colin
/
onnx-mlir
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Support per-block bundling for static memory pools. (#325) 

* Reorganize main function.

* Follow review comments.

* Emit constants are globals in Krnl and LLVM dialects.

* Support the bundling on a per-block basis.

* Format.

* Fix test.

* Fix indent.

* Improve data structure.

* Format.

* Simplify maps for static pools.

* Format.

* Clean-up.
This commit is contained in:
Gheorghe-Teodor Bercea 2020-10-01 14:15:03 -04:00 committed by GitHub
parent 75930ffbcf
commit 0db735f48d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 276 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

88
src/Transform/BundleMemoryPools.cpp
View File

@ -44,6 +44,10 @@ typedef struct ONNXOperandsInitState {
// Helper data structure for the bundling of dynamic AllocOps. // Helper data structure for the bundling of dynamic AllocOps.
std::map<FuncOp, std::unique_ptr<ONNXOperandsInitState>> initMap; std::map<FuncOp, std::unique_ptr<ONNXOperandsInitState>> initMap;
// Handling of static memory pool on a block-basis in each function.
typedef std::map<Block *, AllocOp> BlockToStaticPool;
std::map<FuncOp, std::unique_ptr<BlockToStaticPool>> staticPoolMap;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// Helper functions. // Helper functions.
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
@ -153,11 +157,11 @@ KrnlGetRefOp getCurrentAllocGetRef(AllocOp *allocOp) {
/*! /*!
* RewritePattern that replaces: * RewritePattern that replaces:
* %mem1 = alloc() : memref<<dims1>x<type>> * %mempool = alloc() : memref<<dims1>x<type>>
* %mem2 = alloc() : memref<<dims2>x<type>> * %mem2 = alloc() : memref<<dims2>x<type>>
* %1 = krnl.getref %mem2 0 : memref<<dims2>x<type>> * %1 = krnl.getref %mem2 0 : memref<<dims2>x<type>>
* => * =>
* %mem1 = alloc() : memref<<dims1 + dims2>x<type>> * %mempool = alloc() : memref<<dims1 + dims2>x<type>>
* %1 = krnl.getref %mem1 <dims1> : memref<<dims2>x<type>> * %1 = krnl.getref %mem1 <dims1> : memref<<dims2>x<type>>
* *
* *
@ -171,7 +175,7 @@ KrnlGetRefOp getCurrentAllocGetRef(AllocOp *allocOp) {
* operations are part of memory pooling. * operations are part of memory pooling.
*/ */
class KrnlBundleMemoryPools : public OpRewritePattern<AllocOp> { class KrnlBundleStaticMemoryPools : public OpRewritePattern<AllocOp> {
public: public:
using OpRewritePattern<AllocOp>::OpRewritePattern; using OpRewritePattern<AllocOp>::OpRewritePattern;
@ -199,14 +203,47 @@ public:
if (memRefShape.size() != 1) if (memRefShape.size() != 1)
return failure(); return failure();
int64_t currentMemPoolSize = memRefShape[0]; FuncOp function = getContainingFunction(allocOp);
// Get a KrnlGetRefOp which does not use the current alloc. if (staticPoolMap.count(function) == 0) {
if (KrnlGetRefOp unbundledGetRef = getUnbundledGetRef(&allocOp)) { return failure();
// Make sure that this get ref uses a static alloc. }
auto unbundledGetRefType =
convertToMemRefType(unbundledGetRef.getResult().getType()); std::unique_ptr<BlockToStaticPool> &blockToStaticPool =
if (!hasAllConstantDimensions(unbundledGetRefType)) staticPoolMap.at(function);
// Get parent block.
Block *parentBlock = allocOp.getOperation()->getBlock();
if (blockToStaticPool->count(parentBlock) == 0) {
allocOp.getOperation()->moveBefore(&parentBlock->front());
// Create new entry in the block map.
blockToStaticPool->insert(
std::pair<Block *, AllocOp>(parentBlock, allocOp));
// This is the initial memory pool for this block and it is
// trivially bundled hence it's safe to return success.
return success();
}
// If this parent block has been found present in the map, it means
// a static memory bundle already exists. Fetch it.
AllocOp staticMemPoolAlloc = blockToStaticPool->at(parentBlock);
// If this is the alloc representing the memory pool and the function
// already has an init block, pattern matching must fail to avoid
// processing the dynamic memory pool a second time.
if (allocOp == staticMemPoolAlloc)
