[BACKPORT] Bump LLVM to pick fixes for Gfx12 Hazards

[umangyadav]

Fixes https://github.com/ROCm/rocMLIR-internal/issues/1827

Moves LLVM copy on release branch forward to https://github.com/ROCm/llvm-project/tree/a3a3e6997bd7bd82cce6737c1315019704a846ed which has fix llvm/llvm-project#118750 which dates back to Jan 29, 2025

Previously on release/rocm-rel-6.4 it was pointing to https://github.com/ROCm/llvm-project/tree/656552edc693e2bb4abc9258399c39d190fce2b3 which dates back to Jan 8, 2025

• Between 656552 and a3a3e6, it has eight commits with some breaking ⛓️‍💥 TOSA changes. I've NOT cherry-picked them to avoid dealing with those.
• a3a3e6 doesn't have upstream changes for OCP Fp8, therefore i've cherry picked necessary changes for those. [MLIR][AMDGPU] Add OCP FP8 support for new hardware llvm/llvm-project#127728 OCP Fp8 PR was merged on March 3, 2025.
• Backported Use hip-python API instead of rocm_agent_enumerator #1762 and Add requirements.txt file and modify Dockerfile #1776 without hip-python unit-tests can't be run on rocm-6.4 docker.
• rocMLIR has custom changes for adding "group" attribute on tosa.conv2d which is not upstreamed. I've added those changes [DO NOT SQUASH]Add support to group conv2d #1351. These changes have not been upstreamed. We need to fix those. https://github.com/ROCm/rocMLIR-internal/issues/1610
• runner-pipelines.mlir test is marked as XFAIL as it can't print conver-vector-to-llvm as it was not serializable. It was fixed later in upstream llvm by [mlir] Refactor ConvertVectorToLLVMPass options llvm/llvm-project#128219

Additional Details :

List of TOSA commits between 656552 and a3a3e6 that are not cherry-picked :

commit a58e774fba42e13aa00667d644e96b783fc914b4
Author: Jack Frankland [email protected]
Date: Tue Jan 28 16:25:22 2025 +0000

[mlir][tosa] Make TOSA MUL's Shift an Input (#121953)

The TOSA-v1.0 specification makes the shift attribute of the MUL
(Hammard product) operator an input. Move the `shift` parameter of the
MUL operator in the MILR TOSA dialect from an attribute to an input and
update any lit tests appropriately.

Expand the verifier of the `tosa::MulOp` operation to check the various
constraints defined in the TOSA-v1.0 specification. Specifically, ensure
that all input operands (excluding the optional shift) are of the same
rank. This means that broadcasting tests which previously checked rank-0
tensors would be broadcast are no longer valid and are removed.

Signed-off-by: Jack Frankland <[email protected]>
Co-authored-by: TatWai Chong <[email protected]>

commit 8f17f51deb12456f25d32b9a42ac1f00feabbfbc
Author: Longsheng Mou [email protected]
Date: Mon Jan 27 18:50:53 2025 +0800

[mlir][tosa] Fix comments format(NFC) (#124520)

This PR corrects the formatting of comments in Markdown. The previous
format was as follows:
https://mlir.llvm.org/docs/Dialects/TOSA/#tosaerf-mlirtosaerfop

![image](https://github.com/user-attachments/assets/1d1d10d5-c960-4724-9fb4-29c17ea39b11)

https://mlir.llvm.org/docs/Dialects/TOSA/#tosarescale-mlirtosarescaleop

![image](https://github.com/user-attachments/assets/fb23cbf6-be10-4a60-8b43-b28dc2db6918)

commit 8388040fc9e75d49cd000b3371e2610c6c3548ba
Author: Jack Frankland [email protected]
Date: Thu Jan 23 10:14:00 2025 +0000

[mlir][tosa] Add NaN Propagation Mode Support (#121951)

The TOSA-V1.0 specification adds "nan propagation" modes as attributes
for several operators. Adjust the ODS definitions of the relevant
operations to include this attribute.

The defined modes are "PROPAGATE" and "IGNORE" and the PROPAGATE mode is
set by default.

MAXIMUM, MINIMUM, REDUCE_MAX, REDUCE_MIN, MAX_POOL, CLAMP, and ARGMAX
support this attribute.

Signed-off-by: Jack Frankland <[email protected]>
Co-authored-by: TatWai Chong <[email protected]>

commit 7e622b61320543b3706711609f1f32fd9ea3788d
Author: Jerry-Ge [email protected]
Date: Wed Jan 22 12:36:48 2025 -0800

[TOSA] Change PadOp padding to tosa.shape (#123133)

This patch changes PadOp's padding input to type !tosa.shape<2 * rank>,
(where rank is the rank of the PadOp's input), instead of a <rank x 2>
tensor.

This patch is also a part of TOSA v1.0 effort:
https://discourse.llvm.org/t/rfc-tosa-dialect-increment-to-v1-0/83708

This patch updates the PadOp to match all against the TOSA v1.0 form.

Original Authors include:
@Tai78641
@wonjeon

Co-authored-by: Tai Ly <[email protected]>

commit 7986e0cad10f3bf9efbbe31110ece68af5cb8751
Author: Tai Ly [email protected]
Date: Wed Jan 22 07:29:44 2025 -0600

[TOSA] bug fix infer shape for slice (#113497)

This fixes the infer output shape of TOSA slice op for start/size values
that are out-of-bound or -1

added tests to check:
  - size = -1
  - size is out of bound
  - start is out of bound

Signed-off-by: Tai Ly <[email protected]>

commit 729f958c4f7548c2d5be5f024b7254cd3ea25c64
Author: Tai Ly [email protected]
Date: Wed Jan 22 07:21:04 2025 -0600

[TOSA] Add SameOperandsAndResultRank to TOSA Ops (#104501)

[note: this is blocked by:
https://github.com/tensorflow/tensorflow/pull/73891 otherwise tensorflow
may have lit test failures]

This patch adds SameOperandsAndResultRank trait to TOSA operators with
ResultsBroadcastableShape trait. SameOperandsAndResultRank trait
requiring that all operands and results have matching ranks unless the
operand/result is unranked.

This also renders the TosaMakeBroadcastable pass unnecessary - but this
pass is left in for now just in case it is still used in some flows. The
lit test, broadcast.mlir, is removed.

This also adds verify of the SameOperandsAndResultRank trait in the
TosaInferShapes pass to validate inferred shapes.

Signed-off-by: Tai Ly <[email protected]>

commit 5c6db8c9e0c488585e642e0c17582ba3aaab7518
Author: Matthias Gehre [email protected]
Date: Mon Jan 20 08:48:28 2025 +0100

[MLIR] TosaToLinalg: Prefer to emit identity maps (#123295)

When deciding whether to emit a map like
`#map = affine_map<(d0, d1, d2, d3) -> (0, d1, d2, d3)>` or `#map =
affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>` for an operand of a
`linalg.generic` when lowering element-wise TOSA ops, prefer the latter
unless broadcasting of the operand is really needed.

This helps later transformations which often require the affine map to
be a projected permuatation.

commit f09db6a3af971ab7d9bbc7ba574a8dc0c10b2940
Author: Jerry-Ge [email protected]
Date: Tue Jan 14 13:41:08 2025 -0800

[TOSA] Add Tosa_Shape type and ConstShapeOp (#122547)

Adds:
1. tosa shape type to Tosa dialect e.g., !tosa.shape<4> is a type for
rank-4 shape values (size-4 array of index values)
 2. const_shape operator
3. trait TosaShapeOperator, added to tosa shape operators, and a
verifier that all operands and results of operator are tosa shapes
4. trait TosaResolvableShapeOperands, added to all tosa operators, and a
verifier that every tosa shape operand is produced by a tosa shape
operator (indicated by trait TosaShapeOperator)
5. trait TosaShapeOperatorWithSameRanks, added to
Tosa_ElementwiseShapeOp and a verifier that all operands and result
shapes have same ranks
5. changed TileOp's multiples from attribute to input, of !tosa.shape
type.
 6. add folder for tosa ConstShape operator

This patch was originally authored by Tai Ly <[email protected]>

Signed-off-by: Jerry Ge <[email protected]>
Signed-off-by: Tai Ly <[email protected]>

[dhernandez0]

rocMLIR has custom changes for adding "group" attribute on tosa.conv2d which is not upstreamed. I've added those changes #1351. These changes have not been upstreamed. We need to fix those. https://github.com/ROCm/rocMLIR-internal/issues/1610

I think they didn't accept the patch upstream because tosa specs don't have group convolutions (or didn't have at that time, haven't checked tosa 1.0 specs).

[causten]

Pull in #1792

[causten]

Testing ongoing, approved to merge to expand the testing team.