[TableGen] Add support for variable length instruction in decoder generator

[0x59616e]

Summary

To support variable length instructions, I think of them as fixed length instructions with the "maximum length". For example, if there're three instructions with 2, 6 and 9 bytes, we can fit them into the algorithm by treating them all as 9 bytes.

Also, since we can't know the length of the instruction in advance, there is a function object with type void(APInt &, uint64_t) added in the parameter list of decodeInstruction and fieldFromInstruction. We can use this to supply the additional bits the decoder needs after we know the opcode of the instruction.

Finally, InstrLenTable is added to let the decoder know the length of the instructions.

See D120960 for its usage.

[myhsu]

Thank you for the patch.
A high level question: It seems like generating a decoder requires two phases -- First, generates a list of OperandInfo, which is currently done by populateInstruction, then emit the real (C++) code according to these OperandInfo instances. My question is, can we use our own way -- possibly putting in a separate file -- to generate these OperandInfo before supplying them to the second phase, instead of trying to piggyback everything into the existing FixedLenDecoder framework?
Because, to be honest, I'm not a big fan of overloading the BitsInit (to carry operand info). Using a BitsInit might fit well for fixed-length instructions but I feel like there are more elegant ways to handle var-length instructions. For instance, using CGIOperandList::OperandInfo::ParseOperandName to parse suboperands rather than traversing every single in/out operands in the original instruction definition.

Also, please double check all your modifications and make sure to follow the LLVM coding style, especially the naming convention of local variables.

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

228

https://llvm.org/docs/CodingStandards.html#name-types-functions-variables-and-enumerators-properly

265

"unknown Init kind" would probably be better.

460

why do you want to change these three lines? I don't think RV is used anywhere else.

1347

ditto

2177–2183

The OI argument can be OperandInfo & to avoid copy.

2294

please use llvm::function_ref here

2297

why do we want to enlarge insn (on-demand) upon every field extractions? Can we resize insn ahead of time and do it only once?

2331

ditto

[0x59616e]

In D120958#3380177, @myhsu wrote:

Thank you for the patch.
A high level question: It seems like generating a decoder requires two phases -- First, generates a list of OperandInfo, which is currently done by populateInstruction, then emit the real (C++) code according to these OperandInfo instances. My question is, can we use our own way -- possibly putting in a separate file -- to generate these OperandInfo before supplying them to the second phase, instead of trying to piggyback everything into the existing FixedLenDecoder framework?
Because, to be honest, I'm not a big fan of overloading the BitsInit (to carry operand info). Using a BitsInit might fit well for fixed-length instructions but I feel like there are more elegant ways to handle var-length instructions. For instance, using CGIOperandList::OperandInfo::ParseOperandName to parse suboperands rather than traversing every single in/out operands in the original instruction definition.

Sorry for not having time implementing my ideas, so I just run through it quickly (I will do it maybe this Saturday or Sunday):

We can traverse the Segments of VarLenInst and use CGIOperandList::OperandInfo::ParseOperandName to get the operand number, and then add that Segments into the correspoding OpInfo according to the operand number.

For example, if we find "dst.reg" and use CGIOperandList::OperandInfo::ParseOperandName and know "Oh, this is the third operand", then we can add this into the third OpInfo.

In this way we can avoid overload BitsInit completely.

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

460

My understanding is: SFBits could be null pointer (take a look at the if statement below), which means SoftFail is not necessary. But using getValueAsBitsInit will trigger assertion if SoftFail can not be found.

[0x59616e]

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

1347

Same reason as above: SFBits is not necessary (if not then the if statement below is useless) but getValueAsBitsInit dislike that.

2297

In some cases, for example, case MCD::OPC_CheckField, needs to extract the bit field before we know the length of the instruction.

But only two cases need this IIRC: OPC_ExtractField and OPC_CheckField. Doing this in fieldFromInstruction is because I'm lazy. I will fix this in the next diff.

And, can we resize insn ahead of time and do it only once ? I think it's no. We have to know the length to resize insn. So we can do this only in case MCD::OPC_Decode.

[myhsu]

In D120958#3382237, @0x59616e wrote:

We can traverse the Segments of VarLenInst and use CGIOperandList::OperandInfo::ParseOperandName to get the operand number, and then add that Segments into the correspoding OpInfo according to the operand number.

For example, if we find "dst.reg" and use CGIOperandList::OperandInfo::ParseOperandName and know "Oh, this is the third operand", then we can add this into the third OpInfo.

In this way we can avoid overload BitsInit completely.

This approach sounds reasonable to me

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

460

The use of getValueAsBitsInit might be wrong but it's irrelevant to this patch. I think it's better to put this change into another patch.

[0x59616e]

In D120958#3383624, @myhsu wrote:

In D120958#3382237, @0x59616e wrote:

We can traverse the Segments of VarLenInst and use CGIOperandList::OperandInfo::ParseOperandName to get the operand number, and then add that Segments into the correspoding OpInfo according to the operand number.

For example, if we find "dst.reg" and use CGIOperandList::OperandInfo::ParseOperandName and know "Oh, this is the third operand", then we can add this into the third OpInfo.

In this way we can avoid overload BitsInit completely.

This approach sounds reasonable to me

Great ! I'll materialize it this weekend.

[0x59616e]

address feedback

[0x59616e]

This comment was removed by 0x59616e.

[0x59616e]

fix a minor problem

[myhsu]

The logics look cleaner now.
Now a bigger question is: Should we still calling it FixedLenDecoderEmitter?

llvm/test/TableGen/VarLenEncoder.td

3

I still prefer the decoder's tests either go under the FixedLenDecoderEmitter's folder or at least putting into a separate file.

51

I don't quite understand what this !cond trying to do

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

224

local variables should start with uppercase

1868

I'm not sure non-power-of-two # of (inlined) elements for SmallVector is recommended. If you're not sure what number to put, just use SmallVector<int> as recommended here.

1875

braces here can be removed

1900

please use !OpName.empty() or OpName.size()

1958

Did you use clang-format-diff.py rather than clang-format? I wonder why this line changed...

2362

why change to uint64_t? If you're worry about InsnType being APInt, APInt has operator=(uint64_t)

2389

ditto

2499

ditto

2504

"!= 0" in these two lines are redundant.

2543

why remove '&'?

2677

Could we add some comments here?

llvm/utils/TableGen/VarLenCodeEmitterGen.h

63

I think these forward declarations are obsolete now

[0x59616e]

address some feedbacks

[0x59616e]

move test into another file

[0x59616e]

In D120958#3415717, @myhsu wrote:

The logics look cleaner now.
Now a bigger question is: Should we still calling it FixedLenDecoderEmitter?

What about DecoderEmitter.cpp ?

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

224

Just wondering: there are some variables that is started with lowercase in the original code. For example the parameter "&def" in this function.

Is that a mistake ? or is there any special naming rule that I don't know ?

228

Naming is hard.

1958

I've tried, and it remains the same.

BTW I use "git clang-format". is there any difference between these two ?

2362

But it yells "conversion from ‘uint64_t’ {aka ‘long unsigned int’} to non-scalar type ‘llvm::APInt’ requested"

2504

g++ screams "no match for ‘operator||’ (operand types are ‘llvm::APInt’ and ‘llvm::APInt’)"

[0x59616e]

address feedback

[myhsu]

In D120958#3416426, @0x59616e wrote:

In D120958#3415717, @myhsu wrote:

The logics look cleaner now.
Now a bigger question is: Should we still calling it FixedLenDecoderEmitter?

What about DecoderEmitter.cpp ?

Sounds good to me, but I would suggest you to send a RFC to the forum asking for consensus.
Especially to see if there is any comment from existing FixedLenDecoder users (e.g. AArch64 and ARM). Also, in the RFC, please briefly explain why you want to build this feature on top of FixedLenDecoder rather than writing a separate disassembler.

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

224

it's likely that those code were committed before LLVM coding style was solidified

1958

I've tried, and it remains the same.

that's fine, just a minor issue

BTW I use "git clang-format". is there any difference between these two ?

I don't think so

2362

you're right and I was wrong about APInt::operator=(uint64_t): the operator won't be used in the case of initialization (constructor will be used instead but APInt doesn't have APInt(uint64_t)).

2389

why did you use another PtrLen variable here?

[myhsu]

Also, please make sure all existing disassembler tests, especially targets that use FixedLenDecoder, are passing.

[myhsu]

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

1875

I meant only braces around the for-loop, your current syntax might create ambiguity. Here is the guideline: https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

[0x59616e]

address feedback

[0x59616e]

In D120958#3419921, @myhsu wrote:

In D120958#3416426, @0x59616e wrote:

In D120958#3415717, @myhsu wrote:

The logics look cleaner now.
Now a bigger question is: Should we still calling it FixedLenDecoderEmitter?

What about DecoderEmitter.cpp ?

Sounds good to me, but I would suggest you to send a RFC to the forum asking for consensus.
Especially to see if there is any comment from existing FixedLenDecoder users (e.g. AArch64 and ARM). Also, in the RFC, please briefly explain why you want to build this feature on top of FixedLenDecoder rather than writing a separate disassembler.

No problem.

In D120958#3419939, @myhsu wrote:

Also, please make sure all existing disassembler tests, especially targets that use FixedLenDecoder, are passing.

Sure.

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

224

Ah, that makes sense.

2362

C++ is hard.

2389

quick answer: for debug purpose.

In the original code, Len is the length of the instruction field. But it is changed in the function `decodeULEB128'. That will break the debug message below.

[0x59616e]

Should we rename after this patch ? or before this patch ?

[0x59616e]

rebase

[0x59616e]

I could not remember whether I've modified this patch recently.

So I update this just for ensurance.