return failure();
auto staticMemPoolShape =
convertToMemRefType(staticMemPoolAlloc.getResult().getType())
.getShape();
int64_t currentMemPoolSize = staticMemPoolShape[0];
// Get the getref of the current allocOp. There is exactly one such getref.
KrnlGetRefOp currentAllocGetRef = getCurrentAllocGetRef(&allocOp);
if (!currentAllocGetRef)
return failure(); return failure();
// Current memory pool size is the offset for the newly bundled // Current memory pool size is the offset for the newly bundled
@ -216,8 +253,7 @@ public:
rewriter.getIntegerType(64), currentMemPoolSize)); rewriter.getIntegerType(64), currentMemPoolSize));
// Size in bytes of the output of the krnl.getref operation. // Size in bytes of the output of the krnl.getref operation.
int64_t unbundledTotalSize = int64_t unbundledTotalSize = memRefShape[0];
getMemRefSizeInBytes(unbundledGetRef.getResult());
// Compute new size. // Compute new size.
int64_t bundleTotalSize = unbundledTotalSize + currentMemPoolSize; int64_t bundleTotalSize = unbundledTotalSize + currentMemPoolSize;
@ -227,22 +263,26 @@ public:
newMemPoolShape.emplace_back(bundleTotalSize); newMemPoolShape.emplace_back(bundleTotalSize);
auto bundledMemPoolMemRefType = auto bundledMemPoolMemRefType =
MemRefType::get(newMemPoolShape, rewriter.getIntegerType(8)); MemRefType::get(newMemPoolShape, rewriter.getIntegerType(8));
auto bundledAlloc = auto newStaticMemPoolAlloc =
rewriter.create<AllocOp>(loc, bundledMemPoolMemRefType); rewriter.create<AllocOp>(loc, bundledMemPoolMemRefType);
// The newly bundled MemRef expressed as a KrnlGetRefOp. // The newly bundled MemRef expressed as a KrnlGetRefOp.
auto bundledMemRef = rewriter.create<KrnlGetRefOp>( auto bundledMemRef = rewriter.create<KrnlGetRefOp>(loc,
loc, unbundledGetRef.getResult().getType(), bundledAlloc, offset); currentAllocGetRef.getResult().getType(), newStaticMemPoolAlloc,
rewriter.replaceOp(unbundledGetRef, bundledMemRef.getResult()); offset);
rewriter.replaceOp(currentAllocGetRef, bundledMemRef.getResult());
// Replace old memory pool with new one. // Replace old memory pool with new one.
rewriter.replaceOp(allocOp, bundledAlloc.getResult()); rewriter.replaceOp(staticMemPoolAlloc, newStaticMemPoolAlloc.getResult());
// Update data structure to contain the newly constructed static memory
// pool.
blockToStaticPool->erase(parentBlock);
blockToStaticPool->insert(
std::pair<Block *, AllocOp>(parentBlock, newStaticMemPoolAlloc));
return success(); return success();
} }
return failure();
}
}; };
/*! /*!
@ -448,23 +488,25 @@ class KrnlBundleMemoryPoolsPass
public: public:
void runOnFunction() override { void runOnFunction() override {
auto function = getFunction(); auto function = getFunction();
// ModuleOp module = cast<ModuleOp>(function.getParentOp());
initMap.insert(std::pair<FuncOp, std::unique_ptr<ONNXOperandsInitState>>( initMap.insert(std::pair<FuncOp, std::unique_ptr<ONNXOperandsInitState>>(
function, std::make_unique<ONNXOperandsInitState>())); function, std::make_unique<ONNXOperandsInitState>()));
staticPoolMap.insert(std::pair<FuncOp, std::unique_ptr<BlockToStaticPool>>(
function, std::make_unique<BlockToStaticPool>()));
// Initialize state for this function. // Initialize state for this function.
std::unique_ptr<ONNXOperandsInitState> &initState = initMap.at(function); std::unique_ptr<ONNXOperandsInitState> &initState = initMap.at(function);
initState->initBlock = nullptr; initState->initBlock = nullptr;
ConversionTarget target(getContext()); ConversionTarget target(getContext());
OwningRewritePatternList patterns; OwningRewritePatternList patterns;
patterns.insert<KrnlBundleMemoryPools, KrnlBundleDynamicMemoryPools>( patterns.insert<KrnlBundleStaticMemoryPools, KrnlBundleDynamicMemoryPools>(
&getContext()); &getContext());
applyPatternsAndFoldGreedily(function, patterns); applyPatternsAndFoldGreedily(function, patterns);
initMap.erase(function); initMap.erase(function);
staticPoolMap.erase(function);
} }
}; };
} // namespace } // namespace

