "$Rd = add(#$u6, mpyi($Rs, $Rt))" ,
+ [(set (i32 IntRegs:$Rd),
+ (add (mul (i32 IntRegs:$Rs), (i32 IntRegs:$Rt)), u6ExtPred:$u6))],
+ "", ALU64_tc_3x_SLOT23>, ImmRegRel {
+ bits<5> Rd;
+ bits<6> u6;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1101;
+
+ let Inst{27-23} = 0b01110;
+ let Inst{22-21} = u6{5-4};
+ let Inst{20-16} = Rs;
+ let Inst{13} = u6{3};
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = u6{2-0};
+ let Inst{4-0} = Rd;
+ }
+
+let hasNewValue = 1 in
+class T_AddMpy <bit MajOp, PatLeaf ImmPred, dag ins>
+ : ALU64Inst <(outs IntRegs:$dst), ins,
+ "$dst = add($src1, mpyi("#!if(MajOp,"$src3, #$src2))",
+ "#$src2, $src3))"),
+ [(set (i32 IntRegs:$dst),
+ (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3), ImmPred:$src2)))],
+ "", ALU64_tc_3x_SLOT23> {
+ bits<5> dst;
+ bits<5> src1;
+ bits<8> src2;
+ bits<5> src3;
+
+ let IClass = 0b1101;
+
+ bits<6> ImmValue = !if(MajOp, src2{5-0}, src2{7-2});
+
+ let Inst{27-24} = 0b1111;
+ let Inst{23} = MajOp;
+ let Inst{22-21} = ImmValue{5-4};
+ let Inst{20-16} = src3;
+ let Inst{13} = ImmValue{3};
+ let Inst{12-8} = dst;
+ let Inst{7-5} = ImmValue{2-0};
+ let Inst{4-0} = src1;
+ }
+
+let isCodeGenOnly = 0 in
+def M4_mpyri_addr_u2 : T_AddMpy<0b0, u6_2ImmPred,
+ (ins IntRegs:$src1, u6_2Imm:$src2, IntRegs:$src3)>;
+
+let isExtendable = 1, opExtentBits = 6, opExtendable = 3,
+ CextOpcode = "ADD_MPY", InputType = "imm", isCodeGenOnly = 0 in
+def M4_mpyri_addr : T_AddMpy<0b1, u6ExtPred,
+ (ins IntRegs:$src1, IntRegs:$src3, u6Ext:$src2)>, ImmRegRel;
+
+// Rx=add(Ru,mpyi(Rx,Rs))
+let validSubTargets = HasV4SubT, CextOpcode = "ADD_MPY", InputType = "reg",
+ hasNewValue = 1, isCodeGenOnly = 0 in
+def M4_mpyrr_addr: MInst_acc <(outs IntRegs:$Rx),
+ (ins IntRegs:$Ru, IntRegs:$_src_, IntRegs:$Rs),
+ "$Rx = add($Ru, mpyi($_src_, $Rs))",
+ [(set (i32 IntRegs:$Rx), (add (i32 IntRegs:$Ru),
+ (mul (i32 IntRegs:$_src_), (i32 IntRegs:$Rs))))],
+ "$_src_ = $Rx", M_tc_3x_SLOT23>, ImmRegRel {
+ bits<5> Rx;
+ bits<5> Ru;
+ bits<5> Rs;
+
+ let IClass = 0b1110;
+
+ let Inst{27-21} = 0b0011000;
+ let Inst{12-8} = Rx;
+ let Inst{4-0} = Ru;
+ let Inst{20-16} = Rs;
+ }
// Rd=add(##,mpyi(Rs,#U6))
def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
(HexagonCONST32 tglobaladdr:$src1)),
- (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
+ (i32 (M4_mpyri_addi tglobaladdr:$src1, IntRegs:$src2,
u6ImmPred:$src3))>;
-// Rd=add(#u6,mpyi(Rs,Rt))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
-validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
-def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
- (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst = add(#$src1, mpyi($src2, $src3))",
- [(set (i32 IntRegs:$dst),
- (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
- u6ExtPred:$src1))]>,
- Requires<[HasV4T]>, ImmRegRel;
-
// Rd=add(##,mpyi(Rs,Rt))
def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
(HexagonCONST32 tglobaladdr:$src1)),
- (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
+ (i32 (M4_mpyrr_addi tglobaladdr:$src1, IntRegs:$src2,
IntRegs:$src3))>;
-// Rd=add(Ru,mpyi(#u6:2,Rs))
-let validSubTargets = HasV4SubT in
-def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
- (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
- "$dst = add($src1, mpyi(#$src2, $src3))",
- [(set (i32 IntRegs:$dst),
- (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src3),
- u6_2ImmPred:$src2)))]>,
- Requires<[HasV4T]>;
-
-// Rd=add(Ru,mpyi(Rs,#u6))
-let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
-validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
-def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
- "$dst = add($src1, mpyi($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
- u6ExtPred:$src3)))]>,
- Requires<[HasV4T]>, ImmRegRel;
-
-// Rx=add(Ru,mpyi(Rx,Rs))
-let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
-def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
- "$dst = add($src1, mpyi($src2, $src3))",
- [(set (i32 IntRegs:$dst),
- (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src2 = $dst">,
- Requires<[HasV4T]>, ImmRegRel;
-
-
// Polynomial multiply words
// Rdd=pmpyw(Rs,Rt)
// Rxx^=pmpyw(Rs,Rt)
//===----------------------------------------------------------------------===//
// XTYPE/SHIFT +
//===----------------------------------------------------------------------===//
-
-// Shift by immediate and accumulate.
-// Rx=add(#u8,asl(Rx,#U5))
+// Shift by immediate and accumulate/logical.
+// Rx=add(#u8,asl(Rx,#U5)) Rx=add(#u8,lsr(Rx,#U5))
+// Rx=sub(#u8,asl(Rx,#U5)) Rx=sub(#u8,lsr(Rx,#U5))
+// Rx=and(#u8,asl(Rx,#U5)) Rx=and(#u8,lsr(Rx,#U5))
+// Rx=or(#u8,asl(Rx,#U5)) Rx=or(#u8,lsr(Rx,#U5))
let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = add(#$src1, asl($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+ hasNewValue = 1, opNewValue = 0, validSubTargets = HasV4SubT in
+class T_S4_ShiftOperate<string MnOp, string MnSh, SDNode Op, SDNode Sh,
+ bit asl_lsr, bits<2> MajOp, InstrItinClass Itin>
+ : MInst_acc<(outs IntRegs:$Rd), (ins u8Ext:$u8, IntRegs:$Rx, u5Imm:$U5),
+ "$Rd = "#MnOp#"(#$u8, "#MnSh#"($Rx, #$U5))",
+ [(set (i32 IntRegs:$Rd),
+ (Op (Sh I32:$Rx, u5ImmPred:$U5), u8ExtPred:$u8))],
+ "$Rd = $Rx", Itin> {
-// Rx=add(#u8,lsr(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = add(#$src1, lsr($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+ bits<5> Rd;
+ bits<8> u8;
+ bits<5> Rx;
+ bits<5> U5;
-// Rx=sub(#u8,asl(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = sub(#$src1, asl($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+ let IClass = 0b1101;
+ let Inst{27-24} = 0b1110;
+ let Inst{23-21} = u8{7-5};
+ let Inst{20-16} = Rd;
+ let Inst{13} = u8{4};
+ let Inst{12-8} = U5;
+ let Inst{7-5} = u8{3-1};
+ let Inst{4} = asl_lsr;
+ let Inst{3} = u8{0};
+ let Inst{2-1} = MajOp;
+}
-// Rx=sub(#u8,lsr(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = sub(#$src1, lsr($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+multiclass T_ShiftOperate<string mnemonic, SDNode Op, bits<2> MajOp,
+ InstrItinClass Itin> {
+ def _asl_ri : T_S4_ShiftOperate<mnemonic, "asl", Op, shl, 0, MajOp, Itin>;
+ def _lsr_ri : T_S4_ShiftOperate<mnemonic, "lsr", Op, srl, 1, MajOp, Itin>;
+}
+let AddedComplexity = 200, isCodeGenOnly = 0 in {
+ defm S4_addi : T_ShiftOperate<"add", add, 0b10, ALU64_tc_2_SLOT23>;
+ defm S4_andi : T_ShiftOperate<"and", and, 0b00, ALU64_tc_2_SLOT23>;
+}
-//Shift by immediate and logical.