[0x59616e]

ping. Any more suggestions ?

[RKSimon]

@myhsu Any more comments?

[myhsu]

lgtm. Thanks for the patch!

[ricky26]

It's taken me a very long time to find time to read though this patch (I'm very sorry about that). The approach seems nice and straight forward & the generated code looks good. Thanks a lot for finding the time to get this through. 😅

[0x59616e]

In D120958#3484756, @myhsu wrote:

lgtm. Thanks for the patch!

Thanks for your patience. This patch wouldn't be here if it were not for your help ;)

In D120958#3484851, @ricky26 wrote:

It's taken me a very long time to find time to read though this patch (I'm very sorry about that). The approach seems nice and straight forward & the generated code looks good. Thanks a lot for finding the time to get this through. 😅

My pleasure ;)

[foad]

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

2362

Hi, this is causing a slight problem for us downstream because we're using a custom InsnType (not APInt). Have you seen the big comment emitted at line 2255? It explicitly says that your InsnType needs to "be constructible from a uint64_t". So please either update the comment, or stop using plain APInt as your InsnType :)

2504

Downstream the compiler complained about using mismatched types for operator&, InsnType and uint64_t. So please either change this back or at least update the comment on line 2255.

[foad]

For the record, to get this to build with my downstream target, I had to add:

MyInsnType MyInsnType::operator&(const uint64_t &RHS);
bool MyInsnType::operator!=(const int &RHS);

[gregmiller]

Hello, We are maintaining a downstream version of the monorepo based on the LLVM main branch. In a recent attempt to merge the latest upstream commits
into our monorepo we came across the following test failures after your commit.
Any help would be greatly appreciated.
Thanks
Greg

---

`

FAIL: llvm_regressions :: LLVM/TableGen/VarLenDecoder.td
--------------------------------------------------------------------------------
Script:
--
: 'RUN: at line 1';   /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/llvm-tblgen -gen-disassembler -I /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/../../include /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td | /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/FileCheck /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
--
Exit Code: 1

Command Output (stderr):
--
+ : 'RUN: at line 1'
+ /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/llvm-tblgen -gen-disassembler -I /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/../../include /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
+ /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/FileCheck /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
/scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td:50:16: error: CHECK-NEXT: expected string not found in input
// CHECK-NEXT: MCD::OPC_Decode, 244, 1, 0, // Opcode: FOO16
               ^
<stdin>:72:57: note: scanning from here
/* 3 */ MCD::OPC_FilterValue, 8, 4, 0, 0, // Skip to: 12
                                                        ^
<stdin>:73:9: note: possible intended match here
/* 8 */ MCD::OPC_Decode, 245, 1, 0, // Opcode: FOO16
        ^

Input file: <stdin>
Check file: /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td

-dump-input=help explains the following input dump.

Input was:
<<<<<<
           .
           .
           .
          67:  field.insertBits(bits, startBit, numBits); 
          68: } 
          69:  
          70: static const uint8_t DecoderTable43[] = { 
          71: /* 0 */ MCD::OPC_ExtractField, 3, 5, // Inst{7-3} ... 
          72: /* 3 */ MCD::OPC_FilterValue, 8, 4, 0, 0, // Skip to: 12 
next:50'0                                                             X error: no match found
          73: /* 8 */ MCD::OPC_Decode, 245, 1, 0, // Opcode: FOO16 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
next:50'1             ?                                             possible intended match
          74: /* 12 */ MCD::OPC_FilterValue, 9, 4, 0, 0, // Skip to: 21 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          75: /* 17 */ MCD::OPC_Decode, 246, 1, 1, // Opcode: FOO32 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          76: /* 21 */ MCD::OPC_Fail, 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~
          77:  0 
next:50'0     ~~~
          78: }; 
next:50'0     ~~~
           .
           .
           .
>>>>>>

--
`

[myhsu]

In D120958#3488799, @gregmiller wrote:

Hello, We are maintaining a downstream version of the monorepo based on the LLVM main branch. In a recent attempt to merge the latest upstream commits
into our monorepo we came across the following test failures after your commit.
Any help would be greatly appreciated.
Thanks
Greg

---

`

FAIL: llvm_regressions :: LLVM/TableGen/VarLenDecoder.td
--------------------------------------------------------------------------------
Script:
--
: 'RUN: at line 1';   /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/llvm-tblgen -gen-disassembler -I /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/../../include /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td | /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/FileCheck /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
--
Exit Code: 1

Command Output (stderr):
--
+ : 'RUN: at line 1'
+ /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/llvm-tblgen -gen-disassembler -I /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/../../include /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
+ /scratch/gmiller/tools2/llvm_cgt/arm-llvm/RelWithAsserts/llvm/bin/FileCheck /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td
/scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td:50:16: error: CHECK-NEXT: expected string not found in input
// CHECK-NEXT: MCD::OPC_Decode, 244, 1, 0, // Opcode: FOO16
               ^
<stdin>:72:57: note: scanning from here
/* 3 */ MCD::OPC_FilterValue, 8, 4, 0, 0, // Skip to: 12
                                                        ^
<stdin>:73:9: note: possible intended match here
/* 8 */ MCD::OPC_Decode, 245, 1, 0, // Opcode: FOO16
        ^

Input file: <stdin>
Check file: /scratch/gmiller/tools2/llvm_cgt/llvm-project/llvm/test/TableGen/VarLenDecoder.td

-dump-input=help explains the following input dump.

Input was:
<<<<<<
           .
           .
           .
          67:  field.insertBits(bits, startBit, numBits); 
          68: } 
          69:  
          70: static const uint8_t DecoderTable43[] = { 
          71: /* 0 */ MCD::OPC_ExtractField, 3, 5, // Inst{7-3} ... 
          72: /* 3 */ MCD::OPC_FilterValue, 8, 4, 0, 0, // Skip to: 12 
next:50'0                                                             X error: no match found
          73: /* 8 */ MCD::OPC_Decode, 245, 1, 0, // Opcode: FOO16 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
next:50'1             ?                                             possible intended match
          74: /* 12 */ MCD::OPC_FilterValue, 9, 4, 0, 0, // Skip to: 21 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          75: /* 17 */ MCD::OPC_Decode, 246, 1, 1, // Opcode: FOO32 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          76: /* 21 */ MCD::OPC_Fail, 
next:50'0     ~~~~~~~~~~~~~~~~~~~~~~~~
          77:  0 
next:50'0     ~~~
          78: }; 
next:50'0     ~~~
           .
           .
           .
>>>>>>

--
`

Looks like the opcode of FOO16 and FOO32 are off by one. I think the number of pseudo opcodes is different (from the upstream one). Are include/llvm/Support/TargetOpcodes.def or include/llvm/Target/Target.td in your downstream repo different from upstream?

[0x59616e]

In D120958#3487811, @foad wrote:

For the record, to get this to build with my downstream target, I had to add:

MyInsnType MyInsnType::operator&(const uint64_t &RHS);
bool MyInsnType::operator!=(const int &RHS);

Thanks for your information. I'll update the comment.

[0x59616e]

In D120958#3490184, @0x59616e wrote:

In D120958#3487811, @foad wrote:

For the record, to get this to build with my downstream target, I had to add:

MyInsnType MyInsnType::operator&(const uint64_t &RHS);
bool MyInsnType::operator!=(const int &RHS);

Thanks for your information. I'll update the comment.

@myhsu Should we define our own inst type that wraps APInt, or update the comment and maintain the status quo ?

[snidertm]

llvm/test/TableGen/VarLenDecoder.td

50 ↗

(On Diff #426489)

Is 244 the index into the instruction table where FOO16 is expected to reside?

This check on OPC_Decode seems extremely brittle. If there are any additions either in the upstream Target.td or in a downstream Target.td, won't this check and the one on line 52 will break?

[0x59616e]

I've commited a diff in the hope of fixing the test problem:

https://github.com/llvm/llvm-project/commit/9c2121b843ff7c9846a89305b4b73e3c480fe4e7

[0x59616e]

The comment is updated :
1284ce917b5a

[foad]

In D120958#3502485, @0x59616e wrote:

The comment is updated :
1284ce917b5a

Thanks!

llvm/test/TableGen/VarLenEncoder.td

 // RUN: llvm-tblgen -gen-emitter -I %p/../../include %s | FileCheck %s
+// RUN: llvm-tblgen -gen-disassembler -I %p/../../include %s | \
+// RUN:     FileCheck --check-prefix=CHECK-DECODE %s
 
 // Check if VarLenCodeEmitterGen works correctly.
 
 include "llvm/Target/Target.td"
 
 def ArchInstrInfo : InstrInfo { }
 
 def Arch : Target {
   let InstructionSet = ArchInstrInfo;
 }
 
 def Reg : Register<"reg">;
 
 def RegClass : RegisterClass<"foo", [i64], 0, (add Reg)>;
 
 def GR64 : RegisterOperand<RegClass>;
 
 class MyMemOperand<dag sub_ops> : Operand<iPTR> {
   let MIOperandInfo = sub_ops;
   dag Base;
   dag Extension;
 }
 
+class MemOp16<string op_name> : MyMemOperand<(ops GR64:$reg, i16imm:$offset)> {
+  // Testing sub-operand referencing.
+  let Base = (operand "$"#op_name#".reg", 8);
+  let Extension = (operand "$"#op_name#".offset", 16);
+}
+
+class MemOp32<string op_name> : MyMemOperand<(ops GR64:$reg, i32imm:$offset)> {
+  let Base = (operand "$"#op_name#".reg", 8);
+  // Testing variable-length instruction encoding.
+  let Extension = (operand "$"#op_name#".offset", 32);
+}
+
 class MyVarInst<MyMemOperand memory_op> : Instruction {
   dag Inst;
 
   let OutOperandList = (outs GR64:$dst);
   let InOperandList  = (ins memory_op:$src);
 