198
test/mlir/krnl/krnl_bundle_memory_pool.mlir
View File

@ -31,10 +31,10 @@ func @test_pool_bundling(%arg0: memref<10x10xf32>, %arg1: memref<10x20xf32>) ->
// CHECK-LABEL: test_pool_bundling // CHECK-LABEL: test_pool_bundling
// CHECK: [[CONST_0:%.+]] = constant 0 : i64 // CHECK: [[CONST_0:%.+]] = constant 0 : i64
// CHECK: [[CONST_CST:%.+]] = constant 0.000000e+00 : f32 // CHECK: [[CONST_CST:%.+]] = constant 0.000000e+00 : f32
// CHECK: [[CONST_400:%.+]] = constant 400 : i64
// CHECK: [[CONST_1200:%.+]] = constant 1200 : i64
// CHECK: [[CONST_2000:%.+]] = constant 2000 : i64
// CHECK: [[CONST_2400:%.+]] = constant 2400 : i64 // CHECK: [[CONST_2400:%.+]] = constant 2400 : i64
// CHECK: [[CONST_2000:%.+]] = constant 2000 : i64
// CHECK: [[CONST_1200:%.+]] = constant 1200 : i64
// CHECK: [[CONST_400:%.+]] = constant 400 : i64
// CHECK: [[RES:%.+]] = alloc() : memref<10x20xf32> // CHECK: [[RES:%.+]] = alloc() : memref<10x20xf32>
// CHECK: [[MEMPOOL:%.+]] = alloc() : memref<3200xi8> // CHECK: [[MEMPOOL:%.+]] = alloc() : memref<3200xi8>
// CHECK: [[MEMREF1:%.+]] = "krnl.getref"([[MEMPOOL]], [[CONST_2400]]) : (memref<3200xi8>, i64) -> memref<10x20xf32> // CHECK: [[MEMREF1:%.+]] = "krnl.getref"([[MEMPOOL]], [[CONST_2400]]) : (memref<3200xi8>, i64) -> memref<10x20xf32>
@ -149,12 +149,12 @@ func @test_dynamic_and_static_pool_bundling(%arg0: memref<?x?xf32>, %arg1: memre
return %15 : memref<?x10xf32> return %15 : memref<?x10xf32>
// CHECK-LABEL: test_dynamic_and_static_pool_bundling // CHECK-LABEL: test_dynamic_and_static_pool_bundling
// CHECK: [[C1200_I64:%.+]] = constant 1200 : i64
// CHECK: [[CST:%.+]] = constant 0.000000e+00 : f32 // CHECK: [[CST:%.+]] = constant 0.000000e+00 : f32
// CHECK: [[C0:%.+]] = constant 0 : index // CHECK: [[C0:%.+]] = constant 0 : index
// CHECK: [[C4:%.+]] = constant 4 : index // CHECK: [[C4:%.+]] = constant 4 : index
// CHECK: [[C10:%.+]] = constant 10 : index // CHECK: [[C10:%.+]] = constant 10 : index
// CHECK: [[C400_I64:%.+]] = constant 400 : i64 // CHECK: [[C400_I64:%.+]] = constant 400 : i64
// CHECK: [[C2000_I64:%.+]] = constant 2000 : i64
// CHECK: [[C0_I64:%.+]] = constant 0 : i64 // CHECK: [[C0_I64:%.+]] = constant 0 : i64
// CHECK: [[DIM:%.+]] = dim %arg0, [[C0]] : memref<?x?xf32> // CHECK: [[DIM:%.+]] = dim %arg0, [[C0]] : memref<?x?xf32>
// CHECK: [[SGT:%.+]] = cmpi "sgt", [[DIM]], [[DIM]] : index // CHECK: [[SGT:%.+]] = cmpi "sgt", [[DIM]], [[DIM]] : index
@ -169,17 +169,199 @@ func @test_dynamic_and_static_pool_bundling(%arg0: memref<?x?xf32>, %arg1: memre
// CHECK: [[DATA1:%.+]] = "krnl.getref"([[DYN_MEMPOOL]], [[OFFSET1_I64]]) : (memref<?xi8>, i64) -> memref<?x10xf32> // CHECK: [[DATA1:%.+]] = "krnl.getref"([[DYN_MEMPOOL]], [[OFFSET1_I64]]) : (memref<?xi8>, i64) -> memref<?x10xf32>
// CHECK: [[DATA2:%.+]] = "krnl.getref"([[DYN_MEMPOOL]], [[C0_I64]]) : (memref<?xi8>, i64) -> memref<?x10xf32> // CHECK: [[DATA2:%.+]] = "krnl.getref"([[DYN_MEMPOOL]], [[C0_I64]]) : (memref<?xi8>, i64) -> memref<?x10xf32>
// CHECK: [[STATIC_MEMPOOL:%.+]] = alloc() : memref<2800xi8> // CHECK: [[STATIC_MEMPOOL:%.+]] = alloc() : memref<2800xi8>
// CHECK: [[DATA3:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C2000_I64]]) : (memref<2800xi8>, i64) -> memref<10x20xf32> // CHECK: [[DATA3:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C1200_I64]]) : (memref<2800xi8>, i64) -> memref<10x40xf32>