-//Rx=and(#u8,asl(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = and(#$src1, asl($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+let AddedComplexity = 30, isCodeGenOnly = 0 in
+defm S4_ori : T_ShiftOperate<"or", or, 0b01, ALU64_tc_1_SLOT23>;
-//Rx=and(#u8,lsr(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-validSubTargets = HasV4SubT in
-def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = and(#$src1, lsr($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+let isCodeGenOnly = 0 in
+defm S4_subi : T_ShiftOperate<"sub", sub, 0b11, ALU64_tc_1_SLOT23>;
-//Rx=or(#u8,asl(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-AddedComplexity = 30, validSubTargets = HasV4SubT in
-def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = or(#$src1, asl($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
-//Rx=or(#u8,lsr(Rx,#U5))
-let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
-AddedComplexity = 30, validSubTargets = HasV4SubT in
-def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
- (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
- "$dst = or(#$src1, lsr($src2, #$src3))",
- [(set (i32 IntRegs:$dst),
- (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
- u8ExtPred:$src1))],
- "$src2 = $dst">,
- Requires<[HasV4T]>;
+// Rd=[cround|round](Rs,Rt)
+let hasNewValue = 1, Itinerary = S_3op_tc_2_SLOT23, isCodeGenOnly = 0 in {
+ def A4_cround_rr : T_S3op_3 < "cround", IntRegs, 0b11, 0b00>;
+ def A4_round_rr : T_S3op_3 < "round", IntRegs, 0b11, 0b10>;
+}
+// Rd=round(Rs,Rt):sat
+let hasNewValue = 1, Defs = [USR_OVF], Itinerary = S_3op_tc_2_SLOT23,
+ isCodeGenOnly = 0 in
+def A4_round_rr_sat : T_S3op_3 < "round", IntRegs, 0b11, 0b11, 1>;
+
+// Rdd=[add|sub](Rss,Rtt,Px):carry
+let isPredicateLate = 1, hasSideEffects = 0 in
+class T_S3op_carry <string mnemonic, bits<3> MajOp>
+ : SInst < (outs DoubleRegs:$Rdd, PredRegs:$Px),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt, PredRegs:$Pu),
+ "$Rdd = "#mnemonic#"($Rss, $Rtt, $Pu):carry",
+ [], "$Px = $Pu", S_3op_tc_1_SLOT23 > {
+ bits<5> Rdd;
+ bits<5> Rss;
+ bits<5> Rtt;
+ bits<2> Pu;
-//Shift by register.
-//Rd=lsl(#s6,Rt)
-let validSubTargets = HasV4SubT in {
-def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
- "$dst = lsl(#$src1, $src2)",
- [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
- (i32 IntRegs:$src2)))]>,
- Requires<[HasV4T]>;
+ let IClass = 0b1100;
+ let Inst{27-24} = 0b0010;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ let Inst{6-5} = Pu;
+ let Inst{4-0} = Rdd;
+ }
-//Shift by register and logical.
-//Rxx^=asl(Rss,Rt)
-def ASLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
- "$dst ^= asl($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (xor (i64 DoubleRegs:$src1), (shl (i64 DoubleRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+let isCodeGenOnly = 0 in {
+def A4_addp_c : T_S3op_carry < "add", 0b110 >;
+def A4_subp_c : T_S3op_carry < "sub", 0b111 >;
+}
-//Rxx^=asr(Rss,Rt)
-def ASRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
- "$dst ^= asr($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (xor (i64 DoubleRegs:$src1), (sra (i64 DoubleRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+// Shift an immediate left by register amount.
+let hasNewValue = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
+def S4_lsli: SInst <(outs IntRegs:$Rd), (ins s6Imm:$s6, IntRegs:$Rt),
+ "$Rd = lsl(#$s6, $Rt)" ,
+ [(set (i32 IntRegs:$Rd), (shl s6ImmPred:$s6,
+ (i32 IntRegs:$Rt)))],
+ "", S_3op_tc_1_SLOT23> {
+ bits<5> Rd;
+ bits<6> s6;
+ bits<5> Rt;
-//Rxx^=lsl(Rss,Rt)
-def LSLd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
- "$dst ^= lsl($src2, $src3)",
- [(set (i64 DoubleRegs:$dst), (xor (i64 DoubleRegs:$src1),
- (shl (i64 DoubleRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
+ let IClass = 0b1100;
-//Rxx^=lsr(Rss,Rt)
-def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
- (ins DoubleRegs:$src1, DoubleRegs:$src2, IntRegs:$src3),
- "$dst ^= lsr($src2, $src3)",
- [(set (i64 DoubleRegs:$dst),
- (xor (i64 DoubleRegs:$src1), (srl (i64 DoubleRegs:$src2),
- (i32 IntRegs:$src3))))],
- "$src1 = $dst">,
- Requires<[HasV4T]>;
-}
+ let Inst{27-22} = 0b011010;
+ let Inst{20-16} = s6{5-1};
+ let Inst{12-8} = Rt;
+ let Inst{7-6} = 0b11;
+ let Inst{4-0} = Rd;
+ let Inst{5} = s6{0};
+ }
//===----------------------------------------------------------------------===//
// XTYPE/SHIFT -
(ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
opc#"($base+#$offset)"#memOp#"$delta",
[]>,
- Requires<[HasV4T, UseMEMOP]> {
+ Requires<[UseMEMOP]> {
bits<5> base;
bits<5> delta;
!if (!eq(opcBits, 0b01), offset{6-1},
!if (!eq(opcBits, 0b10), offset{7-2},0)));
+ let opExtentAlign = opcBits;
let IClass = 0b0011;
let Inst{27-24} = 0b1110;
let Inst{22-21} = opcBits;
opc#"($base+#$offset)"#memOp#"#$delta"
#!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
[]>,
- Requires<[HasV4T, UseMEMOP]> {
+ Requires<[UseMEMOP]> {
bits<5> base;
bits<5> delta;
!if (!eq(opcBits, 0b01), offset{6-1},
!if (!eq(opcBits, 0b10), offset{7-2},0)));
+ let opExtentAlign = opcBits;
let IClass = 0b0011;
let Inst{27-24} = 0b1111;
let Inst{22-21} = opcBits;
// multiclass to define MemOp instructions with register operand.
multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
- def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
- def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
- def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
- def _OR#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
+ def L4_add#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
+ def L4_sub#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
+ def L4_and#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
+ def L4_or#NAME : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
}
// multiclass to define MemOp instructions with immediate Operand.
multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
- def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
- def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
- def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
- def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
+ def L4_iadd#NAME : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
+ def L4_isub#NAME : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
+ def L4_iand#NAME : MemOp_ri_base<opc, opcBits, ImmOp, " = clrbit(", 0b10>;
+ def L4_ior#NAME : MemOp_ri_base<opc, opcBits, ImmOp, " = setbit(", 0b11>;
}
multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
- defm r : MemOp_rr <opc, opcBits, ImmOp>;
- defm i : MemOp_ri <opc, opcBits, ImmOp>;
+ defm _#NAME : MemOp_rr <opc, opcBits, ImmOp>;
+ defm _#NAME : MemOp_ri <opc, opcBits, ImmOp>;
}
// Define MemOp instructions.
let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
-validSubTargets =HasV4SubT in {
- let opExtentBits = 6, accessSize = ByteAccess in
- defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
+ validSubTargets =HasV4SubT in {
+ let opExtentBits = 6, accessSize = ByteAccess, isCodeGenOnly = 0 in
+ defm memopb_io : MemOp_base <"memb", 0b00, u6_0Ext>;
- let opExtentBits = 7, accessSize = HalfWordAccess in
- defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
+ let opExtentBits = 7, accessSize = HalfWordAccess, isCodeGenOnly = 0 in
+ defm memoph_io : MemOp_base <"memh", 0b01, u6_1Ext>;
- let opExtentBits = 8, accessSize = WordAccess in
- defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
+ let opExtentBits = 8, accessSize = WordAccess, isCodeGenOnly = 0 in
+ defm memopw_io : MemOp_base <"memw", 0b10, u6_2Ext>;
}
//===----------------------------------------------------------------------===//
multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
// Half Word
defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
- MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
+ L4_iadd_memoph_io, L4_isub_memoph_io>;
// Byte
defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
- MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
+ L4_iadd_memopb_io, L4_isub_memopb_io>;
}
let Predicates = [HasV4T, UseMEMOP] in {
defm : MemOpi_u5ExtType<extloadi8, extloadi16>; // any extend
// Word
- defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
- MemOPw_SUBi_V4>;
+ defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, L4_iadd_memopw_io,
+ L4_isub_memopw_io>;
}
//===----------------------------------------------------------------------===//
multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
// Half Word
defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
- ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
+ ADDRriU6_1, MEMOPIMM_HALF, L4_isub_memoph_io>;
// Byte
defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
- ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
+ ADDRriU6_0, MEMOPIMM_BYTE, L4_isub_memopb_io>;
}
let Predicates = [HasV4T, UseMEMOP] in {
// Word
defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
- ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
+ ADDRriU6_2, MEMOPIMM, L4_isub_memopw_io>;
}
//===----------------------------------------------------------------------===//
multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
// Byte - clrbit
defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
- ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
+ ADDRriU6_0, CLRMEMIMM_BYTE, L4_iand_memopb_io, and>;
// Byte - setbit
defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred, u6ExtPred,
- ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
+ ADDRriU6_0, SETMEMIMM_BYTE, L4_ior_memopb_io, or>;
// Half Word - clrbit
defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
- ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
+ ADDRriU6_1, CLRMEMIMM_SHORT, L4_iand_memoph_io, and>;
// Half Word - setbit
defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
- ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
+ ADDRriU6_1, SETMEMIMM_SHORT, L4_ior_memoph_io, or>;
}
let Predicates = [HasV4T, UseMEMOP] in {
// memw(Rs+#0) = [clrbit|setbit](#U5)
// memw(Rs+#u6:2) = [clrbit|setbit](#U5)
defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
- CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
+ CLRMEMIMM, L4_iand_memopw_io, and>;
defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
- SETMEMIMM, MemOPw_SETBITi_V4, or>;
+ SETMEMIMM, L4_ior_memopw_io, or>;
}
//===----------------------------------------------------------------------===//
multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
// Half Word
defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
- MemOPh_ADDr_V4, MemOPh_SUBr_V4,
- MemOPh_ANDr_V4, MemOPh_ORr_V4>;
+ L4_add_memoph_io, L4_sub_memoph_io,
+ L4_and_memoph_io, L4_or_memoph_io>;
// Byte
defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
- MemOPb_ADDr_V4, MemOPb_SUBr_V4,
- MemOPb_ANDr_V4, MemOPb_ORr_V4>;
+ L4_add_memopb_io, L4_sub_memopb_io,
+ L4_and_memopb_io, L4_or_memopb_io>;
}
// Define 'def Pats' for MemOps with register addend.
defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
defm : MemOPr_ExtType<extloadi8, extloadi16>; // any extend
// Word
- defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
- MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
+ defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, L4_add_memopw_io,
+ L4_sub_memopw_io, L4_and_memopw_io, L4_or_memopw_io >;
}
//===----------------------------------------------------------------------===//
// XTYPE/PRED -
//===----------------------------------------------------------------------===//
-//Deallocate frame and return.
