Bump llvm to ec17161c6e87bb061daee4603ed21ab988f9d03d (#2047)

Xilinx · Feb 16, 2025 · 98bec8f · 98bec8f
1 parent 32179a0
commit 98bec8f
Show file tree

Hide file tree

Showing 11 changed files with 34 additions and 14 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -188,7 +188,7 @@ cmake_dependent_option(AIE_ENABLE_AIRBIN
 # If we need runtime libs, then statically link them.
 set(CMAKE_MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
 
-# add_flag_if_supported("-Werror=sign-compare" WERROR_SIGN_COMPARE)
+add_flag_if_supported("-Werror=sign-compare" WERROR_SIGN_COMPARE)
 add_flag_if_supported("-Werror=unused" WERROR_USED)
 
 # What happens when you have a non-void function with no return?

diff --git a/lib/Dialect/XLLVM/XLLVMOps.cpp b/lib/Dialect/XLLVM/XLLVMOps.cpp
@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "aie/Dialect/XLLVM/XLLVMDialect.h"
+#include "llvm/IR/IRBuilder.h"
 #include "mlir/Dialect/LLVMIR/LLVMTypes.h"
 #include "mlir/IR/OpDefinition.h"
 #include "mlir/IR/TypeUtilities.h"

diff --git a/python/compiler/aiecc/main.py b/python/compiler/aiecc/main.py
@@ -399,10 +399,16 @@ def downgrade_ir_for_chess(llvmir_chesslinked):
             "inaccessiblemem_or_argmemonly writeonly",
         )
         .replace("captures(none)", "nocapture")
+        .replace("getelementptr inbounds nuw", "getelementptr inbounds")
     )
     return llvmir_chesslinked
 
 
+def downgrade_ir_for_peano(llvmir):
+    llvmir = llvmir.replace("getelementptr inbounds nuw", "getelementptr inbounds")
+    return llvmir
+
+
 class FlowRunner:
     def __init__(self, mlir_module_str, opts, tmpdirname):
         self.mlir_module_str = mlir_module_str
@@ -494,6 +500,16 @@ async def chesshack(self, task, llvmir, aie_target):
 
         return llvmir_chesslinked_path
 
+    # In order to run peano on modern ll code, we need a bunch of hacks.
+    async def peanohack(self, llvmir):
+        llvmir_peanohack = llvmir + "peanohack.ll"
+
+        llvmir_ir = await read_file_async(llvmir)
+        llvmir_hacked_ir = downgrade_ir_for_peano(llvmir_ir)
+        await write_file_async(llvmir_hacked_ir, llvmir_peanohack)
+
+        return llvmir_peanohack
+
     async def process_core(
         self,
         core,
@@ -563,8 +579,9 @@ async def process_core(
 
             elif opts.compile:
                 if not opts.unified:
+                    file_core_llvmir_peanohacked = await self.peanohack(file_core_llvmir)
                     file_core_llvmir_stripped = corefile(self.tmpdirname, core, "stripped.ll")
-                    await self.do_call(task, [self.peano_opt_path, "--passes=default<O2>,strip", "-S", file_core_llvmir, "-o", file_core_llvmir_stripped])
+                    await self.do_call(task, [self.peano_opt_path, "--passes=default<O2>,strip", "-S", file_core_llvmir_peanohacked, "-o", file_core_llvmir_stripped])
                     await self.do_call(task, [self.peano_llc_path, file_core_llvmir_stripped, "-O2", "--march=" + aie_target.lower(), "--function-sections", "--filetype=obj", "-o", file_core_obj])
                 else:
                     file_core_obj = self.unified_file_core_obj
@@ -1154,8 +1171,9 @@ async def run_flow(self):
                     file_llvmir_hacked = await self.chesshack(progress_bar.task, file_llvmir, aie_target)
                     await self.do_call(progress_bar.task, ["xchesscc_wrapper", aie_target.lower(), "+w", self.prepend_tmp("work"), "-c", "-d", "+Wclang,-xir", "-f", file_llvmir_hacked, "-o", self.unified_file_core_obj])
                 elif opts.compile:
+                    file_llvmir_hacked = await self.peanohack(file_llvmir)
                     file_llvmir_opt = self.prepend_tmp("input.opt.ll")
-                    await self.do_call(progress_bar.task, [self.peano_opt_path, "--passes=default<O2>", "-inline-threshold=10", "-S", file_llvmir, "-o", file_llvmir_opt])
+                    await self.do_call(progress_bar.task, [self.peano_opt_path, "--passes=default<O2>", "-inline-threshold=10", "-S", file_llvmir_hacked, "-o", file_llvmir_opt])
                     await self.do_call(progress_bar.task, [self.peano_llc_path, file_llvmir_opt, "-O2", "--march=" + aie_target.lower(), "--function-sections", "--filetype=obj", "-o", self.unified_file_core_obj])
             # fmt: on
 