// CHECK: [[DATA4:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C400_I64]]) : (memref<2800xi8>, i64) -> memref<10x40xf32> // CHECK: [[DATA4:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C400_I64]]) : (memref<2800xi8>, i64) -> memref<10x20xf32>
// CHECK: [[DATA5:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C0_I64]]) : (memref<2800xi8>, i64) -> memref<10x10xf32> // CHECK: [[DATA5:%.+]] = "krnl.getref"([[STATIC_MEMPOOL]], [[C0_I64]]) : (memref<2800xi8>, i64) -> memref<10x10xf32>
// CHECK: [[RES:%.+]] = alloc([[DIM]]) : memref<?x10xf32> // CHECK: [[RES:%.+]] = alloc([[DIM]]) : memref<?x10xf32>
// CHECK: affine.store [[CST]], [[DATA1]][0, 0] : memref<?x10xf32> // CHECK: affine.store [[CST]], [[DATA1]][0, 0] : memref<?x10xf32>
// CHECK: affine.store [[CST]], [[DATA2]][0, 0] : memref<?x10xf32> // CHECK: affine.store [[CST]], [[DATA2]][0, 0] : memref<?x10xf32>
// CHECK: affine.store [[CST]], [[RES]][0, 0] : memref<?x10xf32> // CHECK: affine.store [[CST]], [[RES]][0, 0] : memref<?x10xf32>
// CHECK: affine.store [[CST]], [[DATA5]][0, 0] : memref<10x10xf32> // CHECK: affine.store [[CST]], [[DATA5]][0, 0] : memref<10x10xf32>
// CHECK: affine.store [[CST]], [[DATA3]][0, 0] : memref<10x20xf32> // CHECK: affine.store [[CST]], [[DATA4]][0, 0] : memref<10x20xf32>
// CHECK: affine.store [[CST]], [[DATA4]][0, 0] : memref<10x40xf32> // CHECK: affine.store [[CST]], [[DATA3]][0, 0] : memref<10x40xf32>
// CHECK: dealloc [[DYN_MEMPOOL]] : memref<?xi8> // CHECK: dealloc [[DYN_MEMPOOL]] : memref<?xi8>
// CHECK: dealloc [[STATIC_MEMPOOL]] : memref<2800xi8> // CHECK: dealloc [[STATIC_MEMPOOL]] : memref<2800xi8>
// CHECK: return [[RES]] : memref<?x10xf32> // CHECK: return [[RES]] : memref<?x10xf32>
} }
/// Test bundling inside a sub-block.
func @static_mem_pool_rnn_subblock(%arg0: memref<1x3x2xf32>, %arg1: memref<1x4x2xf32>, %arg2: memref<1x4x4xf32>) -> memref<1x3x4xf32> attributes {input_names = ["X", "W", "R"], output_names = ["Y"]} {
%cst = constant 0.000000e+00 : f32
%c0_i64 = constant 0 : i64
%0 = alloc() : memref<1x3x4xf32>
%2 = krnl.define_loops 1
krnl.iterate(%2) with (%2 -> %arg3 = 0 to 1) {
%3:2 = krnl.define_loops 2
krnl.iterate(%3#0, %3#1) with (%3#0 -> %arg4 = 0 to 3, %3#1 -> %arg5 = 0 to 4) {
%4 = alloc() : memref<4xi8>
%5 = "krnl.getref"(%4, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
%6 = affine.load %0[0, symbol(%arg4), symbol(%arg5)] : memref<1x3x4xf32>
%7 = alloc() : memref<4xi8>
%8 = "krnl.getref"(%7, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %cst, %8[] : memref<f32>
%9 = alloc() : memref<4xi8>
%10 = "krnl.getref"(%9, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %cst, %10[] : memref<f32>
%11 = krnl.define_loops 1
krnl.iterate(%11) with (%11 -> %arg6 = 0 to 2) {
%25 = affine.load %arg0[symbol(%arg3), symbol(%arg4), symbol(%arg6)] : memref<1x3x2xf32>
%26 = affine.load %arg1[0, symbol(%arg5), symbol(%arg6)] : memref<1x4x2xf32>
%27 = mulf %25, %26 : f32
%28 = affine.load %8[] : memref<f32>
%29 = addf %28, %27 : f32
affine.store %29, %8[] : memref<f32>
%30 = affine.load %arg2[0, symbol(%arg5), symbol(%arg6)] : memref<1x4x4xf32>
%31 = mulf %6, %30 : f32
%32 = affine.load %10[] : memref<f32>