-// dealloc_return
-let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicable = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_V4 : LD0Inst<(outs), (ins),
- "dealloc_return",
- []>,
- Requires<[HasV4T]>;
+//===----------------------------------------------------------------------===//
+// Multiclass for DeallocReturn
+//===----------------------------------------------------------------------===//
+class L4_RETURN<string mnemonic, bit isNot, bit isPredNew, bit isTak>
+ : LD0Inst<(outs), (ins PredRegs:$src),
+ !if(isNot, "if (!$src", "if ($src")#
+ !if(isPredNew, ".new) ", ") ")#mnemonic#
+ !if(isPredNew, #!if(isTak,":t", ":nt"),""),
+ [], "", LD_tc_3or4stall_SLOT0> {
+
+ bits<2> src;
+ let BaseOpcode = "L4_RETURN";
+ let isPredicatedFalse = isNot;
+ let isPredicatedNew = isPredNew;
+ let isTaken = isTak;
+ let IClass = 0b1001;
+
+ let Inst{27-16} = 0b011000011110;
+
+ let Inst{13} = isNot;
+ let Inst{12} = isTak;
+ let Inst{11} = isPredNew;
+ let Inst{10} = 0b0;
+ let Inst{9-8} = src;
+ let Inst{4-0} = 0b11110;
+ }
+
+// Produce all predicated forms, p, !p, p.new, !p.new, :t, :nt
+multiclass L4_RETURN_PRED<string mnemonic, bit PredNot> {
+ let isPredicated = 1 in {
+ def _#NAME# : L4_RETURN <mnemonic, PredNot, 0, 1>;
+ def _#NAME#new_pnt : L4_RETURN <mnemonic, PredNot, 1, 0>;
+ def _#NAME#new_pt : L4_RETURN <mnemonic, PredNot, 1, 1>;
+ }
+}
+
+multiclass LD_MISC_L4_RETURN<string mnemonic> {
+ let isBarrier = 1, isPredicable = 1 in
+ def NAME : LD0Inst <(outs), (ins), mnemonic, [], "",
+ LD_tc_3or4stall_SLOT0> {
+ let BaseOpcode = "L4_RETURN";
+ let IClass = 0b1001;
+ let Inst{27-16} = 0b011000011110;
+ let Inst{13-10} = 0b0000;
+ let Inst{4-0} = 0b11110;
+ }
+ defm t : L4_RETURN_PRED<mnemonic, 0 >;
+ defm f : L4_RETURN_PRED<mnemonic, 1 >;
}
+let isReturn = 1, isTerminator = 1,
+ Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
+ validSubTargets = HasV4SubT, isCodeGenOnly = 0 in
+defm L4_return: LD_MISC_L4_RETURN <"dealloc_return">, PredNewRel;
+
// Restore registers and dealloc return function call.
let isCall = 1, isBarrier = 1, isReturn = 1, isTerminator = 1,
Defs = [R29, R30, R31, PC] in {
Requires<[HasV4T]>;
}
-// if (Ps) dealloc_return
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cPt_V4 : LD0Inst<(outs),
- (ins PredRegs:$src1),
- "if ($src1) dealloc_return",
- []>,
- Requires<[HasV4T]>;
-}
+//===----------------------------------------------------------------------===//
+// Template class for non predicated store instructions with
+// GP-Relative or absolute addressing.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, isPredicable = 1, isNVStorable = 1 in
+class T_StoreAbsGP <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<2>MajOp, Operand AddrOp, bit isAbs, bit isHalf>
+ : STInst<(outs), (ins AddrOp:$addr, RC:$src),
+ mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src"#!if(isHalf, ".h",""),
+ [], "", V2LDST_tc_st_SLOT01> {
+ bits<19> addr;
+ bits<5> src;
+ bits<16> offsetBits;
-// if (!Ps) dealloc_return
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1, isPredicatedFalse = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cNotPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
- "if (!$src1) dealloc_return",
- []>,
- Requires<[HasV4T]>;
-}
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
+ !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
+ !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
+ /* u16_0Imm */ addr{15-0})));
+ let IClass = 0b0100;
+ let Inst{27} = 1;
+ let Inst{26-25} = offsetBits{15-14};
+ let Inst{24} = 0b0;
+ let Inst{23-22} = MajOp;
+ let Inst{21} = isHalf;
+ let Inst{20-16} = offsetBits{13-9};
+ let Inst{13} = offsetBits{8};
+ let Inst{12-8} = src;
+ let Inst{7-0} = offsetBits{7-0};
+ }
-// if (Ps.new) dealloc_return:nt
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
- "if ($src1.new) dealloc_return:nt",
- []>,
- Requires<[HasV4T]>;
-}
+//===----------------------------------------------------------------------===//
+// Template class for predicated store instructions with
+// GP-Relative or absolute addressing.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, isPredicated = 1, isNVStorable = 1, opExtentBits = 6,
+ opExtendable = 1 in
+class T_StoreAbs_Pred <string mnemonic, RegisterClass RC, bits<2> MajOp,
+ bit isHalf, bit isNot, bit isNew>
+ : STInst<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, RC: $src2),
+ !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ",
+ ") ")#mnemonic#"(#$absaddr) = $src2"#!if(isHalf, ".h",""),
+ [], "", ST_tc_st_SLOT01>, AddrModeRel {
+ bits<2> src1;
+ bits<6> absaddr;
+ bits<5> src2;
-// if (!Ps.new) dealloc_return:nt
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1, isPredicatedFalse = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cNotdnPnt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
- "if (!$src1.new) dealloc_return:nt",
- []>,
- Requires<[HasV4T]>;
-}
+ let isPredicatedNew = isNew;
+ let isPredicatedFalse = isNot;
-// if (Ps.new) dealloc_return:t
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
- "if ($src1.new) dealloc_return:t",
- []>,
- Requires<[HasV4T]>;
-}
+ let IClass = 0b1010;
-// if (!Ps.new) dealloc_return:nt
-let isReturn = 1, isTerminator = 1,
- Defs = [R29, R30, R31, PC], Uses = [R30], hasSideEffects = 0,
- isPredicated = 1, isPredicatedFalse = 1 in {
-let validSubTargets = HasV4SubT in
- def DEALLOC_RET_cNotdnPt_V4 : LD0Inst<(outs), (ins PredRegs:$src1),
- "if (!$src1.new) dealloc_return:t",
- []>,
- Requires<[HasV4T]>;
+ let Inst{27-24} = 0b1111;
+ let Inst{23-22} = MajOp;
+ let Inst{21} = isHalf;
+ let Inst{17-16} = absaddr{5-4};
+ let Inst{13} = isNew;
+ let Inst{12-8} = src2;
+ let Inst{7} = 0b1;
+ let Inst{6-3} = absaddr{3-0};
+ let Inst{2} = isNot;
+ let Inst{1-0} = src1;
+ }
+
+//===----------------------------------------------------------------------===//
+// Template class for predicated store instructions with absolute addressing.
+//===----------------------------------------------------------------------===//
+class T_StoreAbs <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<2> MajOp, bit isHalf>
+ : T_StoreAbsGP <mnemonic, RC, ImmOp, MajOp, u0AlwaysExt, 1, isHalf>,
+ AddrModeRel {
+ string ImmOpStr = !cast<string>(ImmOp);
+ let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 18,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 17,
+ /* u16_0Imm */ 16)));
+
+ let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 2,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 1,
+ /* u16_0Imm */ 0)));
}
-// Load/Store with absolute addressing mode
-// memw(#u6)=Rt
+//===----------------------------------------------------------------------===//
+// Multiclass for store instructions with absolute addressing.
+//===----------------------------------------------------------------------===//
+let validSubTargets = HasV4SubT, addrMode = Absolute, isExtended = 1 in
+multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC,
+ Operand ImmOp, bits<2> MajOp, bit isHalf = 0> {
+ let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
+ let opExtendable = 0, isPredicable = 1 in
+ def S2_#NAME#abs : T_StoreAbs <mnemonic, RC, ImmOp, MajOp, isHalf>;
-multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
- bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME#_V4 : STInst2<(outs),
- (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#mnemonic#"(##$absaddr) = $src2",
- []>,
- Requires<[HasV4T]>;
-}
+ // Predicated
+ def S4_p#NAME#t_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 0, 0>;
+ def S4_p#NAME#f_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 1, 0>;
-multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
- // Predicate new
- defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
+ // .new Predicated
+ def S4_p#NAME#tnew_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 0, 1>;
+ def S4_p#NAME#fnew_abs : T_StoreAbs_Pred<mnemonic, RC, MajOp, isHalf, 1, 1>;
}
}
-let isNVStorable = 1, isExtended = 1, hasSideEffects = 0 in
-multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
- let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
- let opExtendable = 0, isPredicable = 1 in
- def NAME#_V4 : STInst2<(outs),
- (ins u0AlwaysExt:$absaddr, RC:$src),
- mnemonic#"(##$absaddr) = $src",
- []>,
- Requires<[HasV4T]>;
+//===----------------------------------------------------------------------===//
+// Template class for non predicated new-value store instructions with
+// GP-Relative or absolute addressing.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, isPredicable = 1, mayStore = 1, isNVStore = 1,
+ isNewValue = 1, opNewValue = 1 in
+class T_StoreAbsGP_NV <string mnemonic, Operand ImmOp, bits<2>MajOp, bit isAbs>
+ : NVInst_V4<(outs), (ins u0AlwaysExt:$addr, IntRegs:$src),
+ mnemonic # !if(isAbs, "(##", "(#")#"$addr) = $src.new",
+ [], "", V2LDST_tc_st_SLOT0> {
+ bits<19> addr;
+ bits<3> src;
+ bits<16> offsetBits;
- let opExtendable = 1, isPredicated = 1 in {
- defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
- defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
- }
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
+ !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
+ !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
+ /* u16_0Imm */ addr{15-0})));
+ let IClass = 0b0100;
+
+ let Inst{27} = 1;
+ let Inst{26-25} = offsetBits{15-14};
+ let Inst{24-21} = 0b0101;
+ let Inst{20-16} = offsetBits{13-9};
+ let Inst{13} = offsetBits{8};
+ let Inst{12-11} = MajOp;
+ let Inst{10-8} = src;
+ let Inst{7-0} = offsetBits{7-0};
}
-}
-multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
- bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME#_nv_V4 : NVInst_V4<(outs),
- (ins PredRegs:$src1, u0AlwaysExt:$absaddr, RC: $src2),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#mnemonic#"(##$absaddr) = $src2.new",
- []>,
- Requires<[HasV4T]>;
+//===----------------------------------------------------------------------===//
+// Template class for predicated new-value store instructions with
+// absolute addressing.
+//===----------------------------------------------------------------------===//
+let hasSideEffects = 0, isPredicated = 1, mayStore = 1, isNVStore = 1,
+ isNewValue = 1, opNewValue = 2, opExtentBits = 6, opExtendable = 1 in
+class T_StoreAbs_NV_Pred <string mnemonic, bits<2> MajOp, bit isNot, bit isNew>
+ : NVInst_V4<(outs), (ins PredRegs:$src1, u6Ext:$absaddr, IntRegs:$src2),
+ !if(isNot, "if (!$src1", "if ($src1")#!if(isNew, ".new) ",
+ ") ")#mnemonic#"(#$absaddr) = $src2.new",
+ [], "", ST_tc_st_SLOT0>, AddrModeRel {
+ bits<2> src1;
+ bits<6> absaddr;
+ bits<3> src2;
+
+ let isPredicatedNew = isNew;
+ let isPredicatedFalse = isNot;
+
+ let IClass = 0b1010;
+
+ let Inst{27-24} = 0b1111;
+ let Inst{23-21} = 0b101;
+ let Inst{17-16} = absaddr{5-4};
+ let Inst{13} = isNew;
+ let Inst{12-11} = MajOp;
+ let Inst{10-8} = src2;
+ let Inst{7} = 0b1;
+ let Inst{6-3} = absaddr{3-0};
+ let Inst{2} = isNot;
+ let Inst{1-0} = src1;
}
-multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
- // Predicate new
- defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
- }
+//===----------------------------------------------------------------------===//
+// Template class for non-predicated new-value store instructions with
+// absolute addressing.
+//===----------------------------------------------------------------------===//
+class T_StoreAbs_NV <string mnemonic, Operand ImmOp, bits<2> MajOp>
+ : T_StoreAbsGP_NV <mnemonic, ImmOp, MajOp, 1>, AddrModeRel {
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 18,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 17,
+ /* u16_0Imm */ 16)));
+
+ let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 2,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 1,
+ /* u16_0Imm */ 0)));
}
-let mayStore = 1, isNVStore = 1, isExtended = 1, hasSideEffects = 0 in
-multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
+//===----------------------------------------------------------------------===//
+// Multiclass for new-value store instructions with absolute addressing.
+//===----------------------------------------------------------------------===//
+let validSubTargets = HasV4SubT, addrMode = Absolute, isExtended = 1 in
+multiclass ST_Abs_NV <string mnemonic, string CextOp, Operand ImmOp,
+ bits<2> MajOp> {
let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
let opExtendable = 0, isPredicable = 1 in
- def NAME#_nv_V4 : NVInst_V4<(outs),
- (ins u0AlwaysExt:$absaddr, RC:$src),
- mnemonic#"(##$absaddr) = $src.new",
- []>,
- Requires<[HasV4T]>;
+ def S2_#NAME#newabs : T_StoreAbs_NV <mnemonic, ImmOp, MajOp>;
- let opExtendable = 1, isPredicated = 1 in {
- defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
- defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
- }
+ // Predicated
+ def S4_p#NAME#newt_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 0, 0>;
+ def S4_p#NAME#newf_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 1, 0>;
+
+ // .new Predicated
+ def S4_p#NAME#newtnew_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 0, 1>;
+ def S4_p#NAME#newfnew_abs : T_StoreAbs_NV_Pred <mnemonic, MajOp, 1, 1>;
}
}
-let addrMode = Absolute in {
- let accessSize = ByteAccess in
- defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
- ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
+//===----------------------------------------------------------------------===//
+// Stores with absolute addressing
+//===----------------------------------------------------------------------===//
+let accessSize = ByteAccess, isCodeGenOnly = 0 in
+defm storerb : ST_Abs <"memb", "STrib", IntRegs, u16_0Imm, 0b00>,
+ ST_Abs_NV <"memb", "STrib", u16_0Imm, 0b00>;
- let accessSize = HalfWordAccess in
- defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
- ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
+let accessSize = HalfWordAccess, isCodeGenOnly = 0 in
+defm storerh : ST_Abs <"memh", "STrih", IntRegs, u16_1Imm, 0b01>,
+ ST_Abs_NV <"memh", "STrih", u16_1Imm, 0b01>;
- let accessSize = WordAccess in
- defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
- ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
+let accessSize = WordAccess, isCodeGenOnly = 0 in
+defm storeri : ST_Abs <"memw", "STriw", IntRegs, u16_2Imm, 0b10>,
+ ST_Abs_NV <"memw", "STriw", u16_2Imm, 0b10>;
- let accessSize = DoubleWordAccess, isNVStorable = 0 in
- defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
+let isNVStorable = 0, accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+defm storerd : ST_Abs <"memd", "STrid", DoubleRegs, u16_3Imm, 0b11>;
+
+let isNVStorable = 0, accessSize = HalfWordAccess, isCodeGenOnly = 0 in
+defm storerf : ST_Abs <"memh", "STrif", IntRegs, u16_1Imm, 0b01, 1>;
+
+//===----------------------------------------------------------------------===//
+// GP-relative stores.
+// mem[bhwd](#global)=Rt
+// Once predicated, these instructions map to absolute addressing mode.
+// if ([!]Pv[.new]) mem[bhwd](##global)=Rt
+//===----------------------------------------------------------------------===//
+
+let validSubTargets = HasV4SubT in
+class T_StoreGP <string mnemonic, string BaseOp, RegisterClass RC,
+ Operand ImmOp, bits<2> MajOp, bit isHalf = 0>
+ : T_StoreAbsGP <mnemonic, RC, ImmOp, MajOp, globaladdress, 0, isHalf> {
+ // Set BaseOpcode same as absolute addressing instructions so that
+ // non-predicated GP-Rel instructions can have relate with predicated
+ // Absolute instruction.
+ let BaseOpcode = BaseOp#_abs;
+ }
+
+let validSubTargets = HasV4SubT in
+multiclass ST_GP <string mnemonic, string BaseOp, Operand ImmOp,
+ bits<2> MajOp, bit isHalf = 0> {
+ // Set BaseOpcode same as absolute addressing instructions so that
+ // non-predicated GP-Rel instructions can have relate with predicated
+ // Absolute instruction.
+ let BaseOpcode = BaseOp#_abs in {
+ def NAME#gp : T_StoreAbsGP <mnemonic, IntRegs, ImmOp, MajOp,
+ globaladdress, 0, isHalf>;
+ // New-value store
+ def NAME#newgp : T_StoreAbsGP_NV <mnemonic, ImmOp, MajOp, 0> ;
+ }
}
+let accessSize = ByteAccess in
+defm S2_storerb : ST_GP<"memb", "STrib", u16_0Imm, 0b00>, NewValueRel;
+
+let accessSize = HalfWordAccess in
+defm S2_storerh : ST_GP<"memh", "STrih", u16_1Imm, 0b01>, NewValueRel;
+
+let accessSize = WordAccess in
+defm S2_storeri : ST_GP<"memw", "STriw", u16_2Imm, 0b10>, NewValueRel;
+
+let isNVStorable = 0, accessSize = DoubleWordAccess in
+def S2_storerdgp : T_StoreGP <"memd", "STrid", DoubleRegs,
+ u16_3Imm, 0b11>, PredNewRel;
+
+let isNVStorable = 0, accessSize = HalfWordAccess in
+def S2_storerfgp : T_StoreGP <"memh", "STrif", IntRegs,
+ u16_1Imm, 0b01, 1>, PredNewRel;
+
let Predicates = [HasV4T], AddedComplexity = 30 in {
def : Pat<(truncstorei8 (i32 IntRegs:$src1),
(HexagonCONST32 tglobaladdr:$absaddr)),
- (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
+ (S2_storerbabs tglobaladdr: $absaddr, IntRegs: $src1)>;
def : Pat<(truncstorei16 (i32 IntRegs:$src1),
(HexagonCONST32 tglobaladdr:$absaddr)),
- (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
+ (S2_storerhabs tglobaladdr: $absaddr, IntRegs: $src1)>;
def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
- (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
+ (S2_storeriabs tglobaladdr: $absaddr, IntRegs: $src1)>;
def : Pat<(store (i64 DoubleRegs:$src1),
(HexagonCONST32 tglobaladdr:$absaddr)),
- (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
-}
-
-//===----------------------------------------------------------------------===//
-// multiclass for store instructions with GP-relative addressing mode.
-// mem[bhwd](#global)=Rt
-// if ([!]Pv[.new]) mem[bhwd](##global) = Rt
-//===----------------------------------------------------------------------===//
-let mayStore = 1, isNVStorable = 1 in
-multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
- let BaseOpcode = BaseOp, isPredicable = 1 in
- def NAME#_V4 : STInst2<(outs),
- (ins globaladdress:$global, RC:$src),
- mnemonic#"(#$global) = $src",
- []>;
-
- // When GP-relative instructions are predicated, their addressing mode is
- // changed to absolute and they are always constant extended.
- let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
- isPredicated = 1 in {
- defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
- defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
- }
-}
-
-let mayStore = 1, isNVStore = 1 in
-multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
- let BaseOpcode = BaseOp, isPredicable = 1 in
- def NAME#_nv_V4 : NVInst_V4<(outs),
- (ins u0AlwaysExt:$global, RC:$src),
- mnemonic#"(#$global) = $src.new",
- []>,
- Requires<[HasV4T]>;
-
- // When GP-relative instructions are predicated, their addressing mode is
- // changed to absolute and they are always constant extended.
- let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
- isPredicated = 1 in {
- defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
- defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
- }
-}
-
-let validSubTargets = HasV4SubT, hasSideEffects = 0 in {
- let isNVStorable = 0 in
- defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>, PredNewRel;
-
- defm STb_GP : ST_GP<"memb", "STb_GP", IntRegs>,
- ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel;
- defm STh_GP : ST_GP<"memh", "STh_GP", IntRegs>,
- ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel;
- defm STw_GP : ST_GP<"memw", "STw_GP", IntRegs>,
- ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel;
+ (S2_storerdabs tglobaladdr: $absaddr, DoubleRegs: $src1)>;
}
// 64 bit atomic store
def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
(i64 DoubleRegs:$src1)),
- (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
+ (S2_storerdgp tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
Requires<[HasV4T]>;
// Map from store(globaladdress) -> memd(#foo)
let AddedComplexity = 100 in
def : Pat <(store (i64 DoubleRegs:$src1),
(HexagonCONST32_GP tglobaladdr:$global)),
- (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
+ (S2_storerdgp tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
// 8 bit atomic store
def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
(i32 IntRegs:$src1)),
- (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerbgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
// Map from store(globaladdress) -> memb(#foo)
let AddedComplexity = 100 in
def : Pat<(truncstorei8 (i32 IntRegs:$src1),
(HexagonCONST32_GP tglobaladdr:$global)),
- (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerbgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
// Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
// to "r0 = 1; memw(#foo) = r0"
let AddedComplexity = 100 in
def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
- (STb_GP_V4 tglobaladdr:$global, (A2_tfrsi 1))>;
+ (S2_storerbgp tglobaladdr:$global, (A2_tfrsi 1))>;
def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
(i32 IntRegs:$src1)),
- (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerhgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
// Map from store(globaladdress) -> memh(#foo)
let AddedComplexity = 100 in
def : Pat<(truncstorei16 (i32 IntRegs:$src1),
(HexagonCONST32_GP tglobaladdr:$global)),
- (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerhgp tglobaladdr:$global, (i32 IntRegs:$src1))>;
// 32 bit atomic store
def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
(i32 IntRegs:$src1)),
- (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerigp tglobaladdr:$global, (i32 IntRegs:$src1))>;
// Map from store(globaladdress) -> memw(#foo)
let AddedComplexity = 100 in
def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
- (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+ (S2_storerigp tglobaladdr:$global, (i32 IntRegs:$src1))>;
//===----------------------------------------------------------------------===//
-// Multiclass for the load instructions with absolute addressing mode.
+// Template class for non predicated load instructions with
+// absolute addressing mode.
//===----------------------------------------------------------------------===//
-multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
- bit isPredNew> {
- let isPredicatedNew = isPredNew in
- def NAME : LDInst2<(outs RC:$dst),
- (ins PredRegs:$src1, u0AlwaysExt:$absaddr),
- !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
- ") ")#"$dst = "#mnemonic#"(##$absaddr)",
- []>,
- Requires<[HasV4T]>;
-}
+let isPredicable = 1, hasSideEffects = 0, validSubTargets = HasV4SubT in
+class T_LoadAbsGP <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<3> MajOp, Operand AddrOp, bit isAbs>
+ : LDInst <(outs RC:$dst), (ins AddrOp:$addr),
+ "$dst = "#mnemonic# !if(isAbs, "(##", "(#")#"$addr)",
+ [], "", V2LDST_tc_ld_SLOT01> {
+ bits<5> dst;
+ bits<19> addr;
+ bits<16> offsetBits;
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let offsetBits = !if (!eq(ImmOpStr, "u16_3Imm"), addr{18-3},
+ !if (!eq(ImmOpStr, "u16_2Imm"), addr{17-2},
+ !if (!eq(ImmOpStr, "u16_1Imm"), addr{16-1},
+ /* u16_0Imm */ addr{15-0})));
+
+ let IClass = 0b0100;
+
+ let Inst{27} = 0b1;
+ let Inst{26-25} = offsetBits{15-14};
+ let Inst{24} = 0b1;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = offsetBits{13-9};
+ let Inst{13-5} = offsetBits{8-0};
+ let Inst{4-0} = dst;
+ }
-multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
- let isPredicatedFalse = PredNot in {
- defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
- // Predicate new
- defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
+class T_LoadAbs <string mnemonic, RegisterClass RC, Operand ImmOp,
+ bits<3> MajOp>
+ : T_LoadAbsGP <mnemonic, RC, ImmOp, MajOp, u0AlwaysExt, 1>, AddrModeRel {
+
+ string ImmOpStr = !cast<string>(ImmOp);
+ let opExtentBits = !if (!eq(ImmOpStr, "u16_3Imm"), 19,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 18,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 17,
+ /* u16_0Imm */ 16)));
+
+ let opExtentAlign = !if (!eq(ImmOpStr, "u16_3Imm"), 3,
+ !if (!eq(ImmOpStr, "u16_2Imm"), 2,
+ !if (!eq(ImmOpStr, "u16_1Imm"), 1,
+ /* u16_0Imm */ 0)));
}
+//===----------------------------------------------------------------------===//
+// Template class for predicated load instructions with
+// absolute addressing mode.
+//===----------------------------------------------------------------------===//
+let isPredicated = 1, hasNewValue = 1, opExtentBits = 6, opExtendable = 2 in
+class T_LoadAbs_Pred <string mnemonic, RegisterClass RC, bits<3> MajOp,
+ bit isPredNot, bit isPredNew>
+ : LDInst <(outs RC:$dst), (ins PredRegs:$src1, u6Ext:$absaddr),
+ !if(isPredNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+ ") ")#"$dst = "#mnemonic#"(#$absaddr)">, AddrModeRel {
+ bits<5> dst;
+ bits<2> src1;
+ bits<6> absaddr;
+
+ let isPredicatedNew = isPredNew;
+ let isPredicatedFalse = isPredNot;
+
+ let IClass = 0b1001;
+
+ let Inst{27-24} = 0b1111;
+ let Inst{23-21} = MajOp;
+ let Inst{20-16} = absaddr{5-1};
+ let Inst{13} = 0b1;
+ let Inst{12} = isPredNew;
+ let Inst{11} = isPredNot;
+ let Inst{10-9} = src1;
+ let Inst{8} = absaddr{0};
+ let Inst{7} = 0b1;
+ let Inst{4-0} = dst;
+ }
+
+//===----------------------------------------------------------------------===//
+// Multiclass for the load instructions with absolute addressing mode.
+//===----------------------------------------------------------------------===//
+multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bits<3> MajOp,
+ bit PredNot> {
+ def _abs : T_LoadAbs_Pred <mnemonic, RC, MajOp, PredNot, 0>;
+ // Predicate new
+ def new_abs : T_LoadAbs_Pred <mnemonic, RC, MajOp, PredNot, 1>;
}
-let isExtended = 1, hasSideEffects = 0 in
-multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
+let addrMode = Absolute, isExtended = 1 in
+multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC,
+ Operand ImmOp, bits<3> MajOp> {
let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
- let opExtendable = 1, isPredicable = 1 in
- def NAME#_V4 : LDInst2<(outs RC:$dst),
- (ins u0AlwaysExt:$absaddr),
- "$dst = "#mnemonic#"(##$absaddr)",
- []>,
- Requires<[HasV4T]>;
+ let opExtendable = 1, isPredicable = 1 in
+ def L4_#NAME#_abs: T_LoadAbs <mnemonic, RC, ImmOp, MajOp>;
- let opExtendable = 2, isPredicated = 1 in {
- defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
- defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
- }
+ // Predicated
+ defm L4_p#NAME#t : LD_Abs_Pred<mnemonic, RC, MajOp, 0>;
+ defm L4_p#NAME#f : LD_Abs_Pred<mnemonic, RC, MajOp, 1>;
}
}
-let addrMode = Absolute in {
- let accessSize = ByteAccess in {
- defm LDrib_abs : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
- defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
- }
- let accessSize = HalfWordAccess in {
- defm LDrih_abs : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
- defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
+let accessSize = ByteAccess, hasNewValue = 1, isCodeGenOnly = 0 in {
+ defm loadrb : LD_Abs<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>;
+ defm loadrub : LD_Abs<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>;
+}
+
+let accessSize = HalfWordAccess, hasNewValue = 1, isCodeGenOnly = 0 in {
+ defm loadrh : LD_Abs<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>;
+ defm loadruh : LD_Abs<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>;
+}
+
+let accessSize = WordAccess, hasNewValue = 1, isCodeGenOnly = 0 in
+defm loadri : LD_Abs<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>;
+
+let accessSize = DoubleWordAccess, isCodeGenOnly = 0 in
+defm loadrd : LD_Abs<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>;
+
+//===----------------------------------------------------------------------===//
+// multiclass for load instructions with GP-relative addressing mode.
+// Rx=mem[bhwd](##global)
+// Once predicated, these instructions map to absolute addressing mode.
+// if ([!]Pv[.new]) Rx=mem[bhwd](##global)
+//===----------------------------------------------------------------------===//
+
+class T_LoadGP <string mnemonic, string BaseOp, RegisterClass RC, Operand ImmOp,
+ bits<3> MajOp>
+ : T_LoadAbsGP <mnemonic, RC, ImmOp, MajOp, globaladdress, 0>, PredNewRel {
+ let BaseOpcode = BaseOp#_abs;
}
- let accessSize = WordAccess in
- defm LDriw_abs : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
- let accessSize = DoubleWordAccess in
- defm LDrid_abs : LD_Abs<"memd", "LDrid", DoubleRegs>, AddrModeRel;
+let accessSize = ByteAccess, hasNewValue = 1 in {
+ def L2_loadrbgp : T_LoadGP<"memb", "LDrib", IntRegs, u16_0Imm, 0b000>;
+ def L2_loadrubgp : T_LoadGP<"memub", "LDriub", IntRegs, u16_0Imm, 0b001>;
}
+let accessSize = HalfWordAccess, hasNewValue = 1 in {
+ def L2_loadrhgp : T_LoadGP<"memh", "LDrih", IntRegs, u16_1Imm, 0b010>;
+ def L2_loadruhgp : T_LoadGP<"memuh", "LDriuh", IntRegs, u16_1Imm, 0b011>;
+}
+
+let accessSize = WordAccess, hasNewValue = 1 in
+def L2_loadrigp : T_LoadGP<"memw", "LDriw", IntRegs, u16_2Imm, 0b100>;
+
+let accessSize = DoubleWordAccess in
+def L2_loadrdgp : T_LoadGP<"memd", "LDrid", DoubleRegs, u16_3Imm, 0b110>;
+
let Predicates = [HasV4T], AddedComplexity = 30 in {
def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
- (LDriw_abs_V4 tglobaladdr: $absaddr)>;
+ (L4_loadri_abs tglobaladdr: $absaddr)>;
def : Pat<(i32 (sextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
- (LDrib_abs_V4 tglobaladdr:$absaddr)>;
+ (L4_loadrb_abs tglobaladdr:$absaddr)>;
def : Pat<(i32 (zextloadi8 (HexagonCONST32 tglobaladdr:$absaddr))),
- (LDriub_abs_V4 tglobaladdr:$absaddr)>;
+ (L4_loadrub_abs tglobaladdr:$absaddr)>;
def : Pat<(i32 (sextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
- (LDrih_abs_V4 tglobaladdr:$absaddr)>;
+ (L4_loadrh_abs tglobaladdr:$absaddr)>;
def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
- (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
-}
-
-//===----------------------------------------------------------------------===//
-// multiclass for load instructions with GP-relative addressing mode.
-// Rx=mem[bhwd](##global)
-// if ([!]Pv[.new]) Rx=mem[bhwd](##global)
-//===----------------------------------------------------------------------===//
-let hasSideEffects = 0, validSubTargets = HasV4SubT in
-multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
- let BaseOpcode = BaseOp in {
- let isPredicable = 1 in
- def NAME#_V4 : LDInst2<(outs RC:$dst),
- (ins globaladdress:$global),
- "$dst = "#mnemonic#"(#$global)",
- []>;
-
- let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
- defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
- defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
- }
- }
+ (L4_loadruh_abs tglobaladdr:$absaddr)>;
}
-defm LDd_GP : LD_GP<"memd", "LDd_GP", DoubleRegs>, PredNewRel;
-defm LDb_GP : LD_GP<"memb", "LDb_GP", IntRegs>, PredNewRel;
-defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>, PredNewRel;
-defm LDh_GP : LD_GP<"memh", "LDh_GP", IntRegs>, PredNewRel;
-defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>, PredNewRel;
-defm LDw_GP : LD_GP<"memw", "LDw_GP", IntRegs>, PredNewRel;
-
def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
- (i64 (LDd_GP_V4 tglobaladdr:$global))>;
+ (i64 (L2_loadrdgp tglobaladdr:$global))>;
def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
- (i32 (LDw_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrigp tglobaladdr:$global))>;
def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
- (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadruhgp tglobaladdr:$global))>;
def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
- (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrubgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memw(#foo + 0)
let AddedComplexity = 100 in
def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
- (i64 (LDd_GP_V4 tglobaladdr:$global))>;
+ (i64 (L2_loadrdgp tglobaladdr:$global))>;
// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
let AddedComplexity = 100 in
def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
- (i1 (C2_tfrrp (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
+ (i1 (C2_tfrrp (i32 (L2_loadrbgp tglobaladdr:$global))))>;
// When the Interprocedural Global Variable optimizer realizes that a certain
// global variable takes only two constant values, it shrinks the global to
// a boolean. Catch those loads here in the following 3 patterns.
let AddedComplexity = 100 in
def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrbgp tglobaladdr:$global))>;
let AddedComplexity = 100 in
def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrbgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memb(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrbgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memb(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrbgp tglobaladdr:$global))>;
let AddedComplexity = 100 in
def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrubgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memub(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrubgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memh(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDh_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrhgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memh(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDh_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrhgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memuh(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadruhgp tglobaladdr:$global))>;
// Map from load(globaladdress) -> memw(#foo)
let AddedComplexity = 100 in
def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
- (i32 (LDw_GP_V4 tglobaladdr:$global))>;
+ (i32 (L2_loadrigp tglobaladdr:$global))>;
// Transfer global address into a register
let Predicates = [HasV4T], AddedComplexity = 30 in {
def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
- (STrib_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
+ (S2_storerbabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
- (STrih_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
+ (S2_storerhabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
- (STriw_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
+ (S2_storeriabs u0AlwaysExtPred:$src2, IntRegs: $src1)>;
}
let Predicates = [HasV4T], AddedComplexity = 30 in {
def : Pat<(i32 (load u0AlwaysExtPred:$src)),
- (LDriw_abs_V4 u0AlwaysExtPred:$src)>;
+ (L4_loadri_abs u0AlwaysExtPred:$src)>;
def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
- (LDrib_abs_V4 u0AlwaysExtPred:$src)>;
+ (L4_loadrb_abs u0AlwaysExtPred:$src)>;
def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
- (LDriub_abs_V4 u0AlwaysExtPred:$src)>;
+ (L4_loadrub_abs u0AlwaysExtPred:$src)>;
def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
- (LDrih_abs_V4 u0AlwaysExtPred:$src)>;
+ (L4_loadrh_abs u0AlwaysExtPred:$src)>;
def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
- (LDriuh_abs_V4 u0AlwaysExtPred:$src)>;
+ (L4_loadruh_abs u0AlwaysExtPred:$src)>;
}
// Indexed store word - global address.
// zextloadi8.
let Predicates = [HasV4T], AddedComplexity = 120 in {
def: Pat <(i64 (extloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDrib_abs_V4 tglobaladdr:$addr)))>;
+ (i64 (A4_combineir 0, (L4_loadrb_abs tglobaladdr:$addr)))>;
def: Pat <(i64 (zextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDriub_abs_V4 tglobaladdr:$addr)))>;
+ (i64 (A4_combineir 0, (L4_loadrub_abs tglobaladdr:$addr)))>;
def: Pat <(i64 (sextloadi8 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A2_sxtw (LDrib_abs_V4 tglobaladdr:$addr)))>;
+ (i64 (A2_sxtw (L4_loadrb_abs tglobaladdr:$addr)))>;
def: Pat <(i64 (extloadi8 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
+ (i64 (A4_combineir 0, (L4_loadrb_abs FoldGlobalAddr:$addr)))>;
def: Pat <(i64 (zextloadi8 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDriub_abs_V4 FoldGlobalAddr:$addr)))>;
+ (i64 (A4_combineir 0, (L4_loadrub_abs FoldGlobalAddr:$addr)))>;
def: Pat <(i64 (sextloadi8 FoldGlobalAddr:$addr)),
- (i64 (A2_sxtw (LDrib_abs_V4 FoldGlobalAddr:$addr)))>;
+ (i64 (A2_sxtw (L4_loadrb_abs FoldGlobalAddr:$addr)))>;
}
// i16 -> i64 loads
// We need a complexity of 120 here to override preceding handling of
// zextloadi16.
let AddedComplexity = 120 in {
def: Pat <(i64 (extloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDrih_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadrh_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (zextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDriuh_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadruh_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (sextloadi16 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A2_sxtw (LDrih_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A2_sxtw (L4_loadrh_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (extloadi16 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadrh_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (zextloadi16 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDriuh_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadruh_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (sextloadi16 FoldGlobalAddr:$addr)),
- (i64 (A2_sxtw (LDrih_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A2_sxtw (L4_loadrh_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
}
// i32->i64 loads
// zextloadi32.