   // Testing `ascend` and `descend`
   let Inst = (ascend
-    (descend 0b10110111, memory_op.Base),
+    (descend 0b10110111, memory_op.Base,
+      !cond(
+        !isa<MemOp16>(memory_op) : 0b0,
+        !isa<MemOp32>(memory_op) : 0b1,
+      )
+    ),
     memory_op.Extension,
     // Testing operand referencing.
     (operand "$dst", 4),
     // Testing operand referencing with a certain bit range.
     (slice "$dst", 3, 1),
     // Testing custom encoder
     (operand "$dst", 2, (encoder "myCustomEncoder"))
   );
 }
 
-class MemOp16<string op_name> : MyMemOperand<(ops GR64:$reg, i16imm:$offset)> {
-  // Testing sub-operand referencing.
-  let Base = (operand "$"#op_name#".reg", 8);
-  let Extension = (operand "$"#op_name#".offset", 16);
-}
-
-class MemOp32<string op_name> : MyMemOperand<(ops GR64:$reg, i32imm:$offset)> {
-  let Base = (operand "$"#op_name#".reg", 8);
-  // Testing variable-length instruction encoding.
-  let Extension = (operand "$"#op_name#".offset", 32);
-}
-
 def FOO16 : MyVarInst<MemOp16<"src">>;
 def FOO32 : MyVarInst<MemOp32<"src">>;
 
 // The fixed bits part
-// CHECK: {/*NumBits*/41,
+// CHECK: {/*NumBits*/42,
 // CHECK-SAME: // FOO16
-// CHECK: {/*NumBits*/57,
+// CHECK: {/*NumBits*/58,
 // CHECK-SAME: // FOO32
-// CHECK: UINT64_C(46848), // FOO16
-// CHECK: UINT64_C(46848), // FOO32
+// CHECK: UINT64_C(93696), // FOO16
+// CHECK: UINT64_C(93697), // FOO32
 
 // CHECK-LABEL: case ::FOO16: {
-// CHECK: Scratch = Scratch.zextOrSelf(41);
+// CHECK: Scratch = Scratch.zextOrSelf(42);
 // src.reg
-// CHECK: getMachineOpValue(MI, MI.getOperand(1), /*Pos=*/0, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(8, 0), 0);
+// CHECK: getMachineOpValue(MI, MI.getOperand(1), /*Pos=*/1, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(8, 0), 1);
 // src.offset
-// CHECK: getMachineOpValue(MI, MI.getOperand(2), /*Pos=*/16, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(16, 0), 16);
+// CHECK: getMachineOpValue(MI, MI.getOperand(2), /*Pos=*/17, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(16, 0), 17);
 // 1st dst
-// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/32, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(4, 0), 32);
+// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/33, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(4, 0), 33);
 // 2nd dst
-// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/36, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(3, 1), 36);
+// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/37, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(3, 1), 37);
 // dst w/ custom encoder
-// CHECK: myCustomEncoder(MI, /*OpIdx=*/0, /*Pos=*/39, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(2, 0), 39);
+// CHECK: myCustomEncoder(MI, /*OpIdx=*/0, /*Pos=*/40, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(2, 0), 40);
 
 // CHECK-LABEL: case ::FOO32: {
-// CHECK: Scratch = Scratch.zextOrSelf(57);
+// CHECK: Scratch = Scratch.zextOrSelf(58);
 // src.reg
-// CHECK: getMachineOpValue(MI, MI.getOperand(1), /*Pos=*/0, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(8, 0), 0);
+// CHECK: getMachineOpValue(MI, MI.getOperand(1), /*Pos=*/1, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(8, 0), 1);
 // src.offset
-// CHECK: getMachineOpValue(MI, MI.getOperand(2), /*Pos=*/16, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(32, 0), 16);
+// CHECK: getMachineOpValue(MI, MI.getOperand(2), /*Pos=*/17, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(32, 0), 17);
 // 1st dst
-// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/48, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(4, 0), 48);
+// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/49, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(4, 0), 49);
 // 2nd dst
-// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/52, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(3, 1), 52);
+// CHECK: getMachineOpValue(MI, MI.getOperand(0), /*Pos=*/53, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(3, 1), 53);
 // dst w/ custom encoder
-// CHECK: myCustomEncoder(MI, /*OpIdx=*/0, /*Pos=*/55, Scratch, Fixups, STI);
-// CHECK: Inst.insertBits(Scratch.extractBits(2, 0), 55);
+// CHECK: myCustomEncoder(MI, /*OpIdx=*/0, /*Pos=*/56, Scratch, Fixups, STI);
+// CHECK: Inst.insertBits(Scratch.extractBits(2, 0), 56);
+
+// CHECK-DECODE: MCD::OPC_ExtractField, 0, 1,  // Inst{0} ...
+// CHECK-DECODE-NEXT: MCD::OPC_FilterValue, 0, 12, 0, 0, // Skip to: 20
+// CHECK-DECODE-NEXT: MCD::OPC_CheckField, 9, 8, 183, 1, 21, 0, 0, // Skip to: 37
+// CHECK-DECODE-NEXT: MCD::OPC_Decode, 244, 1, 0, // Opcode: FOO16
+// CHECK-DECODE-NEXT: MCD::OPC_FilterValue, 1, 12, 0, 0, // Skip to: 37
+// CHECK-DECODE-NEXT: MCD::OPC_CheckField, 9, 8, 183, 1, 4, 0, 0, // Skip to: 37
+// CHECK-DECODE-NEXT: MCD::OPC_Decode, 245, 1, 1, // Opcode: FOO32
+// CHECK-DECODE-NEXT: MCD::OPC_Fail,
+
+// Instruction length table
+// CHECK-DECODE: 42,
+// CHECK-DECODE-NEXT: 58,
+// CHECK-DECODE-NEXT: };
+
+// CHECK-DECODE-LABEL: case 0:
+// CHECK-DECODE: tmp = 0x0;
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 33, 4), 0, 4);
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 37, 3), 1, 3);
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 40, 2), 0, 2);
+// CHECK-DECODE-NEXT: if (DecodeRegClassRegisterClass(MI, tmp, Address, Decoder) == MCDisassembler::Fail) { return MCDisassembler::Fail; }
+// CHECK-DECODE-NEXT: tmp = fieldFromInstruction(insn, 1, 8);
+// CHECK-DECODE-NEXT: if (DecodeRegClassRegisterClass(MI, tmp, Address, Decoder) == MCDisassembler::Fail) { return MCDisassembler::Fail; }
+// CHECK-DECODE-NEXT: tmp = fieldFromInstruction(insn, 17, 16);
+// CHECK-DECODE-NEXT: MI.addOperand(MCOperand::createImm(tmp));
+// CHECK-DECODE-NEXT: return S;
+// CHECK-DECODE-NEXT: case 1:
+// CHECK-DECODE-NEXT: tmp = 0x0;
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 49, 4), 0, 4);
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 53, 3), 1, 3);
+// CHECK-DECODE-NEXT: insertBits(tmp, fieldFromInstruction(insn, 56, 2), 0, 2);
+// CHECK-DECODE-NEXT: if (DecodeRegClassRegisterClass(MI, tmp, Address, Decoder) == MCDisassembler::Fail) { return MCDisassembler::Fail; }
+// CHECK-DECODE-NEXT: tmp = fieldFromInstruction(insn, 1, 8);
+// CHECK-DECODE-NEXT: if (DecodeRegClassRegisterClass(MI, tmp, Address, Decoder) == MCDisassembler::Fail) { return MCDisassembler::Fail; }
+// CHECK-DECODE-NEXT: tmp = fieldFromInstruction(insn, 17, 32);
+// CHECK-DECODE-NEXT: MI.addOperand(MCOperand::createImm(tmp));
+// CHECK-DECODE-NEXT: return S;
+
+// CHECK-DECODE-LABEL: case MCD::OPC_ExtractField: {
+// CHECK-DECODE: makeUp(insn, Start + Len);
+
+// CHECK-DECODE-LABEL: case MCD::OPC_CheckField: {
+// CHECK-DECODE: makeUp(insn, Start + Len);
+
+// CHECK-DECODE-LABEL: case MCD::OPC_Decode: {
+// CHECK-DECODE: Len = InstrLenTable[Opc];
+// CHECK-DECODE-NEXT: makeUp(insn, Len);

llvm/utils/TableGen/FixedLenDecoderEmitter.cpp

 //===------------ FixedLenDecoderEmitter.cpp - Decoder Generator ----------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // It contains the tablegen backend that emits the decoder functions for
 // targets with fixed length instruction set.
 //
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenInstruction.h"
 #include "CodeGenTarget.h"
 #include "InfoByHwMode.h"
+#include "VarLenCodeEmitterGen.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/CachedHashString.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringExtras.h"
FixedLenDecoderEmitter(RecordKeeper &R, std::string PredicateNamespace,
         GuardPrefix(std::move(GPrefix)), GuardPostfix(std::move(GPostfix)),
         ReturnOK(std::move(ROK)), ReturnFail(std::move(RFail)),
         Locals(std::move(L)) {}
 
   // Emit the decoder state machine table.
   void emitTable(formatted_raw_ostream &o, DecoderTable &Table,
                  unsigned Indentation, unsigned BitWidth,
                  StringRef Namespace) const;
+  void emitInstrLenTable(formatted_raw_ostream &OS,
+                         std::vector<unsigned> &InstrLen) const;
   void emitPredicateFunction(formatted_raw_ostream &OS,
                              PredicateSet &Predicates,
                              unsigned Indentation) const;
   void emitDecoderFunction(formatted_raw_ostream &OS,
                            DecoderSet &Decoders,
                            unsigned Indentation) const;
 
   // run - Output the code emitter
case BIT_UNSET:
       break;
     default:
       llvm_unreachable("unexpected return value from bitFromBits");
     }
   }
 }
 
 static BitsInit &getBitsField(const Record &def, StringRef str) {
-  BitsInit *bits = def.getValueAsBitsInit(str);
+  const RecordVal *RV = def.getValue(str);
+  if (BitsInit *bits = dyn_cast<BitsInit>(RV->getValue()))
     return *bits;
+
+  // variable length instruction
+  DagInit *DI = cast<DagInit>(RV->getValue());
+  VarLenInst VLI = VarLenInst(DI, RV);
+  SmallVector<Init *, 16> Bits;
+
+  for (auto &SI : VLI) {
+    if (const BitsInit *BI = dyn_cast<BitsInit>(SI.Value)) {
+      for (unsigned Idx = 0U; Idx < BI->getNumBits(); ++Idx) {
+        Bits.push_back(BI->getBit(Idx));
+      }
+    } else if (const BitInit *BI = dyn_cast<BitInit>(SI.Value)) {
+      Bits.push_back(const_cast<BitInit *>(BI));
+    } else {
+      for (unsigned Idx = 0U; Idx < SI.BitWidth; ++Idx)
+        Bits.push_back(UnsetInit::get());
+    }
+  }
+
+  return *BitsInit::get(Bits);
 }
 
 // Representation of the instruction to work on.
 typedef std::vector<bit_value_t> insn_t;
 
 namespace {
 
 static const uint64_t NO_FIXED_SEGMENTS_SENTINEL = -1ULL;
 
 class FilterChooser;
 
 /// Filter - Filter works with FilterChooser to produce the decoding tree for
 /// the ISA.
 ///
 /// It is useful to think of a Filter as governing the switch stmts of the
 /// decoding tree in a certain level.  Each case stmt delegates to an inferior
 /// FilterChooser to decide what further decoding logic to employ, or in another
 /// words, what other remaining bits to look at.  The FilterChooser eventually
 /// chooses a best Filter to do its job.
 ///
 /// This recursive scheme ends when the number of Opcodes assigned to the
 /// FilterChooser becomes 1 or if there is a conflict.  A conflict happens when
 /// the Filter/FilterChooser combo does not know how to distinguish among the
 /// Opcodes assigned.
 ///
 /// An example of a conflict is
 ///
 /// Conflict:
public:
   void operator=(const FilterChooser &) = delete;
 
   unsigned getBitWidth() const { return BitWidth; }
 
 protected:
   // Populates the insn given the uid.
   void insnWithID(insn_t &Insn, unsigned Opcode) const {
     BitsInit &Bits = getBitsField(*AllInstructions[Opcode].EncodingDef, "Inst");
-
+    Insn.resize(BitWidth > Bits.getNumBits() ? BitWidth : Bits.getNumBits(),
+                BIT_UNSET);
     // We may have a SoftFail bitmask, which specifies a mask where an encoding
     // may differ from the value in "Inst" and yet still be valid, but the
     // disassembler should return SoftFail instead of Success.
     //
     // This is used for marking UNPREDICTABLE instructions in the ARM world.
     const RecordVal *RV =
         AllInstructions[Opcode].EncodingDef->getValue("SoftFail");
     const BitsInit *SFBits = RV ? dyn_cast<BitsInit>(RV->getValue()) : nullptr;
-    for (unsigned i = 0; i < BitWidth; ++i) {
+    for (unsigned i = 0; i < Bits.getNumBits(); ++i) {
       if (SFBits && bitFromBits(*SFBits, i) == BIT_TRUE)
-        Insn.push_back(BIT_UNSET);
+        Insn[i] = BIT_UNSET;
       else
-        Insn.push_back(bitFromBits(Bits, i));
+        Insn[i] = bitFromBits(Bits, i);
     }
   }
 
   // Emit the name of the encoding/instruction pair.
   void emitNameWithID(raw_ostream &OS, unsigned Opcode) const {
     const Record *EncodingDef = AllInstructions[Opcode].EncodingDef;
     const Record *InstDef = AllInstructions[Opcode].Inst->TheDef;
     if (EncodingDef != InstDef)
void FixedLenDecoderEmitter::emitTable(formatted_raw_ostream &OS,
   }
   OS.indent(Indentation) << "0\n";
 
   Indentation -= 2;
 
   OS.indent(Indentation) << "};\n\n";
 }
 
+void FixedLenDecoderEmitter::emitInstrLenTable(
+    formatted_raw_ostream &OS, std::vector<unsigned> &InstrLen) const {
+  OS << "static const uint8_t InstrLenTable[] = {\n";
+  for (unsigned &Len : InstrLen) {
+    OS << Len << ",\n";
+  }
+  OS << "};\n\n";
+}
+
 void FixedLenDecoderEmitter::
 emitPredicateFunction(formatted_raw_ostream &OS, PredicateSet &Predicates,
                       unsigned Indentation) const {
   // The predicate function is just a big switch statement based on the
   // input predicate index.
   OS.indent(Indentation) << "static bool checkDecoderPredicate(unsigned Idx, "
     << "const FeatureBitset &Bits) {\n";
   Indentation += 2;
void FilterChooser::emitPredicateTableEntry(DecoderTableInfo &TableInfo,
   TableInfo.Table.push_back(0);
   TableInfo.Table.push_back(0);
   TableInfo.Table.push_back(0);
 }
 
 void FilterChooser::emitSoftFailTableEntry(DecoderTableInfo &TableInfo,
                                            unsigned Opc) const {
   const RecordVal *RV = AllInstructions[Opc].EncodingDef->getValue("SoftFail");
   BitsInit *SFBits = RV ? dyn_cast<BitsInit>(RV->getValue()) : nullptr;
 
   if (!SFBits) return;
   BitsInit *InstBits =
       AllInstructions[Opc].EncodingDef->getValueAsBitsInit("Inst");
 
   APInt PositiveMask(BitWidth, 0ULL);
   APInt NegativeMask(BitWidth, 0ULL);
   for (unsigned i = 0; i < BitWidth; ++i) {
for (auto Opcode : Opcodes) {
     errs() << " ";
     dumpBits(
         errs(),
         getBitsField(*AllInstructions[Opcode.EncodingID].EncodingDef, "Inst"));
     errs() << '\n';
   }
 }
 
-static std::string findOperandDecoderMethod(TypedInit *TI) {
+static std::string findOperandDecoderMethod(Record *Record) {
   std::string Decoder;
 
-  Record *Record = cast<DefInit>(TI)->getDef();
-
   RecordVal *DecoderString = Record->getValue("DecoderMethod");
   StringInit *String = DecoderString ?
     dyn_cast<StringInit>(DecoderString->getValue()) : nullptr;
   if (String) {
     Decoder = std::string(String->getValue());
     if (!Decoder.empty())
       return Decoder;
   }
 
   if (Record->isSubClassOf("RegisterOperand"))
     Record = Record->getValueAsDef("RegClass");
 
   if (Record->isSubClassOf("RegisterClass")) {
     Decoder = "Decode" + Record->getName().str() + "RegisterClass";
   } else if (Record->isSubClassOf("PointerLikeRegClass")) {
     Decoder = "DecodePointerLikeRegClass" +
       utostr(Record->getValueAsInt("RegClassKind"));
   }
 
   return Decoder;
 }
 
-static bool
+OperandInfo getOpInfo(Record *TypeRecord) {
+  std::string Decoder = findOperandDecoderMethod(TypeRecord);
+
+  RecordVal *HasCompleteDecoderVal = TypeRecord->getValue("hasCompleteDecoder");
+  BitInit *HasCompleteDecoderBit =
+      HasCompleteDecoderVal
+          ? dyn_cast<BitInit>(HasCompleteDecoderVal->getValue())
+          : nullptr;
+  bool HasCompleteDecoder =
+      HasCompleteDecoderBit ? HasCompleteDecoderBit->getValue() : true;
+
+  return OperandInfo(Decoder, HasCompleteDecoder);
+}
+
+void parseVarLenInstOperand(const Record &Def,
+                            std::vector<OperandInfo> &Operands,
+                            const CodeGenInstruction &CGI) {
+
+  const RecordVal *RV = Def.getValue("Inst");
+  VarLenInst VLI(cast<DagInit>(RV->getValue()), RV);
+  SmallVector<int, 3> TiedTo;
+
+  for (unsigned Idx = 0; Idx < CGI.Operands.size(); ++Idx) {
+    auto &Op = CGI.Operands[Idx];
+    if (Op.MIOperandInfo && Op.MIOperandInfo->getNumArgs() > 0) {
+      for (auto *Arg : Op.MIOperandInfo->getArgs()) {
+        Operands.push_back(getOpInfo(cast<DefInit>(Arg)->getDef()));
+      }
+    } else {
+      Operands.push_back(getOpInfo(Op.Rec));
+    }
+
+    int TiedReg = Op.getTiedRegister();
+    TiedTo.push_back(-1);
+    if (TiedReg != -1) {
+      TiedTo[Idx] = TiedReg;
+      TiedTo[TiedReg] = Idx;
+    }
+  }
+
+  unsigned CurrBitPos = 0;
+  for (auto &EncodingSegment : VLI) {
+    unsigned Offset = 0;
+    StringRef OpName;
+
+    if (const StringInit *SI = dyn_cast<StringInit>(EncodingSegment.Value)) {
+      OpName = SI->getValue();
+    } else if (const DagInit *DI = dyn_cast<DagInit>(EncodingSegment.Value)) {
+      OpName = cast<StringInit>(DI->getArg(0))->getValue();
+      Offset = cast<IntInit>(DI->getArg(2))->getValue();
+    }
+
+    if (OpName != "") {
+      auto OpSubOpPair =
+          const_cast<CodeGenInstruction &>(CGI).Operands.ParseOperandName(
+              OpName);
+      unsigned OpIdx = CGI.Operands.getFlattenedOperandNumber(OpSubOpPair);
+      Operands[OpIdx].addField(CurrBitPos, EncodingSegment.BitWidth, Offset);
+
+      int TiedReg = TiedTo[OpSubOpPair.first];
+      if (TiedReg != -1) {
+        unsigned OpIdx = CGI.Operands.getFlattenedOperandNumber(
+            std::make_pair(TiedReg, OpSubOpPair.second));
+        Operands[OpIdx].addField(CurrBitPos, EncodingSegment.BitWidth, Offset);
+      }
+    }
+
+    CurrBitPos += EncodingSegment.BitWidth;
+  }
+}
+
+static unsigned
 populateInstruction(CodeGenTarget &Target, const Record &EncodingDef,
                     const CodeGenInstruction &CGI, unsigned Opc,
-                    std::map<unsigned, std::vector<OperandInfo>> &Operands) {
+                    std::map<unsigned, std::vector<OperandInfo>> &Operands,
+                    bool IsVarLenInst) {
   const Record &Def = *CGI.TheDef;
   // If all the bit positions are not specified; do not decode this instruction.
   // We are bound to fail!  For proper disassembly, the well-known encoding bits
   // of the instruction must be fully specified.
 
