[Generic][AIE2] Combiner for shufflevectors that use build vector #129

ValentijnvdBeek · 2024-07-19T13:20:04Z

Transforms a shufflevector that uses a build vector or undefined into just a build vector. This can be done is because a shuffle vector lowering is an unmerge and then merge. Since build is a merge, the merge and unmerge cancel each other out and we can just merge the vector directly.

Example:

    %1:_(s32) = COPY $r0
    %3:_(<8 x s32>) = G_IMPLICIT_DEF
    %5:_(s32) = G_IMPLICIT_DEF
    %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32)
    %0:_(<8 x s32>) = G_SHUFFLE_VECTOR %2(<8 x s32>), %3, shufflemask(0, 0, 0, 0, 0, 0, 0, 0)
    ===>
    %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32)

konstantinschwarz

Mostly nits, combiner looks good to me

konstantinschwarz · 2024-07-22T19:18:33Z

llvm/include/llvm/CodeGen/GlobalISel/GenericMachineInstrs.h

      return true;
    default:
      return false;
    }
  }
 };

+/// Represents a G_SHUFFLE_VECTOR
+class GShuffleVector : public GMergeLikeInstr {


G_SHUFFLE_VECTOR takes the shufflemask as the last operand, which shouldn't be part of the getNumSources I think?
To me that's what makes G_SHUFFLE_VECTOR different compared to the other "merge like" instructions.
I'd rather not make it part of that group, as this can have surprising effects to other existing combines

I can see that argument. I was looking at it from the perspective of what the result of the function would be. I'll separate them, thanks.

konstantinschwarz · 2024-07-22T19:22:39Z

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

+  // Our inputs need to be either be build vectors or undefined, register inputs
+  // break this optimization. You could maybe do something clever were you
+  // concatenate vectors to save half a build vector.
+  if ((SrcInstr1 == 0 && IsUndef1 == 0) || (SrcInstr2 == 0 && IsUndef2 == 0))


These are pointers, just use (!SrcInstr1 && !IsUndef1) || ...)

Sure, will do. I personally prefer explicit NULL pointer checking for clarity (which, to be fair, this isn't yet), but doing it like that is fine with me.

konstantinschwarz · 2024-07-22T19:38:57Z

llvm/test/CodeGen/AIE/aie2/GlobalISel/prelegalizercombiner-shufflevector-buildvector.mir

+    ; CHECK: [[DEF:%[0-9]+]]:_(<16 x s32>) = G_IMPLICIT_DEF
+    ; CHECK-NEXT: $x0 = COPY [[DEF]](<16 x s32>)
+    ; CHECK-NEXT: PseudoRET implicit $lr, implicit $x0
+    %0:_(s32) = G_CONSTANT i32 28


Nit: unused G_CONSTANTs here and in the tests below

Fair, thanks Konstantin.

konstantinschwarz · 2024-07-22T19:42:11Z

llvm/test/CodeGen/AIE/aie2/GlobalISel/prelegalizercombiner-shufflevector-buildvector.mir

+    %5:_(s32) = G_IMPLICIT_DEF
+    %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32)
+    %0:_(<8 x s32>) = G_SHUFFLE_VECTOR %2(<8 x s32>), %3, shufflemask(0, 0, 0, 0, 0, 0, 0, 0)
+    PseudoRET implicit $lr, implicit %0


Looks like this is missing the closing ... at the end, and then I think this would fail, because this is further combined into G_AIE_BROADCAST?

The shufflevector branch has become really divergent in the past month or so, so that hasn't landed yet. Let's try to get all of that merged as quickly as we can, I'll manually merge them to make sure that spectests are updated before enter the main branch.

Missing the closing ... is deliberate, trailing ellipses have caused errors with the unit tests scripts for me before.

Ran into it again just now. If you have a trailing ellipsis, you need a trailing newline, or otherwise the update mir script will error out. If you skip it, then it doesn't matter whether you have a newline or not. That is why I tend to commit mine without the trailing ellipsis.

llvm/test/CodeGen/AIE/aie2/GlobalISel/prelegalizercombiner-shufflevector-buildvector.mir

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

This implements the simple legalization that lowers G_SHUFFLE_VECTOR into extracts of the elements based on the mask and then combining them using a G_BUILD_VECTOR. Our architecture has a VSHUFFLE instruction which could be used to implement some patterns more efficiently.

… CONCAT_VECTOR We check for iterative shift masks which corresponds to the CONCAT_VECTOR instruction.

…et size

…hunks of a vector

…rs together

… of two vectors together

Transforms a shufflevector that uses a build vector or undefined into just a build vector. This can be done is because a shuffle vector lowering is an unmerge and then merge. Since build is a merge, the merge and unmerge cancel each other out and we can just merge the vector directly. Example: ``` %1:_(s32) = COPY $r0 %3:_(<8 x s32>) = G_IMPLICIT_DEF %5:_(s32) = G_IMPLICIT_DEF %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32) %0:_(<8 x s32>) = G_SHUFFLE_VECTOR %2(<8 x s32>), %3, shufflemask(0, 0, 0, 0, 0, 0, 0, 0) ===> %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32) ```

ValentijnvdBeek · 2024-08-05T13:39:17Z

llvm/lib/Target/AArch64/AArch64Combine.td

                        select_to_minmax, or_to_bsp, combine_concat_vector,
-                        commute_constant_to_rhs]> {
+                        commute_constant_to_rhs, shufflevector_merge]> {
 }


This is needed since some ARM64 tests relies on the legalizer changing the inputs of the shufflevector and if you don't run them afterwards you get worse code.