%33 = addf %32, %31 : f32
affine.store %33, %10[] : memref<f32>
}
%12 = affine.load %8[] : memref<f32>
%13 = affine.load %10[] : memref<f32>
%14 = addf %12, %13 : f32
%15 = alloc() : memref<4xi8>
%16 = "krnl.getref"(%15, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %14, %16[] : memref<f32>
%17 = affine.load %16[] : memref<f32>
%18 = subf %cst, %17 : f32
%19 = exp %17 : f32
%20 = exp %18 : f32
%21 = subf %19, %20 : f32
%22 = addf %19, %20 : f32
%23 = divf %21, %22 : f32
affine.store %23, %5[] : memref<f32>
%24 = affine.load %5[] : memref<f32>
affine.store %24, %0[0, symbol(%arg4), symbol(%arg5)] : memref<1x3x4xf32>
dealloc %15 : memref<4xi8>
dealloc %9 : memref<4xi8>
dealloc %7 : memref<4xi8>
dealloc %4 : memref<4xi8>
}
}
return %0 : memref<1x3x4xf32>
// CHECK-LABEL: static_mem_pool_rnn_subblock
// CHECK: [[CST:%.+]] = constant 0.000000e+00 : f32
// CHECK: [[C0:%.+]] = constant 0 : i64
// CHECK: [[C12:%.+]] = constant 12 : i64
// CHECK: [[C8:%.+]] = constant 8 : i64
// CHECK: [[C4:%.+]] = constant 4 : i64
// CHECK: [[RES:%.+]] = alloc() : memref<1x3x4xf32>
// CHECK: krnl.define_loops 1
// CHECK: krnl.iterate
// CHECK: krnl.define_loops 2
// CHECK: krnl.iterate
// CHECK: [[STATIC_MEM_POOL:%.+]] = alloc() : memref<16xi8>
// CHECK: [[REF1:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C12]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.load
// CHECK: [[REF2:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C8]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.store
// CHECK: [[REF3:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C4]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.store
// CHECK: krnl.define_loops 1
// CHECK: krnl.iterate
// CHECK: [[REF4:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C0]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: dealloc [[STATIC_MEM_POOL]] : memref<16xi8>
// CHECK: return [[RES]] : memref<1x3x4xf32>
}
/// Test bundling inside a sub-block and in the main block.
func @static_mem_pool_rnn_sub_and_main_block(%arg0: memref<1x3x2xf32>, %arg1: memref<1x4x2xf32>, %arg2: memref<1x4x4xf32>) -> memref<1x3x4xf32> attributes {input_names = ["X", "W", "R"], output_names = ["Y"]} {
%cst = constant 0.000000e+00 : f32
%c0_i64 = constant 0 : i64
%0 = alloc() : memref<1x3x4xf32>
%mem0 = alloc() : memref<4xi8>
%ref0 = "krnl.getref"(%mem0, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
%mem1 = alloc() : memref<4xi8>
%ref1 = "krnl.getref"(%mem1, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
%2 = krnl.define_loops 1
krnl.iterate(%2) with (%2 -> %arg3 = 0 to 1) {
%3:2 = krnl.define_loops 2
krnl.iterate(%3#0, %3#1) with (%3#0 -> %arg4 = 0 to 3, %3#1 -> %arg5 = 0 to 4) {
%4 = alloc() : memref<4xi8>
%5 = "krnl.getref"(%4, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
%6 = affine.load %0[0, symbol(%arg4), symbol(%arg5)] : memref<1x3x4xf32>
%7 = alloc() : memref<4xi8>
%8 = "krnl.getref"(%7, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %cst, %8[] : memref<f32>
%9 = alloc() : memref<4xi8>
%10 = "krnl.getref"(%9, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %cst, %10[] : memref<f32>
%11 = krnl.define_loops 1
krnl.iterate(%11) with (%11 -> %arg6 = 0 to 2) {
%25 = affine.load %arg0[symbol(%arg3), symbol(%arg4), symbol(%arg6)] : memref<1x3x2xf32>