   BitsInit &Bits = getBitsField(EncodingDef, "Inst");
-  if (Bits.allInComplete()) return false;
+  if (Bits.allInComplete())
+    return 0;
 
   std::vector<OperandInfo> InsnOperands;
 
   // If the instruction has specified a custom decoding hook, use that instead
   // of trying to auto-generate the decoder.
   StringRef InstDecoder = EncodingDef.getValueAsString("DecoderMethod");
   if (InstDecoder != "") {
     bool HasCompleteInstDecoder = EncodingDef.getValueAsBit("hasCompleteDecoder");
     InsnOperands.push_back(
         OperandInfo(std::string(InstDecoder), HasCompleteInstDecoder));
     Operands[Opc] = InsnOperands;
-    return true;
+    return Bits.getNumBits();
   }
 
   // Generate a description of the operand of the instruction that we know
   // how to decode automatically.
   // FIXME: We'll need to have a way to manually override this as needed.
 
   // Gather the outputs/inputs of the instruction, so we can find their
   // positions in the encoding.  This assumes for now that they appear in the
   // MCInst in the order that they're listed.
   std::vector<std::pair<Init*, StringRef>> InOutOperands;
   DagInit *Out  = Def.getValueAsDag("OutOperandList");
   DagInit *In  = Def.getValueAsDag("InOperandList");
   for (unsigned i = 0; i < Out->getNumArgs(); ++i)
-    InOutOperands.push_back(std::make_pair(Out->getArg(i),
-                                           Out->getArgNameStr(i)));
+    InOutOperands.push_back(
+        std::make_pair(Out->getArg(i), Out->getArgNameStr(i)));
   for (unsigned i = 0; i < In->getNumArgs(); ++i)
-    InOutOperands.push_back(std::make_pair(In->getArg(i),
-                                           In->getArgNameStr(i)));
+    InOutOperands.push_back(
+        std::make_pair(In->getArg(i), In->getArgNameStr(i)));
 
   // Search for tied operands, so that we can correctly instantiate
   // operands that are not explicitly represented in the encoding.
   std::map<std::string, std::string> TiedNames;
   for (unsigned i = 0; i < CGI.Operands.size(); ++i) {
     int tiedTo = CGI.Operands[i].getTiedRegister();
     if (tiedTo != -1) {
       std::pair<unsigned, unsigned> SO =
         CGI.Operands.getSubOperandNumber(tiedTo);
       TiedNames[std::string(InOutOperands[i].second)] =
           std::string(InOutOperands[SO.first].second);
       TiedNames[std::string(InOutOperands[SO.first].second)] =
           std::string(InOutOperands[i].second);
     }
   }
 
+  if (IsVarLenInst) {
+    parseVarLenInstOperand(EncodingDef, InsnOperands, CGI);
+  } else {
     std::map<std::string, std::vector<OperandInfo>> NumberedInsnOperands;
     std::set<std::string> NumberedInsnOperandsNoTie;
-  if (Target.getInstructionSet()->
-        getValueAsBit("decodePositionallyEncodedOperands")) {
+    if (Target.getInstructionSet()->getValueAsBit(
+            "decodePositionallyEncodedOperands")) {
       const std::vector<RecordVal> &Vals = Def.getValues();
       unsigned NumberedOp = 0;
 
       std::set<unsigned> NamedOpIndices;
-    if (Target.getInstructionSet()->
-         getValueAsBit("noNamedPositionallyEncodedOperands"))
+      if (Target.getInstructionSet()->getValueAsBit(
+              "noNamedPositionallyEncodedOperands"))
         // Collect the set of operand indices that might correspond to named
         // operand, and skip these when assigning operands based on position.
         for (unsigned i = 0, e = Vals.size(); i != e; ++i) {
           unsigned OpIdx;
           if (!CGI.Operands.hasOperandNamed(Vals[i].getName(), OpIdx))
             continue;
 
           NamedOpIndices.insert(OpIdx);
         }
 
       for (unsigned i = 0, e = Vals.size(); i != e; ++i) {
         // Ignore fixed fields in the record, we're looking for values like:
         //    bits<5> RST = { ?, ?, ?, ?, ? };
         if (Vals[i].isNonconcreteOK() || Vals[i].getValue()->isComplete())
           continue;
 
         // Determine if Vals[i] actually contributes to the Inst encoding.
         unsigned bi = 0;
         for (; bi < Bits.getNumBits(); ++bi) {
           VarInit *Var = nullptr;
           VarBitInit *BI = dyn_cast<VarBitInit>(Bits.getBit(bi));
           if (BI)
             Var = dyn_cast<VarInit>(BI->getBitVar());
           else
             Var = dyn_cast<VarInit>(Bits.getBit(bi));
 
           if (Var && Var->getName() == Vals[i].getName())
             break;
         }
 
         if (bi == Bits.getNumBits())
           continue;
 
         // Skip variables that correspond to explicitly-named operands.
         unsigned OpIdx;
         if (CGI.Operands.hasOperandNamed(Vals[i].getName(), OpIdx))
           continue;
 
         // Get the bit range for this operand:
         unsigned bitStart = bi++, bitWidth = 1;
         for (; bi < Bits.getNumBits(); ++bi) {
           VarInit *Var = nullptr;
           VarBitInit *BI = dyn_cast<VarBitInit>(Bits.getBit(bi));
           if (BI)
             Var = dyn_cast<VarInit>(BI->getBitVar());
           else
             Var = dyn_cast<VarInit>(Bits.getBit(bi));
 
           if (!Var)
             break;
 
           if (Var->getName() != Vals[i].getName())
             break;
 
           ++bitWidth;
         }
 
         unsigned NumberOps = CGI.Operands.size();
         while (NumberedOp < NumberOps &&
                (CGI.Operands.isFlatOperandNotEmitted(NumberedOp) ||
-              (!NamedOpIndices.empty() && NamedOpIndices.count(
+                (!NamedOpIndices.empty() &&
+                 NamedOpIndices.count(
                      CGI.Operands.getSubOperandNumber(NumberedOp).first))))
           ++NumberedOp;
 
         OpIdx = NumberedOp++;
 
         // OpIdx now holds the ordered operand number of Vals[i].
         std::pair<unsigned, unsigned> SO =
             CGI.Operands.getSubOperandNumber(OpIdx);
         const std::string &Name = CGI.Operands[SO.first].Name;
 
         LLVM_DEBUG(dbgs() << "Numbered operand mapping for " << Def.getName()
-                        << ": " << Name << "(" << SO.first << ", " << SO.second
-                        << ") => " << Vals[i].getName() << "\n");
+                          << ": " << Name << "(" << SO.first << ", "
+                          << SO.second << ") => " << Vals[i].getName() << "\n");
 
         std::string Decoder;
         Record *TypeRecord = CGI.Operands[SO.first].Rec;
 
         RecordVal *DecoderString = TypeRecord->getValue("DecoderMethod");
-      StringInit *String = DecoderString ?
-        dyn_cast<StringInit>(DecoderString->getValue()) : nullptr;
+        StringInit *String =
+            DecoderString ? dyn_cast<StringInit>(DecoderString->getValue())
+                          : nullptr;
         if (String && String->getValue() != "")
           Decoder = std::string(String->getValue());
 
-      if (Decoder == "" &&
-          CGI.Operands[SO.first].MIOperandInfo &&
+        if (Decoder == "" && CGI.Operands[SO.first].MIOperandInfo &&
             CGI.Operands[SO.first].MIOperandInfo->getNumArgs()) {
-        Init *Arg = CGI.Operands[SO.first].MIOperandInfo->
-                      getArg(SO.second);
+          Init *Arg = CGI.Operands[SO.first].MIOperandInfo->getArg(SO.second);
           if (DefInit *DI = cast<DefInit>(Arg))
             TypeRecord = DI->getDef();
         }
 
         bool isReg = false;
         if (TypeRecord->isSubClassOf("RegisterOperand"))
           TypeRecord = TypeRecord->getValueAsDef("RegClass");
         if (TypeRecord->isSubClassOf("RegisterClass")) {
           Decoder = "Decode" + TypeRecord->getName().str() + "RegisterClass";
           isReg = true;
         } else if (TypeRecord->isSubClassOf("PointerLikeRegClass")) {
           Decoder = "DecodePointerLikeRegClass" +
                     utostr(TypeRecord->getValueAsInt("RegClassKind"));
           isReg = true;
         }
 
         DecoderString = TypeRecord->getValue("DecoderMethod");
-      String = DecoderString ?
-        dyn_cast<StringInit>(DecoderString->getValue()) : nullptr;
+        String = DecoderString ? dyn_cast<StringInit>(DecoderString->getValue())
+                               : nullptr;
         if (!isReg && String && String->getValue() != "")
           Decoder = std::string(String->getValue());
 
         RecordVal *HasCompleteDecoderVal =
             TypeRecord->getValue("hasCompleteDecoder");
-      BitInit *HasCompleteDecoderBit = HasCompleteDecoderVal ?
-        dyn_cast<BitInit>(HasCompleteDecoderVal->getValue()) : nullptr;
-      bool HasCompleteDecoder = HasCompleteDecoderBit ?
-        HasCompleteDecoderBit->getValue() : true;
+        BitInit *HasCompleteDecoderBit =
+            HasCompleteDecoderVal
+                ? dyn_cast<BitInit>(HasCompleteDecoderVal->getValue())
+                : nullptr;
+        bool HasCompleteDecoder =
+            HasCompleteDecoderBit ? HasCompleteDecoderBit->getValue() : true;
 
         OperandInfo OpInfo(Decoder, HasCompleteDecoder);
         OpInfo.addField(bitStart, bitWidth, 0);
 
         NumberedInsnOperands[Name].push_back(OpInfo);
 
         // FIXME: For complex operands with custom decoders we can't handle tied
         // sub-operands automatically. Skip those here and assume that this is
         // fixed up elsewhere.
         if (CGI.Operands[SO.first].MIOperandInfo &&
-          CGI.Operands[SO.first].MIOperandInfo->getNumArgs() > 1 &&
-          String && String->getValue() != "")
+            CGI.Operands[SO.first].MIOperandInfo->getNumArgs() > 1 && String &&
+            String->getValue() != "")
           NumberedInsnOperandsNoTie.insert(Name);
       }
     }
 
     // For each operand, see if we can figure out where it is encoded.
     for (const auto &Op : InOutOperands) {
       if (!NumberedInsnOperands[std::string(Op.second)].empty()) {
         llvm::append_range(InsnOperands,
                            NumberedInsnOperands[std::string(Op.second)]);
         continue;
       }
       if (!NumberedInsnOperands[TiedNames[std::string(Op.second)]].empty()) {
-      if (!NumberedInsnOperandsNoTie.count(TiedNames[std::string(Op.second)])) {
+        if (!NumberedInsnOperandsNoTie.count(
+                TiedNames[std::string(Op.second)])) {
           // Figure out to which (sub)operand we're tied.
           unsigned i =
               CGI.Operands.getOperandNamed(TiedNames[std::string(Op.second)]);
           int tiedTo = CGI.Operands[i].getTiedRegister();
           if (tiedTo == -1) {
             i = CGI.Operands.getOperandNamed(Op.second);
             tiedTo = CGI.Operands[i].getTiedRegister();
           }
 
           if (tiedTo != -1) {
             std::pair<unsigned, unsigned> SO =
                 CGI.Operands.getSubOperandNumber(tiedTo);
 
             InsnOperands.push_back(
                 NumberedInsnOperands[TiedNames[std::string(Op.second)]]
                                     [SO.second]);
           }
         }
         continue;
       }
 
-    TypedInit *TI = cast<TypedInit>(Op.first);
-
       // At this point, we can locate the decoder field, but we need to know how
       // to interpret it.  As a first step, require the target to provide
       // callbacks for decoding register classes.
-    std::string Decoder = findOperandDecoderMethod(TI);
-    Record *TypeRecord = cast<DefInit>(TI)->getDef();
-
-    RecordVal *HasCompleteDecoderVal =
-      TypeRecord->getValue("hasCompleteDecoder");
-    BitInit *HasCompleteDecoderBit = HasCompleteDecoderVal ?
-      dyn_cast<BitInit>(HasCompleteDecoderVal->getValue()) : nullptr;
-    bool HasCompleteDecoder = HasCompleteDecoderBit ?
-      HasCompleteDecoderBit->getValue() : true;
 
-    OperandInfo OpInfo(Decoder, HasCompleteDecoder);
+      OperandInfo OpInfo = getOpInfo(cast<DefInit>(Op.first)->getDef());
 
       // Some bits of the operand may be required to be 1 depending on the
       // instruction's encoding. Collect those bits.
       if (const RecordVal *EncodedValue = EncodingDef.getValue(Op.second))
-      if (const BitsInit *OpBits = dyn_cast<BitsInit>(EncodedValue->getValue()))
+        if (const BitsInit *OpBits =
+                dyn_cast<BitsInit>(EncodedValue->getValue()))
           for (unsigned I = 0; I < OpBits->getNumBits(); ++I)
             if (const BitInit *OpBit = dyn_cast<BitInit>(OpBits->getBit(I)))
               if (OpBit->getValue())
                 OpInfo.InitValue |= 1ULL << I;
 
       unsigned Base = ~0U;
       unsigned Width = 0;
       unsigned Offset = 0;
 
       for (unsigned bi = 0; bi < Bits.getNumBits(); ++bi) {
         VarInit *Var = nullptr;
         VarBitInit *BI = dyn_cast<VarBitInit>(Bits.getBit(bi));
         if (BI)
           Var = dyn_cast<VarInit>(BI->getBitVar());
         else
           Var = dyn_cast<VarInit>(Bits.getBit(bi));
 
         if (!Var) {
           if (Base != ~0U) {
             OpInfo.addField(Base, Width, Offset);
             Base = ~0U;
             Width = 0;
             Offset = 0;
           }
           continue;
         }
 
-      if (Var->getName() != Op.second &&
-          Var->getName() != TiedNames[std::string(Op.second)]) {
+        if ((Var->getName() != Op.second &&
+             Var->getName() != TiedNames[std::string(Op.second)])) {
           if (Base != ~0U) {
             OpInfo.addField(Base, Width, Offset);
             Base = ~0U;
             Width = 0;
             Offset = 0;
           }
           continue;
         }
 
         if (Base == ~0U) {
           Base = bi;
           Width = 1;
           Offset = BI ? BI->getBitNum() : 0;
         } else if (BI && BI->getBitNum() != Offset + Width) {
           OpInfo.addField(Base, Width, Offset);
           Base = bi;
           Width = 1;
           Offset = BI->getBitNum();
         } else {
           ++Width;
         }
       }
 
       if (Base != ~0U)
         OpInfo.addField(Base, Width, Offset);
 
       if (OpInfo.numFields() > 0)
         InsnOperands.push_back(OpInfo);
     }
+  }
 
   Operands[Opc] = InsnOperands;
 
 #if 0
   LLVM_DEBUG({
       // Dumps the instruction encoding bits.
       dumpBits(errs(), Bits);
 
       errs() << '\n';
 
       // Dumps the list of operand info.
       for (unsigned i = 0, e = CGI.Operands.size(); i != e; ++i) {
         const CGIOperandList::OperandInfo &Info = CGI.Operands[i];
         const std::string &OperandName = Info.Name;
         const Record &OperandDef = *Info.Rec;
 
         errs() << "\t" << OperandName << " (" << OperandDef.getName() << ")\n";
       }
     });
 #endif
 
-  return true;
+  return Bits.getNumBits();
 }
 
 // emitFieldFromInstruction - Emit the templated helper function
 // fieldFromInstruction().
 // On Windows we make sure that this function is not inlined when
 // using the VS compiler. It has a bug which causes the function
 // to be optimized out in some circustances. See llvm.org/pr38292
 static void emitFieldFromInstruction(formatted_raw_ostream &OS) {
OS << "// Helper functions for extracting fields from encoded instructions.\n"
      << "}\n"
      << "\n"
      << "template <typename InsnType>\n"
      << "static std::enable_if_t<!std::is_integral<InsnType>::value, "
         "uint64_t>\n"
      << "fieldFromInstruction(const InsnType &insn, unsigned startBit,\n"
      << "                     unsigned numBits) {\n"
      << "  return insn.extractBitsAsZExtValue(numBits, startBit);\n"
      << "}\n\n";
 }
 
 // emitInsertBits - Emit the templated helper function insertBits().
 static void emitInsertBits(formatted_raw_ostream &OS) {
   OS << "// Helper function for inserting bits extracted from an encoded "
         "instruction into\n"
      << "// a field.\n"
      << "template <typename InsnType>\n"
      << "static std::enable_if_t<std::is_integral<InsnType>::value>\n"
      << "insertBits(InsnType &field, InsnType bits, unsigned startBit, "
         "unsigned numBits) {\n"
      << "  assert(startBit + numBits <= sizeof field * 8);\n"
      << "  field |= (InsnType)bits << startBit;\n"
      << "}\n"
      << "\n"
      << "template <typename InsnType>\n"
      << "static std::enable_if_t<!std::is_integral<InsnType>::value>\n"
      << "insertBits(InsnType &field, uint64_t bits, unsigned startBit, "
         "unsigned numBits) {\n"
      << "  field.insertBits(bits, startBit, numBits);\n"
      << "}\n\n";
 }
 
 // emitDecodeInstruction - Emit the templated helper function
 // decodeInstruction().
-static void emitDecodeInstruction(formatted_raw_ostream &OS) {
+static void emitDecodeInstruction(formatted_raw_ostream &OS,
+                                  bool IsVarLenInst) {
   OS << "template <typename InsnType>\n"
      << "static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], "
         "MCInst &MI,\n"
      << "                                      InsnType insn, uint64_t "
         "Address,\n"
      << "                                      const void *DisAsm,\n"
-     << "                                      const MCSubtargetInfo &STI) {\n"
+     << "                                      const MCSubtargetInfo &STI";
+  if (IsVarLenInst) {
+    OS << ",\n"
+       << "                                      llvm::function_ref<void(APInt "
+          "&,"
+       << " uint64_t)> makeUp";
+  }
+  OS << ") {\n"
      << "  const FeatureBitset &Bits = STI.getFeatureBits();\n"
      << "\n"
      << "  const uint8_t *Ptr = DecodeTable;\n"
-     << "  InsnType CurFieldValue = 0;\n"
+     << "  uint64_t CurFieldValue = 0;\n"
      << "  DecodeStatus S = MCDisassembler::Success;\n"
      << "  while (true) {\n"
      << "    ptrdiff_t Loc = Ptr - DecodeTable;\n"
      << "    switch (*Ptr) {\n"
      << "    default:\n"
      << "      errs() << Loc << \": Unexpected decode table opcode!\\n\";\n"
      << "      return MCDisassembler::Fail;\n"
      << "    case MCD::OPC_ExtractField: {\n"
      << "      unsigned Start = *++Ptr;\n"
      << "      unsigned Len = *++Ptr;\n"
-     << "      ++Ptr;\n"
-     << "      CurFieldValue = fieldFromInstruction(insn, Start, Len);\n"
+     << "      ++Ptr;\n";
+  if (IsVarLenInst)
+    OS << "      makeUp(insn, Start + Len);\n";
+  OS << "      CurFieldValue = fieldFromInstruction(insn, Start, Len);\n"
      << "      LLVM_DEBUG(dbgs() << Loc << \": OPC_ExtractField(\" << Start << "
         "\", \"\n"
      << "                   << Len << \"): \" << CurFieldValue << \"\\n\");\n"
      << "      break;\n"
      << "    }\n"
      << "    case MCD::OPC_FilterValue: {\n"
      << "      // Decode the field value.\n"
      << "      unsigned Len;\n"
-     << "      InsnType Val = decodeULEB128(++Ptr, &Len);\n"
+     << "      uint64_t Val = decodeULEB128(++Ptr, &Len);\n"
      << "      Ptr += Len;\n"
      << "      // NumToSkip is a plain 24-bit integer.\n"
      << "      unsigned NumToSkip = *Ptr++;\n"
      << "      NumToSkip |= (*Ptr++) << 8;\n"
      << "      NumToSkip |= (*Ptr++) << 16;\n"
      << "\n"
      << "      // Perform the filter operation.\n"
      << "      if (Val != CurFieldValue)\n"
      << "        Ptr += NumToSkip;\n"
      << "      LLVM_DEBUG(dbgs() << Loc << \": OPC_FilterValue(\" << Val << "
         "\", \" << NumToSkip\n"
      << "                   << \"): \" << ((Val != CurFieldValue) ? \"FAIL:\" "
         ": \"PASS:\")\n"
      << "                   << \" continuing at \" << (Ptr - DecodeTable) << "
         "\"\\n\");\n"
      << "\n"
      << "      break;\n"
      << "    }\n"
      << "    case MCD::OPC_CheckField: {\n"
      << "      unsigned Start = *++Ptr;\n"
-     << "      unsigned Len = *++Ptr;\n"
-     << "      InsnType FieldValue = fieldFromInstruction(insn, Start, Len);\n"
+     << "      unsigned Len = *++Ptr;\n";
+  if (IsVarLenInst)
+    OS << "      makeUp(insn, Start + Len);\n";
+  OS << "      uint64_t FieldValue = fieldFromInstruction(insn, Start, Len);\n"
      << "      // Decode the field value.\n"
-     << "      InsnType ExpectedValue = decodeULEB128(++Ptr, &Len);\n"
-     << "      Ptr += Len;\n"
+     << "      unsigned PtrLen = 0;\n"
+     << "      uint64_t ExpectedValue = decodeULEB128(++Ptr, &PtrLen);\n"
+     << "      Ptr += PtrLen;\n"
      << "      // NumToSkip is a plain 24-bit integer.\n"
      << "      unsigned NumToSkip = *Ptr++;\n"
      << "      NumToSkip |= (*Ptr++) << 8;\n"
      << "      NumToSkip |= (*Ptr++) << 16;\n"
      << "\n"
      << "      // If the actual and expected values don't match, skip.\n"
      << "      if (ExpectedValue != FieldValue)\n"
      << "        Ptr += NumToSkip;\n"
OS << "      uint64_t FieldValue = fieldFromInstruction(insn, Start, Len);\n"
      << "      // Decode the Opcode value.\n"
      << "      unsigned Opc = decodeULEB128(++Ptr, &Len);\n"
      << "      Ptr += Len;\n"
      << "      unsigned DecodeIdx = decodeULEB128(Ptr, &Len);\n"
      << "      Ptr += Len;\n"
      << "\n"
      << "      MI.clear();\n"
      << "      MI.setOpcode(Opc);\n"
-     << "      bool DecodeComplete;\n"
-     << "      S = decodeToMCInst(S, DecodeIdx, insn, MI, Address, DisAsm, "
+     << "      bool DecodeComplete;\n";
+  if (IsVarLenInst) {
+    OS << "      Len = InstrLenTable[Opc];\n"
+       << "      makeUp(insn, Len);\n";
+  }
+  OS << "      S = decodeToMCInst(S, DecodeIdx, insn, MI, Address, DisAsm, "
         "DecodeComplete);\n"
      << "      assert(DecodeComplete);\n"
      << "\n"
      << "      LLVM_DEBUG(dbgs() << Loc << \": OPC_Decode: opcode \" << Opc\n"
      << "                   << \", using decoder \" << DecodeIdx << \": \"\n"
      << "                   << (S != MCDisassembler::Fail ? \"PASS\" : "
         "\"FAIL\") << \"\\n\");\n"
      << "      return S;\n"
OS << "      S = decodeToMCInst(S, DecodeIdx, insn, MI, Address, DisAsm, "
      << "        // set before the decode attempt.\n"
      << "        S = MCDisassembler::Success;\n"
      << "      }\n"
      << "      break;\n"
      << "    }\n"
      << "    case MCD::OPC_SoftFail: {\n"
      << "      // Decode the mask values.\n"
      << "      unsigned Len;\n"
-     << "      InsnType PositiveMask = decodeULEB128(++Ptr, &Len);\n"
+     << "      uint64_t PositiveMask = decodeULEB128(++Ptr, &Len);\n"
      << "      Ptr += Len;\n"
-     << "      InsnType NegativeMask = decodeULEB128(Ptr, &Len);\n"
+     << "      uint64_t NegativeMask = decodeULEB128(Ptr, &Len);\n"
      << "      Ptr += Len;\n"
-     << "      bool Fail = (insn & PositiveMask) || (~insn & NegativeMask);\n"
+     << "      bool Fail = (insn & PositiveMask) != 0 || (~insn & "
+        "NegativeMask) != 0;\n"
      << "      if (Fail)\n"
      << "        S = MCDisassembler::SoftFail;\n"
      << "      LLVM_DEBUG(dbgs() << Loc << \": OPC_SoftFail: \" << (Fail ? "
         "\"FAIL\\n\" : \"PASS\\n\"));\n"
      << "      break;\n"
      << "    }\n"
      << "    case MCD::OPC_Fail: {\n"
      << "      LLVM_DEBUG(dbgs() << Loc << \": OPC_Fail\\n\");\n"
void FixedLenDecoderEmitter::run(raw_ostream &o) {
 
   emitFieldFromInstruction(OS);
   emitInsertBits(OS);
 
   Target.reverseBitsForLittleEndianEncoding();
 
   // Parameterize the decoders based on namespace and instruction width.
   std::set<StringRef> HwModeNames;
-  const auto &NumberedInstructions = Target.getInstructionsByEnumValue();
+  const auto NumberedInstructions = Target.getInstructionsByEnumValue();
   NumberedEncodings.reserve(NumberedInstructions.size());
   DenseMap<Record *, unsigned> IndexOfInstruction;
   // First, collect all HwModes referenced by the target.
   for (const auto &NumberedInstruction : NumberedInstructions) {
     IndexOfInstruction[NumberedInstruction->TheDef] = NumberedEncodings.size();
 
     if (const RecordVal *RV =
             NumberedInstruction->TheDef->getValue("EncodingInfos")) {
void FixedLenDecoderEmitter::run(raw_ostream &o) {
   for (const auto &NumberedAlias : RK.getAllDerivedDefinitions("AdditionalEncoding"))
     NumberedEncodings.emplace_back(
         NumberedAlias,
         &Target.getInstruction(NumberedAlias->getValueAsDef("AliasOf")));
 
   std::map<std::pair<std::string, unsigned>, std::vector<EncodingIDAndOpcode>>
       OpcMap;
   std::map<unsigned, std::vector<OperandInfo>> Operands;
+  std::vector<unsigned> InstrLen;
+
+  bool IsVarLenInst =
+      any_of(NumberedInstructions, [](const CodeGenInstruction *CGI) {
+        RecordVal *RV = CGI->TheDef->getValue("Inst");
+        return RV && isa<DagInit>(RV->getValue());
+      });
+  unsigned MaxInstLen = 0;
 
   for (unsigned i = 0; i < NumberedEncodings.size(); ++i) {
     const Record *EncodingDef = NumberedEncodings[i].EncodingDef;
     const CodeGenInstruction *Inst = NumberedEncodings[i].Inst;
     const Record *Def = Inst->TheDef;
     unsigned Size = EncodingDef->getValueAsInt("Size");
     if (Def->getValueAsString("Namespace") == "TargetOpcode" ||
         Def->getValueAsBit("isPseudo") ||
         Def->getValueAsBit("isAsmParserOnly") ||
         Def->getValueAsBit("isCodeGenOnly")) {
       NumEncodingsLackingDisasm++;
       continue;
     }
 
     if (i < NumberedInstructions.size())
       NumInstructions++;
     NumEncodings++;
 