diff --git a/python/requirements.txt b/python/requirements.txt
@@ -15,4 +15,4 @@ ml_dtypes
 matplotlib
 notebook
 pre-commit
-nanobind
+nanobind>=2.5
diff --git a/python/requirements_extras.txt b/python/requirements_extras.txt
@@ -1,3 +1,4 @@
  # This can't go in the normal requirements file because the way the wheels build parses requirements.txt
  # does not support github packages
-git+https://github.com/makslevental/mlir-python-extras@8aba4e7
+git+https://github.com/makslevental/mlir-python-extras@15cb803
+-f https://github.com/llvm/eudsl/releases/expanded_assets/latest
diff --git a/test/Integration/lit.local.cfg b/test/Integration/lit.local.cfg
@@ -10,7 +10,7 @@ if config.aie_include_integration_tests != 'ON':
 
 tosa_to_linalg = '--pass-pipeline="builtin.module(func.func(tosa-to-linalg-named, tosa-to-linalg, tosa-to-tensor))"'
 transform_linalg_for_aievec_cpp = '--dynamic-size-no-implicit-broadcast'
-linalg_to_affine = '--linalg-fuse-elementwise-ops --linalg-fold-unit-extent-dims --eliminate-empty-tensors --empty-tensor-to-alloc-tensor --one-shot-bufferize="allow-return-allocs-from-loops allow-unknown-ops bufferize-function-boundaries function-boundary-type-conversion=identity-layout-map unknown-type-conversion=identity-layout-map" --drop-equivalent-buffer-results --buffer-results-to-out-params --buffer-deallocation --canonicalize --cse --convert-linalg-to-affine-loops'
+linalg_to_affine = '--linalg-fuse-elementwise-ops --linalg-fold-unit-extent-dims --eliminate-empty-tensors --empty-tensor-to-alloc-tensor --one-shot-bufferize="allow-return-allocs-from-loops allow-unknown-ops bufferize-function-boundaries function-boundary-type-conversion=identity-layout-map unknown-type-conversion=identity-layout-map" --drop-equivalent-buffer-results --buffer-results-to-out-params --buffer-deallocation-pipeline --canonicalize --cse --convert-linalg-to-affine-loops'
 
 config.substitutions.append(('%tosa-to-linalg%', tosa_to_linalg))
 config.substitutions.append(('%linalg-to-vector-v64%', transform_linalg_for_aievec_cpp+' '+linalg_to_affine+' --affine-super-vectorize="virtual-vector-size=64"'))

diff --git a/test/Targets/AIEVecToCpp/polygeist_conv2d_i32.mlir b/test/Targets/AIEVecToCpp/polygeist_conv2d_i32.mlir
@@ -7,7 +7,7 @@
 // (c) Copyright 2024 Advanced Micro Devices, Inc. or its affiliates
 //
 //===----------------------------------------------------------------------===//
-// RUN: aie-opt --affine-loop-unroll="unroll-full unroll-full-threshold=3" --canonicalize -affine-super-vectorize="virtual-vector-size=8 vectorize-reductions" --aie-vectorize | aie-translate --aievec-to-cpp > %t.cpp
+// RUN: aie-opt -pass-pipeline="builtin.module(func.func(affine-loop-unroll{unroll-full=true unroll-full-threshold=3}), canonicalize, func.func(affine-super-vectorize{virtual-vector-size=8 vectorize-reductions}), aie-vectorize)" | aie-translate --aievec-to-cpp > %t.cpp
 