%26 = affine.load %arg1[0, symbol(%arg5), symbol(%arg6)] : memref<1x4x2xf32>
%27 = mulf %25, %26 : f32
%28 = affine.load %8[] : memref<f32>
%29 = addf %28, %27 : f32
affine.store %29, %8[] : memref<f32>
%30 = affine.load %arg2[0, symbol(%arg5), symbol(%arg6)] : memref<1x4x4xf32>
%31 = mulf %6, %30 : f32
%32 = affine.load %10[] : memref<f32>
%33 = addf %32, %31 : f32
affine.store %33, %10[] : memref<f32>
affine.store %33, %ref0[] : memref<f32>
}
%12 = affine.load %8[] : memref<f32>
%13 = affine.load %10[] : memref<f32>
%14 = addf %12, %13 : f32
%15 = alloc() : memref<4xi8>
%16 = "krnl.getref"(%15, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
affine.store %14, %16[] : memref<f32>
%17 = affine.load %16[] : memref<f32>
%18 = subf %cst, %17 : f32
%19 = exp %17 : f32
%20 = exp %18 : f32
%21 = subf %19, %20 : f32
%22 = addf %19, %20 : f32
%23 = divf %21, %22 : f32
affine.store %23, %5[] : memref<f32>
%24 = affine.load %5[] : memref<f32>
affine.store %24, %0[0, symbol(%arg4), symbol(%arg5)] : memref<1x3x4xf32>
affine.store %24, %ref1[] : memref<f32>
dealloc %15 : memref<4xi8>
dealloc %9 : memref<4xi8>
dealloc %7 : memref<4xi8>
dealloc %4 : memref<4xi8>
}
}
%mem2 = alloc() : memref<4xi8>
%ref2 = "krnl.getref"(%mem2, %c0_i64) : (memref<4xi8>, i64) -> memref<f32>
%val = affine.load %ref1[] : memref<f32>
affine.store %val, %ref2[] : memref<f32>
dealloc %mem2 : memref<4xi8>
dealloc %mem1 : memref<4xi8>
dealloc %mem0 : memref<4xi8>
return %0 : memref<1x3x4xf32>
// CHECK-LABEL: static_mem_pool_rnn_sub_and_main_block
// CHECK: [[CST:%.+]] = constant 0.000000e+00 : f32
// CHECK: [[C0:%.+]] = constant 0 : i64
// CHECK: [[C12:%.+]] = constant 12 : i64
// CHECK: [[C8:%.+]] = constant 8 : i64
// CHECK: [[C4:%.+]] = constant 4 : i64
// CHECK: [[RES:%.+]] = alloc() : memref<1x3x4xf32>
// CHECK: [[STATIC_MEM_POOL_MAIN:%.+]] = alloc() : memref<12xi8>
// CHECK: [[MAIN_REF_0:%.+]] = "krnl.getref"([[STATIC_MEM_POOL_MAIN]], [[C8]]) : (memref<12xi8>, i64) -> memref<f32>
// CHECK: [[MAIN_REF_1:%.+]] = "krnl.getref"([[STATIC_MEM_POOL_MAIN]], [[C4]]) : (memref<12xi8>, i64) -> memref<f32>
// CHECK: krnl.define_loops 1
// CHECK: krnl.iterate
// CHECK: krnl.define_loops 2
// CHECK: krnl.iterate
// CHECK: [[STATIC_MEM_POOL:%.+]] = alloc() : memref<16xi8>
// CHECK: [[REF1:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C12]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.load
// CHECK: [[REF2:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C8]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.store
// CHECK: [[REF3:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C4]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: affine.store
// CHECK: krnl.define_loops 1
// CHECK: krnl.iterate
// CHECK: [[REF4:%.+]] = "krnl.getref"([[STATIC_MEM_POOL]], [[C0]]) : (memref<16xi8>, i64) -> memref<f32>
// CHECK: dealloc [[STATIC_MEM_POOL]] : memref<16xi8>
// CHECK: [[MAIN_REF_2:%.+]] = "krnl.getref"([[STATIC_MEM_POOL_MAIN]], [[C0]]) : (memref<12xi8>, i64) -> memref<f32>
// CHECK: [[LOAD:%.+]] = affine.load [[MAIN_REF_1]][] : memref<f32>
// CHECK: affine.store [[LOAD]], [[MAIN_REF_2]][] : memref<f32>
// CHECK: dealloc [[STATIC_MEM_POOL_MAIN]] : memref<12xi8>
// CHECK: return [[RES]] : memref<1x3x4xf32>
}