-    if (!Size)
+    if (!Size && !IsVarLenInst)
       continue;
 
-    if (populateInstruction(Target, *EncodingDef, *Inst, i, Operands)) {
+    if (IsVarLenInst)
+      InstrLen.resize(NumberedInstructions.size(), 0);
+
+    if (unsigned Len = populateInstruction(Target, *EncodingDef, *Inst, i,
+                                           Operands, IsVarLenInst)) {
+      if (IsVarLenInst) {
+        MaxInstLen = std::max(MaxInstLen, Len);
+        InstrLen[i] = Len;
+      }
       std::string DecoderNamespace =
           std::string(EncodingDef->getValueAsString("DecoderNamespace"));
       if (!NumberedEncodings[i].HwModeName.empty())
         DecoderNamespace +=
             std::string("_") + NumberedEncodings[i].HwModeName.str();
       OpcMap[std::make_pair(DecoderNamespace, Size)].emplace_back(
           i, IndexOfInstruction.find(Def)->second);
     } else {
       NumEncodingsOmitted++;
     }
   }
 
   DecoderTableInfo TableInfo;
   for (const auto &Opc : OpcMap) {
     // Emit the decoder for this namespace+width combination.
     ArrayRef<EncodingAndInst> NumberedEncodingsRef(
         NumberedEncodings.data(), NumberedEncodings.size());
     FilterChooser FC(NumberedEncodingsRef, Opc.second, Operands,
-                     8 * Opc.first.second, this);
+                     IsVarLenInst ? MaxInstLen : 8 * Opc.first.second, this);
 
     // The decode table is cleared for each top level decoder function. The
     // predicates and decoders themselves, however, are shared across all
     // decoders to give more opportunities for uniqueing.
     TableInfo.Table.clear();
     TableInfo.FixupStack.clear();
     TableInfo.Table.reserve(16384);
     TableInfo.FixupStack.emplace_back();
     FC.emitTableEntries(TableInfo);
     // Any NumToSkip fixups in the top level scope can resolve to the
     // OPC_Fail at the end of the table.
     assert(TableInfo.FixupStack.size() == 1 && "fixup stack phasing error!");
     // Resolve any NumToSkip fixups in the current scope.
     resolveTableFixups(TableInfo.Table, TableInfo.FixupStack.back(),
                        TableInfo.Table.size());
     TableInfo.FixupStack.clear();
 
     TableInfo.Table.push_back(MCD::OPC_Fail);
 
     // Print the table to the output stream.
     emitTable(OS, TableInfo.Table, 0, FC.getBitWidth(), Opc.first.first);
     OS.flush();
   }
 
+  if (IsVarLenInst)
+    emitInstrLenTable(OS, InstrLen);
   // Emit the predicate function.
   emitPredicateFunction(OS, TableInfo.Predicates, 0);
 
   // Emit the decoder function.
   emitDecoderFunction(OS, TableInfo.Decoders, 0);
 
   // Emit the main entry point for the decoder, decodeInstruction().
-  emitDecodeInstruction(OS);
+  emitDecodeInstruction(OS, IsVarLenInst);
 
   OS << "\n} // end namespace llvm\n";
 }
 
 namespace llvm {
 
 void EmitFixedLenDecoder(RecordKeeper &RK, raw_ostream &OS,
                          const std::string &PredicateNamespace,
                          const std::string &GPrefix,
                          const std::string &GPostfix, const std::string &ROK,
                          const std::string &RFail, const std::string &L) {
   FixedLenDecoderEmitter(RK, PredicateNamespace, GPrefix, GPostfix,
                          ROK, RFail, L).run(OS);
 }
 
 } // end namespace llvm

llvm/utils/TableGen/VarLenCodeEmitterGen.h

 //===- VarLenCodeEmitterGen.h - CEG for variable-length insts ---*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file declare the CodeEmitterGen component for variable-length
 // instructions. See the .cpp file for more details.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_UTILS_TABLEGEN_VARLENCODEEMITTERGEN_H
 #define LLVM_UTILS_TABLEGEN_VARLENCODEEMITTERGEN_H
 
+#include "llvm/TableGen/Record.h"
+
 namespace llvm {
 
+struct EncodingSegment {
+  unsigned BitWidth;
+  const Init *Value;
+  StringRef CustomEncoder = "";
+};
+
+class VarLenInst {
+  const RecordVal *TheDef;
+  size_t NumBits;
+
+  // Set if any of the segment is not fixed value.
+  bool HasDynamicSegment;
+
+  SmallVector<EncodingSegment, 4> Segments;
+
+  void buildRec(const DagInit *DI);
+
+  StringRef getCustomEncoderName(const Init *EI) const {
+    if (const auto *DI = dyn_cast<DagInit>(EI)) {
+      if (DI->getNumArgs() && isa<StringInit>(DI->getArg(0)))
+        return cast<StringInit>(DI->getArg(0))->getValue();
+    }
+    return "";
+  }
+
+public:
+  VarLenInst() : TheDef(nullptr), NumBits(0U), HasDynamicSegment(false) {}
+
+  explicit VarLenInst(const DagInit *DI, const RecordVal *TheDef);
+
+  /// Number of bits
+  size_t size() const { return NumBits; }
+
+  using const_iterator = decltype(Segments)::const_iterator;
+
+  const_iterator begin() const { return Segments.begin(); }
+  const_iterator end() const { return Segments.end(); }
+  size_t getNumSegments() const { return Segments.size(); }
+
+  bool isFixedValueOnly() const { return !HasDynamicSegment; }
+};
+
 class RecordKeeper;
 class raw_ostream;
 
 void emitVarLenCodeEmitter(RecordKeeper &R, raw_ostream &OS);
 
 } // end namespace llvm
 #endif

llvm/utils/TableGen/VarLenCodeEmitterGen.cpp

 #include "CodeGenHwModes.h"
 #include "CodeGenInstruction.h"
 #include "CodeGenTarget.h"
 #include "InfoByHwMode.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/TableGen/Error.h"
-#include "llvm/TableGen/Record.h"
 
 using namespace llvm;
 
 namespace {
 
 class VarLenCodeEmitterGen {
   RecordKeeper &Records;
 
-  struct EncodingSegment {
-    unsigned BitWidth;
-    const Init *Value;
-    StringRef CustomEncoder = "";
-  };
-
-  class VarLenInst {
-    RecordVal *TheDef;
-    size_t NumBits;
-
-    // Set if any of the segment is not fixed value.
-    bool HasDynamicSegment;
-
-    SmallVector<EncodingSegment, 4> Segments;
-
-    void buildRec(const DagInit *DI);
-
-    StringRef getCustomEncoderName(const Init *EI) const {
-      if (const auto *DI = dyn_cast<DagInit>(EI)) {
-        if (DI->getNumArgs() && isa<StringInit>(DI->getArg(0)))
-          return cast<StringInit>(DI->getArg(0))->getValue();
-      }
-      return "";
-    }
-
-  public:
-    VarLenInst() : TheDef(nullptr), NumBits(0U), HasDynamicSegment(false) {}
-
-    explicit VarLenInst(const DagInit *DI, RecordVal *TheDef);
-
-    /// Number of bits
-    size_t size() const { return NumBits; }
-
-    using const_iterator = decltype(Segments)::const_iterator;
-
-    const_iterator begin() const { return Segments.begin(); }
-    const_iterator end() const { return Segments.end(); }
-    size_t getNumSegments() const { return Segments.size(); }
-
-    bool isFixedValueOnly() const { return !HasDynamicSegment; }
-  };
-
   DenseMap<Record *, VarLenInst> VarLenInsts;
 
   // Emit based values (i.e. fixed bits in the encoded instructions)
   void emitInstructionBaseValues(
       raw_ostream &OS,
       ArrayRef<const CodeGenInstruction *> NumberedInstructions,
       CodeGenTarget &Target, int HwMode = -1);
 
   std::string getInstructionCase(Record *R, CodeGenTarget &Target);
   std::string getInstructionCaseForEncoding(Record *R, Record *EncodingDef,
                                             CodeGenTarget &Target);
 
 public:
   explicit VarLenCodeEmitterGen(RecordKeeper &R) : Records(R) {}
 
   void run(raw_ostream &OS);
 };
 
 } // end anonymous namespace
 
-VarLenCodeEmitterGen::VarLenInst::VarLenInst(const DagInit *DI,
-                                             RecordVal *TheDef)
+VarLenInst::VarLenInst(const DagInit *DI, const RecordVal *TheDef)
     : TheDef(TheDef), NumBits(0U) {
   buildRec(DI);
   for (const auto &S : Segments)
     NumBits += S.BitWidth;
 }
 
-void VarLenCodeEmitterGen::VarLenInst::buildRec(const DagInit *DI) {
+void VarLenInst::buildRec(const DagInit *DI) {
   assert(TheDef && "The def record is nullptr ?");
 
   std::string Op = DI->getOperator()->getAsString();
 
   if (Op == "ascend" || Op == "descend") {
     bool Reverse = Op == "descend";
     int i = Reverse ? DI->getNumArgs() - 1 : 0;
     int e = Reverse ? -1 : DI->getNumArgs();