 module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<i64, dense<64> : vector<2xi32>>, #dlti.dl_entry<f80, dense<128> : vector<2xi32>>, #dlti.dl_entry<i1, dense<8> : vector<2xi32>>, #dlti.dl_entry<i8, dense<8> : vector<2xi32>>, #dlti.dl_entry<i16, dense<16> : vector<2xi32>>, #dlti.dl_entry<i32, dense<32> : vector<2xi32>>, #dlti.dl_entry<f16, dense<16> : vector<2xi32>>, #dlti.dl_entry<f64, dense<64> : vector<2xi32>>, #dlti.dl_entry<f128, dense<128> : vector<2xi32>>>, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} {
   func.func @conv2d(%arg0: memref<?x272xi32>, %arg1: memref<?x3xi32>, %arg2: memref<?x256xi32>) attributes {llvm.linkage = #llvm.linkage<external>} {
@@ -29,4 +29,4 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<"dlti.endianness"
     }
     return
   }
-}
+}
diff --git a/test/dialect/AIEVec/precanonicalization-aieml-llvmir.mlir b/test/dialect/AIEVec/precanonicalization-aieml-llvmir.mlir
@@ -15,7 +15,7 @@ func.func @scalar_extsi_to_broadcast_swap(%s: i8) -> vector<32xi32> {
 // CHECK-LABEL: @extsi_to_broadcast_swap(
 // CHECK-SAME: %[[VIN:.*]]: vector<8xi8>
 func.func @extsi_to_broadcast_swap(%v: vector<8xi8>) -> vector<4x8xi32> {
-    // CHECK: %[[ZV:.*]] = arith.constant dense<0> : vector<4x8xi8>
+    // CHECK: %[[ZV:.*]] = ub.poison : vector<4x8xi8>
     // CHECK: %[[I0:.*]] = vector.insert %[[VIN]], %[[ZV]] [0] : vector<8xi8> into vector<4x8xi8>
     // CHECK: %[[I1:.*]] = vector.insert %[[VIN]], %[[I0]] [1] : vector<8xi8> into vector<4x8xi8>
     // CHECK: %[[I2:.*]] = vector.insert %[[VIN]], %[[I1]] [2] : vector<8xi8> into vector<4x8xi8>
@@ -31,7 +31,7 @@ func.func @extsi_to_broadcast_swap(%v: vector<8xi8>) -> vector<4x8xi32> {
 // CHECK-LABEL: @broadcast_to_insert(
 // CHECK-SAME: %[[V:.*]]: vector<8xbf16>
 func.func @broadcast_to_insert(%v: vector<8xbf16>) -> vector<1x4x8xbf16> {
-    // CHECK: %[[ZV:.*]] = arith.constant dense<0.000000e+00> : vector<4x8xbf16>
+    // CHECK: %[[ZV:.*]] = ub.poison : vector<4x8xbf16>
     // CHECK: %[[I0:.*]] = vector.insert %[[V]], %[[ZV]] [0] : vector<8xbf16> into vector<4x8xbf16>
     // CHECK: %[[I1:.*]] = vector.insert %[[V]], %[[I0]] [1] : vector<8xbf16> into vector<4x8xbf16>
     // CHECK: %[[I2:.*]] = vector.insert %[[V]], %[[I1]] [2] : vector<8xbf16> into vector<4x8xbf16>

diff --git a/test/python/aievec.py b/test/python/aievec.py
@@ -267,7 +267,7 @@ def dut(A: T.tensor(1024, T.i8()), B: T.tensor(1024, T.i8())):
         )
         .drop_equivalent_buffer_results()
         .buffer_results_to_out_params()
-        .Func(p().buffer_deallocation())
+        .add_pass("buffer-deallocation-pipeline")
         .canonicalize()
         .cse()
         .convert_linalg_to_affine_loops()

diff --git a/utils/clone-llvm.sh b/utils/clone-llvm.sh
@@ -13,8 +13,8 @@
 ##===----------------------------------------------------------------------===##
 
 # The LLVM commit to use.
-LLVM_PROJECT_COMMIT=e6d16f93b329f2f9618d18d0b99c6060e206cf08
-DATETIME=2025013107
+LLVM_PROJECT_COMMIT=ec17161c6e87bb061daee4603ed21ab988f9d03d
+DATETIME=2025021418
 WHEEL_VERSION=21.0.0.$DATETIME+${LLVM_PROJECT_COMMIT:0:8}
 
 ############################################################################################

diff --git a/utils/mlir_aie_wheels/requirements.txt b/utils/mlir_aie_wheels/requirements.txt
@@ -8,4 +8,4 @@ pybind11[global]>=2.10.4
 rich
 setuptools>=42
 wheel
-nanobind>=2.4
+nanobind>=2.5