let AddedComplexity = 120 in {
def: Pat <(i64 (extloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadri_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (zextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A4_combineir 0, (LDriw_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadri_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (sextloadi32 (NumUsesBelowThresCONST32 tglobaladdr:$addr))),
- (i64 (A2_sxtw (LDriw_abs_V4 tglobaladdr:$addr)))>,
+ (i64 (A2_sxtw (L4_loadri_abs tglobaladdr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (extloadi32 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadri_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (zextloadi32 FoldGlobalAddr:$addr)),
- (i64 (A4_combineir 0, (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A4_combineir 0, (L4_loadri_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
def: Pat <(i64 (sextloadi32 FoldGlobalAddr:$addr)),
- (i64 (A2_sxtw (LDriw_abs_V4 FoldGlobalAddr:$addr)))>,
+ (i64 (A2_sxtw (L4_loadri_abs FoldGlobalAddr:$addr)))>,
Requires<[HasV4T]>;
}
let AddedComplexity = 100 in
def : Pat<(store (i64 DoubleRegs:$src1),
FoldGlobalAddrGP:$addr),
- (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
+ (S2_storerdabs FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
Requires<[HasV4T]>;
def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
(i64 DoubleRegs:$src1)),
- (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
+ (S2_storerdabs FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
Requires<[HasV4T]>;
// Map from store(globaladdress + x) -> memb(#foo + x)
let AddedComplexity = 100 in
def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
- (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storerbabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
- (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storerbabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
// Map from store(globaladdress + x) -> memh(#foo + x)
let AddedComplexity = 100 in
def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
- (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storerhabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
- (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storerhabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
// Map from store(globaladdress + x) -> memw(#foo + x)
let AddedComplexity = 100 in
def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
- (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storeriabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
- (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+ (S2_storeriabs FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memd(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
- (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i64 (L4_loadrd_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
- (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i64 (L4_loadrd_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memb(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
- (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrb_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memb(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
- (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrb_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
//let AddedComplexity = 100 in
let AddedComplexity = 100 in
def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
- (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrh_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memh(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
- (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrh_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memuh(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
- (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadruh_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
- (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadruh_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memub(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
- (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrub_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
- (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadrub_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
// Map from load(globaladdress + x) -> memw(#foo + x)
let AddedComplexity = 100 in
def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
- (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadri_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
- (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
+ (i32 (L4_loadri_abs FoldGlobalAddrGP:$addr))>,
Requires<[HasV4T]>;
+
+//===----------------------------------------------------------------------===//
+// :raw for of boundscheck:hi:lo insns
+//===----------------------------------------------------------------------===//
+
+// A4_boundscheck_lo: Detect if a register is within bounds.
+let hasSideEffects = 0, isCodeGenOnly = 0 in
+def A4_boundscheck_lo: ALU64Inst <
+ (outs PredRegs:$Pd),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Pd = boundscheck($Rss, $Rtt):raw:lo"> {
+ bits<2> Pd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1101;
+
+ let Inst{27-23} = 0b00100;
+ let Inst{13} = 0b1;
+ let Inst{7-5} = 0b100;
+ let Inst{1-0} = Pd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+// A4_boundscheck_hi: Detect if a register is within bounds.
+let hasSideEffects = 0, isCodeGenOnly = 0 in
+def A4_boundscheck_hi: ALU64Inst <
+ (outs PredRegs:$Pd),
+ (ins DoubleRegs:$Rss, DoubleRegs:$Rtt),
+ "$Pd = boundscheck($Rss, $Rtt):raw:hi"> {
+ bits<2> Pd;
+ bits<5> Rss;
+ bits<5> Rtt;
+
+ let IClass = 0b1101;
+
+ let Inst{27-23} = 0b00100;
+ let Inst{13} = 0b1;
+ let Inst{7-5} = 0b101;
+ let Inst{1-0} = Pd;
+ let Inst{20-16} = Rss;
+ let Inst{12-8} = Rtt;
+ }
+
+let hasSideEffects = 0 in
+def A4_boundscheck : MInst <
+ (outs PredRegs:$Pd), (ins IntRegs:$Rs, DoubleRegs:$Rtt),
+ "$Pd=boundscheck($Rs,$Rtt)">;
+
+// A4_tlbmatch: Detect if a VA/ASID matches a TLB entry.
+let isPredicateLate = 1, hasSideEffects = 0, isCodeGenOnly = 0 in
+def A4_tlbmatch : ALU64Inst<(outs PredRegs:$Pd),
+ (ins DoubleRegs:$Rs, IntRegs:$Rt),
+ "$Pd = tlbmatch($Rs, $Rt)",
+ [], "", ALU64_tc_2early_SLOT23> {
+ bits<2> Pd;
+ bits<5> Rs;
+ bits<5> Rt;
+
+ let IClass = 0b1101;
+ let Inst{27-23} = 0b00100;
+ let Inst{20-16} = Rs;
+ let Inst{13} = 0b1;
+ let Inst{12-8} = Rt;
+ let Inst{7-5} = 0b011;
+ let Inst{1-0} = Pd;
+ }
+
+// We need custom lowering of ISD::PREFETCH into HexagonISD::DCFETCH
+// because the SDNode ISD::PREFETCH has properties MayLoad and MayStore.
+// We don't really want either one here.
+def SDTHexagonDCFETCH : SDTypeProfile<0, 2, [SDTCisPtrTy<0>,SDTCisInt<1>]>;
+def HexagonDCFETCH : SDNode<"HexagonISD::DCFETCH", SDTHexagonDCFETCH,
+ [SDNPHasChain]>;
+
+// Use LD0Inst for dcfetch, but set "mayLoad" to 0 because this doesn't
+// really do a load.
+let hasSideEffects = 1, mayLoad = 0, isCodeGenOnly = 0 in
+def Y2_dcfetchbo : LD0Inst<(outs), (ins IntRegs:$Rs, u11_3Imm:$u11_3),
+ "dcfetch($Rs + #$u11_3)",
+ [(HexagonDCFETCH IntRegs:$Rs, u11_3ImmPred:$u11_3)],
+ "", LD_tc_ld_SLOT0> {
+ bits<5> Rs;
+ bits<14> u11_3;
+
+ let IClass = 0b1001;
+ let Inst{27-21} = 0b0100000;
+ let Inst{20-16} = Rs;
+ let Inst{13} = 0b0;
+ let Inst{10-0} = u11_3{13-3};
+}
+
+//===----------------------------------------------------------------------===//
+// Compound instructions
+//===----------------------------------------------------------------------===//
+
+let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
+ isPredicated = 1, isPredicatedNew = 1, isExtendable = 1,
+ opExtentBits = 11, opExtentAlign = 2, opExtendable = 1,
+ isTerminator = 1, validSubTargets = HasV4SubT in
+class CJInst_tstbit_R0<string px, bit np, string tnt>
+ : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2),
+ ""#px#" = tstbit($Rs, #0); if ("
+ #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
+ [], "", COMPOUND, TypeCOMPOUND> {
+ bits<4> Rs;
+ bits<11> r9_2;
+
+ // np: !p[01]
+ let isPredicatedFalse = np;
+ // tnt: Taken/Not Taken
+ let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
+ let isTaken = !if (!eq(tnt, "t"), 1, 0);
+
+ let IClass = 0b0001;
+ let Inst{27-26} = 0b00;
+ let Inst{25} = !if (!eq(px, "!p1"), 1,
+ !if (!eq(px, "p1"), 1, 0));
+ let Inst{24-23} = 0b11;
+ let Inst{22} = np;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{19-16} = Rs;
+ let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
+ let Inst{9-8} = 0b11;
+ let Inst{7-1} = r9_2{8-2};
+}
+
+let Defs = [PC, P0], Uses = [P0], isCodeGenOnly = 0 in {
+ def J4_tstbit0_tp0_jump_nt : CJInst_tstbit_R0<"p0", 0, "nt">;
+ def J4_tstbit0_tp0_jump_t : CJInst_tstbit_R0<"p0", 0, "t">;
+ def J4_tstbit0_fp0_jump_nt : CJInst_tstbit_R0<"p0", 1, "nt">;
+ def J4_tstbit0_fp0_jump_t : CJInst_tstbit_R0<"p0", 1, "t">;
+}
+
+let Defs = [PC, P1], Uses = [P1], isCodeGenOnly = 0 in {
+ def J4_tstbit0_tp1_jump_nt : CJInst_tstbit_R0<"p1", 0, "nt">;
+ def J4_tstbit0_tp1_jump_t : CJInst_tstbit_R0<"p1", 0, "t">;
+ def J4_tstbit0_fp1_jump_nt : CJInst_tstbit_R0<"p1", 1, "nt">;
+ def J4_tstbit0_fp1_jump_t : CJInst_tstbit_R0<"p1", 1, "t">;
+}
+
+
+let isBranch = 1, hasSideEffects = 0,
+ isExtentSigned = 1, isPredicated = 1, isPredicatedNew = 1,
+ isExtendable = 1, opExtentBits = 11, opExtentAlign = 2,
+ opExtendable = 2, isTerminator = 1, validSubTargets = HasV4SubT in
+class CJInst_RR<string px, string op, bit np, string tnt>
+ : InstHexagon<(outs), (ins IntRegs:$Rs, IntRegs:$Rt, brtarget:$r9_2),
+ ""#px#" = cmp."#op#"($Rs, $Rt); if ("
+ #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
+ [], "", COMPOUND, TypeCOMPOUND> {
+ bits<4> Rs;
+ bits<4> Rt;
+ bits<11> r9_2;
+
+ // np: !p[01]
+ let isPredicatedFalse = np;
+ // tnt: Taken/Not Taken
+ let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
+ let isTaken = !if (!eq(tnt, "t"), 1, 0);
+
+ let IClass = 0b0001;
+ let Inst{27-23} = !if (!eq(op, "eq"), 0b01000,
+ !if (!eq(op, "gt"), 0b01001,
+ !if (!eq(op, "gtu"), 0b01010, 0)));
+ let Inst{22} = np;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{19-16} = Rs;
+ let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
+ // px: Predicate reg 0/1
+ let Inst{12} = !if (!eq(px, "!p1"), 1,
+ !if (!eq(px, "p1"), 1, 0));
+ let Inst{11-8} = Rt;
+ let Inst{7-1} = r9_2{8-2};
+}
+
+// P[10] taken/not taken.