ValentijnvdBeek · 2024-08-12T12:30:59Z

llvm/test/CodeGen/AArch64/GlobalISel/legalize-shuffle-vector-widen-crash.ll

-; CHECK-NEXT:    mov.h v0[1], v1[0]
+; CHECK-NEXT:    mov w8, #0 ; =0x0
+; CHECK-NEXT:    fmov s0, w8
+; CHECK-NEXT:    mov.16b v1, v0


Add copyright header!

ValentijnvdBeek · 2024-10-11T13:37:22Z

llvm/include/llvm/Target/GlobalISel/Combine.td

+def shufflevector_merge_matchinfo : GIDefMatchData<"SmallVector<Register, 8>">;
+def shufflevector_merge : GICombineRule<
+  (defs root:$d, shufflevector_merge_matchinfo:$info),
+  (match (wip_match_opcode G_SHUFFLE_VECTOR): $d,


A recent move in LLVM is that you shouldn't use wip_match_opcode anymore since it slows down compilation.

https://llvm.org/docs/GlobalISel/MIRPatterns.html#gallery

ValentijnvdBeek added llvm:globalisel Code that modifies the Global Intruction Selection backend:aie Code that modifies AIE code vectorization Support for vector instructions llvm:instcombine Code that modifies the combiner llvm:core Modifies non-AIE specific code backend:aie2 labels Jul 19, 2024

ValentijnvdBeek self-assigned this Jul 19, 2024

ValentijnvdBeek requested review from abhinay-anubola, abnikant, andcarminati, gbossu, khallouh, konstantinschwarz, martien-de-jong, SagarMaheshwari99 and stephenneuendorffer as code owners July 19, 2024 13:20

konstantinschwarz reviewed Jul 22, 2024

View reviewed changes

niwinanto reviewed Jul 23, 2024

View reviewed changes

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp Show resolved Hide resolved

ValentijnvdBeek force-pushed the vvandebe.sv.bv.combiner branch from e0a5aae to cc0f633 Compare July 29, 2024 12:54

ValentijnvdBeek added 11 commits July 30, 2024 11:15

[AIE2] Tests for instrinsic lowering using shufflevector

0d0489b

[AIE2] Enable G_CONCAT_VECTOR optimizations for AIE2

2a44a46

[GISel][CombinerHelper] Use a stream to check mask patterns to detect…

898de6d

… CONCAT_VECTOR We check for iterative shift masks which corresponds to the CONCAT_VECTOR instruction.

[GISel][CombinerHelper] Add a helper that unmerges a vector to a targ…

59e90e2

…et size

[GISel][CombinerHelper] Add two patterns that extract the first two c…

70eb536

…hunks of a vector

[GISel][CombinerHelper] Add a function that chains a list of generato…

f2d4bb6

…rs together

[GISel][CombinerHelper] Add a combiner to concatenate the first halfs…

dd2555b

… of two vectors together

[GISel][CombinerHelper] Shuffle pattern for reversing vector order

4836f6a

[AIE2] AIE2 custom shuffle vector mask support

de8762a

[AIE2] Replace 8x8->8x8 tranpose shuffle vector with vshuffle

b4aca80

[AIE2] Implement vshuffle selection

84f3995

ValentijnvdBeek force-pushed the vvandebe.vshuffle.impl branch from 5654047 to 84f3995 Compare August 2, 2024 11:22

ValentijnvdBeek force-pushed the vvandebe.sv.bv.combiner branch from cc0f633 to fee2d99 Compare August 5, 2024 13:36

ValentijnvdBeek commented Aug 5, 2024

View reviewed changes

ValentijnvdBeek force-pushed the vvandebe.vshuffle.impl branch from 84f3995 to 5c3b1a6 Compare August 7, 2024 18:35

ValentijnvdBeek commented Aug 12, 2024

View reviewed changes

ValentijnvdBeek force-pushed the vvandebe.vshuffle.impl branch 3 times, most recently from aec1600 to d1d0a3a Compare August 15, 2024 13:35

ValentijnvdBeek commented Oct 11, 2024

View reviewed changes

ValentijnvdBeek mentioned this pull request Oct 28, 2024

[LLVM] Optimize G_SHUFFLE_VECTOR into more efficient generic opcodes #41

Open

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Generic][AIE2] Combiner for shufflevectors that use build vector #129

[Generic][AIE2] Combiner for shufflevectors that use build vector #129

ValentijnvdBeek commented Jul 19, 2024

konstantinschwarz left a comment

konstantinschwarz Jul 22, 2024

ValentijnvdBeek Jul 29, 2024

konstantinschwarz Jul 22, 2024

ValentijnvdBeek Jul 29, 2024

konstantinschwarz Jul 22, 2024

ValentijnvdBeek Jul 29, 2024

konstantinschwarz Jul 22, 2024

ValentijnvdBeek Jul 29, 2024

ValentijnvdBeek Jul 29, 2024

ValentijnvdBeek Aug 5, 2024

ValentijnvdBeek Aug 12, 2024

ValentijnvdBeek Oct 11, 2024

[Generic][AIE2] Combiner for shufflevectors that use build vector #129

Are you sure you want to change the base?

[Generic][AIE2] Combiner for shufflevectors that use build vector #129

Conversation

ValentijnvdBeek commented Jul 19, 2024

konstantinschwarz left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment