[mlir][sparse][CRunnerUtils] Add shuffle in CRunnerUtils #77124

yinying-lisa-li · 2024-01-05T18:43:28Z

Shuffle can generate an array of unique and random numbers from 0 to size-1. It can be used to generate tensors with specified sparsity level.

llvmbot · 2024-01-05T18:54:21Z

@llvm/pr-subscribers-mlir-sparse

Author: Yinying Li (yinying-lisa-li)

Changes

Shuffle can generate an array of unique and random numbers from 0 to size-1. It can be used to generate tensors with specified sparsity level.

Full diff: https://github.com/llvm/llvm-project/pull/77124.diff

3 Files Affected:

(modified) mlir/include/mlir/ExecutionEngine/CRunnerUtils.h (+4)
(modified) mlir/lib/ExecutionEngine/CRunnerUtils.cpp (+16)
(added) mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir (+108)

diff --git a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
index 76b04145b482e4..747e5ca40ca6f6 100644
--- a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
+++ b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
@@ -486,6 +486,10 @@ extern "C" MLIR_CRUNNERUTILS_EXPORT void *rtsrand(uint64_t s);
 extern "C" MLIR_CRUNNERUTILS_EXPORT uint64_t rtrand(void *, uint64_t m);
 // Deletes the random number generator.
 extern "C" MLIR_CRUNNERUTILS_EXPORT void rtdrand(void *);
+// Returns a pointer to an array of random numbers in the range of [0, s).
+extern "C" MLIR_CRUNNERUTILS_EXPORT void *shuffle(uint64_t s, void *g);
+// Deletes the array of random numbers.
+extern "C" MLIR_CRUNNERUTILS_EXPORT void shuffleFree(void *a);
 
 //===----------------------------------------------------------------------===//
 // Runtime support library to allow the use of std::sort in MLIR program.
diff --git a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
index e28e75eb110303..3a3261d1ad4e03 100644
--- a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
+++ b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
@@ -160,6 +160,22 @@ extern "C" void mlirAlignedFree(void *ptr) {
 #endif
 }
 
+/// Generates an array with unique and random numbers from 0 to s-1.
+extern "C" void *shuffle(uint64_t s, void *g) {
+  std::mt19937 *generator = static_cast<std::mt19937 *>(g);
+  uint64_t *output = new uint64_t[s];
+  std::vector<uint64_t> arr(s);
+  std::iota(arr.begin(), arr.end(), 0);
+  std::shuffle(arr.begin(), arr.end(), *generator);
+  std::copy(arr.begin(), arr.end(), output);
+  return output;
+}
+
+extern "C" void shuffleFree(void *a) {
+  uint64_t *arr = static_cast<uint64_t *>(a);
+  delete[] arr;
+}
+
 extern "C" void *rtsrand(uint64_t s) {
   // Standard mersenne_twister_engine seeded with s.
   return new std::mt19937(s);
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
new file mode 100644
index 00000000000000..b2bf3eb74e8f52
--- /dev/null
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
@@ -0,0 +1,108 @@
+//--------------------------------------------------------------------------------------------------
+// WHEN CREATING A NEW TEST, PLEASE JUST COPY & PASTE WITHOUT EDITS.
+//
+// Set-up that's shared across all tests in this directory. In principle, this
+// config could be moved to lit.local.cfg. However, there are downstream users that
+//  do not use these LIT config files. Hence why this is kept inline.
+//
+// DEFINE: %{sparsifier_opts} = enable-runtime-library=true
+// DEFINE: %{sparsifier_opts_sve} = enable-arm-sve=true %{sparsifier_opts}
+// DEFINE: %{compile} = mlir-opt %s --sparsifier="%{sparsifier_opts}"
+// DEFINE: %{compile_sve} = mlir-opt %s --sparsifier="%{sparsifier_opts_sve}"
+// DEFINE: %{run_libs} = -shared-libs=%mlir_c_runner_utils,%mlir_runner_utils
+// DEFINE: %{run_opts} = -e entry -entry-point-result=void
+// DEFINE: %{run} = mlir-cpu-runner %{run_opts} %{run_libs}
+// DEFINE: %{run_sve} = %mcr_aarch64_cmd --march=aarch64 --mattr="+sve" %{run_opts} %{run_libs}
+//
+// DEFINE: %{env} =
+//--------------------------------------------------------------------------------------------------
+
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and vectorization.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and VLA vectorization.
+// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}
+
+//
+// Integration test that generates a tensor with specified sparsity level.
+//
+
+!Generator = !llvm.ptr
+!Array = !llvm.ptr
+
+#SparseVector = #sparse_tensor.encoding<{
+  map = (d0) -> (d0 : compressed)
+}>
+
+module {
+  func.func private @rtsrand(index) -> (!Generator)
+  func.func private @rtrand(!Generator, index) -> (index)
+  func.func private @rtdrand(!Generator) -> ()
+  func.func private @shuffle(index, !Generator) -> (!Array)
+  func.func private @shuffleFree(!Array) -> ()
+
+  //
+  // Main driver.
+  //
+  func.func @entry() {
+    %c0 = arith.constant 0 : index
+    %c1 = arith.constant 1 : index
+    %f0 = arith.constant 0.0 : f64
+    %c99 = arith.constant 99 : index
+    %c100 = arith.constant 100 : index
+
+    // Set up input size and sparsity level.
+    %size = arith.constant 50 : index
+    %sparsity = arith.constant 90 : index
+    %zeros = arith.muli %size, %sparsity : index
+    %nz = arith.floordivsi %zeros, %c100 : index
+    %nse = arith.subi %size, %nz : index
+
+    // Set up an empty vector.
+    %empty = tensor.empty(%size) : tensor<?xf64>
+    %zero_vec = linalg.fill ins(%f0 : f64) outs(%empty : tensor<?xf64>) -> tensor<?xf64>
+
+    // Generate shuffled indices in the range of [0, %size).
+    %g = func.call @rtsrand(%c0) : (index) ->(!Generator)
+    %res = func.call @shuffle(%size, %g) : (index, !Generator) -> !Array
+
+    // Iterate through the number of nse indices to insert values.
+    %output = scf.for %iv = %c0 to %nse step %c1 iter_args(%iter = %zero_vec) -> tensor<?xf64> {
+      // Fetch the index to insert value from shuffled index array.
+      %r = arith.index_cast %iv : index to i64
+      %arr = llvm.getelementptr %res[%r] : (!llvm.ptr, i64) -> !llvm.ptr, i64
+      %val = llvm.load %arr : !Array -> i64
+      %idx = arith.index_cast %val : i64 to index
+      // Generate a random number from 1 to 100.
+      %ri0 = func.call @rtrand(%g, %c99) : (!Generator, index) -> (index)
+      %ri1 = arith.addi %ri0, %c1 : index
+      %r0 = arith.index_cast %ri1 : index to i64
+      %fr = arith.uitofp %r0 : i64 to f64
+      // Insert the random number to current index.
+      %out = tensor.insert %fr into %iter[%idx] : tensor<?xf64>
+      scf.yield %out : tensor<?xf64>
+    }
+
+    %sv = sparse_tensor.convert %output : tensor<?xf64> to tensor<?xf64, #SparseVector>
+    %n0 = sparse_tensor.number_of_entries %sv : tensor<?xf64, #SparseVector>
+
+    // Print the number of non-zeros for verification.
+    //
+    // CHECK: 5
+    vector.print %n0 : index
+
+    // Release the resources.
+    bufferization.dealloc_tensor %sv : tensor<?xf64, #SparseVector>
+    func.call @shuffleFree(%res) : (!Array) -> ()
+    func.call @rtdrand(%g) : (!Generator) -> ()
+
+    return
+  }
+}

llvmbot · 2024-01-05T18:54:21Z

@llvm/pr-subscribers-mlir

Author: Yinying Li (yinying-lisa-li)

Changes

Shuffle can generate an array of unique and random numbers from 0 to size-1. It can be used to generate tensors with specified sparsity level.