6
test/mlir/onnx/onnx_bundle_memory_pool.mlir
View File

@ -12,10 +12,10 @@ func @test_bundle_memory_pool(%arg0: tensor<10x10xf32>, %arg1: tensor<10x20xf32>
// CHECK-LABEL: test_bundle_memory_pool // CHECK-LABEL: test_bundle_memory_pool
// CHECK: [[CONST0:%.+]] = constant 0 : i64 // CHECK: [[CONST0:%.+]] = constant 0 : i64
// CHECK: [[CONST00:%.+]] = constant 0.000000e+00 : f32 // CHECK: [[CONST00:%.+]] = constant 0.000000e+00 : f32
// CHECK: [[CONST400:%.+]] = constant 400 : i64
// CHECK: [[CONST1200:%.+]] = constant 1200 : i64
// CHECK: [[CONST2000:%.+]] = constant 2000 : i64
// CHECK: [[CONST2400:%.+]] = constant 2400 : i64 // CHECK: [[CONST2400:%.+]] = constant 2400 : i64
// CHECK: [[CONST2000:%.+]] = constant 2000 : i64
// CHECK: [[CONST1200:%.+]] = constant 1200 : i64
// CHECK: [[CONST400:%.+]] = constant 400 : i64
// CHECK: [[RES:%.+]] = alloc() : memref<10x20xf32> // CHECK: [[RES:%.+]] = alloc() : memref<10x20xf32>
// CHECK: [[MEMPOOL:%.+]] = alloc() : memref<3200xi8> // CHECK: [[MEMPOOL:%.+]] = alloc() : memref<3200xi8>
// CHECK: "krnl.getref"([[MEMPOOL]], [[CONST2400]]) : (memref<3200xi8>, i64) -> memref<10x20xf32> // CHECK: "krnl.getref"([[MEMPOOL]], [[CONST2400]]) : (memref<3200xi8>, i64) -> memref<10x20xf32>