+multiclass T_tnt_CJInst_RR<string op, bit np> {
+ let Defs = [PC, P0], Uses = [P0] in {
+ def NAME#p0_jump_nt : CJInst_RR<"p0", op, np, "nt">;
+ def NAME#p0_jump_t : CJInst_RR<"p0", op, np, "t">;
+ }
+ let Defs = [PC, P1], Uses = [P1] in {
+ def NAME#p1_jump_nt : CJInst_RR<"p1", op, np, "nt">;
+ def NAME#p1_jump_t : CJInst_RR<"p1", op, np, "t">;
+ }
+}
+// Predicate / !Predicate
+multiclass T_pnp_CJInst_RR<string op>{
+ defm J4_cmp#NAME#_t : T_tnt_CJInst_RR<op, 0>;
+ defm J4_cmp#NAME#_f : T_tnt_CJInst_RR<op, 1>;
+}
+// TypeCJ Instructions compare RR and jump
+let isCodeGenOnly = 0 in {
+defm eq : T_pnp_CJInst_RR<"eq">;
+defm gt : T_pnp_CJInst_RR<"gt">;
+defm gtu : T_pnp_CJInst_RR<"gtu">;
+}
+
+let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
+ isPredicated = 1, isPredicatedNew = 1, isExtendable = 1, opExtentBits = 11,
+ opExtentAlign = 2, opExtendable = 2, isTerminator = 1,
+ validSubTargets = HasV4SubT in
+class CJInst_RU5<string px, string op, bit np, string tnt>
+ : InstHexagon<(outs), (ins IntRegs:$Rs, u5Imm:$U5, brtarget:$r9_2),
+ ""#px#" = cmp."#op#"($Rs, #$U5); if ("
+ #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
+ [], "", COMPOUND, TypeCOMPOUND> {
+ bits<4> Rs;
+ bits<5> U5;
+ bits<11> r9_2;
+
+ // np: !p[01]
+ let isPredicatedFalse = np;
+ // tnt: Taken/Not Taken
+ let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
+ let isTaken = !if (!eq(tnt, "t"), 1, 0);
+
+ let IClass = 0b0001;
+ let Inst{27-26} = 0b00;
+ // px: Predicate reg 0/1
+ let Inst{25} = !if (!eq(px, "!p1"), 1,
+ !if (!eq(px, "p1"), 1, 0));
+ let Inst{24-23} = !if (!eq(op, "eq"), 0b00,
+ !if (!eq(op, "gt"), 0b01,
+ !if (!eq(op, "gtu"), 0b10, 0)));
+ let Inst{22} = np;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{19-16} = Rs;
+ let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
+ let Inst{12-8} = U5;
+ let Inst{7-1} = r9_2{8-2};
+}
+// P[10] taken/not taken.
+multiclass T_tnt_CJInst_RU5<string op, bit np> {
+ let Defs = [PC, P0], Uses = [P0] in {
+ def NAME#p0_jump_nt : CJInst_RU5<"p0", op, np, "nt">;
+ def NAME#p0_jump_t : CJInst_RU5<"p0", op, np, "t">;
+ }
+ let Defs = [PC, P1], Uses = [P1] in {
+ def NAME#p1_jump_nt : CJInst_RU5<"p1", op, np, "nt">;
+ def NAME#p1_jump_t : CJInst_RU5<"p1", op, np, "t">;
+ }
+}
+// Predicate / !Predicate
+multiclass T_pnp_CJInst_RU5<string op>{
+ defm J4_cmp#NAME#i_t : T_tnt_CJInst_RU5<op, 0>;
+ defm J4_cmp#NAME#i_f : T_tnt_CJInst_RU5<op, 1>;
+}
+// TypeCJ Instructions compare RI and jump
+let isCodeGenOnly = 0 in {
+defm eq : T_pnp_CJInst_RU5<"eq">;
+defm gt : T_pnp_CJInst_RU5<"gt">;
+defm gtu : T_pnp_CJInst_RU5<"gtu">;
+}
+
+let isBranch = 1, hasSideEffects = 0, isExtentSigned = 1,
+ isPredicated = 1, isPredicatedFalse = 1, isPredicatedNew = 1,
+ isExtendable = 1, opExtentBits = 11, opExtentAlign = 2, opExtendable = 1,
+ isTerminator = 1, validSubTargets = HasV4SubT in
+class CJInst_Rn1<string px, string op, bit np, string tnt>
+ : InstHexagon<(outs), (ins IntRegs:$Rs, brtarget:$r9_2),
+ ""#px#" = cmp."#op#"($Rs,#-1); if ("
+ #!if(np, "!","")#""#px#".new) jump:"#tnt#" $r9_2",
+ [], "", COMPOUND, TypeCOMPOUND> {
+ bits<4> Rs;
+ bits<11> r9_2;
+
+ // np: !p[01]
+ let isPredicatedFalse = np;
+ // tnt: Taken/Not Taken
+ let isBrTaken = !if (!eq(tnt, "t"), "true", "false");
+ let isTaken = !if (!eq(tnt, "t"), 1, 0);
+
+ let IClass = 0b0001;
+ let Inst{27-26} = 0b00;
+ let Inst{25} = !if (!eq(px, "!p1"), 1,
+ !if (!eq(px, "p1"), 1, 0));
+
+ let Inst{24-23} = 0b11;
+ let Inst{22} = np;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{19-16} = Rs;
+ let Inst{13} = !if (!eq(tnt, "t"), 1, 0);
+ let Inst{9-8} = !if (!eq(op, "eq"), 0b00,
+ !if (!eq(op, "gt"), 0b01, 0));
+ let Inst{7-1} = r9_2{8-2};
+}
+
+// P[10] taken/not taken.
+multiclass T_tnt_CJInst_Rn1<string op, bit np> {
+ let Defs = [PC, P0], Uses = [P0] in {
+ def NAME#p0_jump_nt : CJInst_Rn1<"p0", op, np, "nt">;
+ def NAME#p0_jump_t : CJInst_Rn1<"p0", op, np, "t">;
+ }
+ let Defs = [PC, P1], Uses = [P1] in {
+ def NAME#p1_jump_nt : CJInst_Rn1<"p1", op, np, "nt">;
+ def NAME#p1_jump_t : CJInst_Rn1<"p1", op, np, "t">;
+ }
+}
+// Predicate / !Predicate
+multiclass T_pnp_CJInst_Rn1<string op>{
+ defm J4_cmp#NAME#n1_t : T_tnt_CJInst_Rn1<op, 0>;
+ defm J4_cmp#NAME#n1_f : T_tnt_CJInst_Rn1<op, 1>;
+}
+// TypeCJ Instructions compare -1 and jump
+let isCodeGenOnly = 0 in {
+defm eq : T_pnp_CJInst_Rn1<"eq">;
+defm gt : T_pnp_CJInst_Rn1<"gt">;
+}
+
+// J4_jumpseti: Direct unconditional jump and set register to immediate.
+let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1,
+ isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11,
+ opExtentAlign = 2, opExtendable = 2, validSubTargets = HasV4SubT,
+ isCodeGenOnly = 0 in
+def J4_jumpseti: CJInst <
+ (outs IntRegs:$Rd),
+ (ins u6Imm:$U6, brtarget:$r9_2),
+ "$Rd = #$U6 ; jump $r9_2"> {
+ bits<4> Rd;
+ bits<6> U6;
+ bits<11> r9_2;
+
+ let IClass = 0b0001;
+ let Inst{27-24} = 0b0110;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{19-16} = Rd;
+ let Inst{13-8} = U6;
+ let Inst{7-1} = r9_2{8-2};
+ }
+
+// J4_jumpsetr: Direct unconditional jump and transfer register.
+let Defs = [PC], isBranch = 1, hasSideEffects = 0, hasNewValue = 1,
+ isExtentSigned = 1, opNewValue = 0, isExtendable = 1, opExtentBits = 11,
+ opExtentAlign = 2, opExtendable = 2, validSubTargets = HasV4SubT,
+ isCodeGenOnly = 0 in
+def J4_jumpsetr: CJInst <
+ (outs IntRegs:$Rd),
+ (ins IntRegs:$Rs, brtarget:$r9_2),
+ "$Rd = $Rs ; jump $r9_2"> {
+ bits<4> Rd;
+ bits<4> Rs;
+ bits<11> r9_2;
+
+ let IClass = 0b0001;
+ let Inst{27-24} = 0b0111;
+ let Inst{21-20} = r9_2{10-9};
+ let Inst{11-8} = Rd;
+ let Inst{19-16} = Rs;
+ let Inst{7-1} = r9_2{8-2};
+ }