Full diff: https://github.com/llvm/llvm-project/pull/77124.diff

3 Files Affected:

(modified) mlir/include/mlir/ExecutionEngine/CRunnerUtils.h (+4)
(modified) mlir/lib/ExecutionEngine/CRunnerUtils.cpp (+16)
(added) mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir (+108)

diff --git a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
index 76b04145b482e4..747e5ca40ca6f6 100644
--- a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
+++ b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
@@ -486,6 +486,10 @@ extern "C" MLIR_CRUNNERUTILS_EXPORT void *rtsrand(uint64_t s);
 extern "C" MLIR_CRUNNERUTILS_EXPORT uint64_t rtrand(void *, uint64_t m);
 // Deletes the random number generator.
 extern "C" MLIR_CRUNNERUTILS_EXPORT void rtdrand(void *);
+// Returns a pointer to an array of random numbers in the range of [0, s).
+extern "C" MLIR_CRUNNERUTILS_EXPORT void *shuffle(uint64_t s, void *g);
+// Deletes the array of random numbers.
+extern "C" MLIR_CRUNNERUTILS_EXPORT void shuffleFree(void *a);
 
 //===----------------------------------------------------------------------===//
 // Runtime support library to allow the use of std::sort in MLIR program.
diff --git a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
index e28e75eb110303..3a3261d1ad4e03 100644
--- a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
+++ b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
@@ -160,6 +160,22 @@ extern "C" void mlirAlignedFree(void *ptr) {
 #endif
 }
 
+/// Generates an array with unique and random numbers from 0 to s-1.
+extern "C" void *shuffle(uint64_t s, void *g) {
+  std::mt19937 *generator = static_cast<std::mt19937 *>(g);
+  uint64_t *output = new uint64_t[s];
+  std::vector<uint64_t> arr(s);
+  std::iota(arr.begin(), arr.end(), 0);
+  std::shuffle(arr.begin(), arr.end(), *generator);
+  std::copy(arr.begin(), arr.end(), output);
+  return output;
+}
+
+extern "C" void shuffleFree(void *a) {
+  uint64_t *arr = static_cast<uint64_t *>(a);
+  delete[] arr;
+}
+
 extern "C" void *rtsrand(uint64_t s) {
   // Standard mersenne_twister_engine seeded with s.
   return new std::mt19937(s);
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
new file mode 100644
index 00000000000000..b2bf3eb74e8f52
--- /dev/null
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
@@ -0,0 +1,108 @@
+//--------------------------------------------------------------------------------------------------
+// WHEN CREATING A NEW TEST, PLEASE JUST COPY & PASTE WITHOUT EDITS.
+//
+// Set-up that's shared across all tests in this directory. In principle, this
+// config could be moved to lit.local.cfg. However, there are downstream users that
+//  do not use these LIT config files. Hence why this is kept inline.
+//
+// DEFINE: %{sparsifier_opts} = enable-runtime-library=true
+// DEFINE: %{sparsifier_opts_sve} = enable-arm-sve=true %{sparsifier_opts}
+// DEFINE: %{compile} = mlir-opt %s --sparsifier="%{sparsifier_opts}"
+// DEFINE: %{compile_sve} = mlir-opt %s --sparsifier="%{sparsifier_opts_sve}"
+// DEFINE: %{run_libs} = -shared-libs=%mlir_c_runner_utils,%mlir_runner_utils
+// DEFINE: %{run_opts} = -e entry -entry-point-result=void
+// DEFINE: %{run} = mlir-cpu-runner %{run_opts} %{run_libs}
+// DEFINE: %{run_sve} = %mcr_aarch64_cmd --march=aarch64 --mattr="+sve" %{run_opts} %{run_libs}
+//
+// DEFINE: %{env} =
+//--------------------------------------------------------------------------------------------------
+
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and vectorization.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and VLA vectorization.
+// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}
+
+//
+// Integration test that generates a tensor with specified sparsity level.
+//
+
+!Generator = !llvm.ptr
+!Array = !llvm.ptr
+
+#SparseVector = #sparse_tensor.encoding<{
+  map = (d0) -> (d0 : compressed)
+}>
+
+module {
+  func.func private @rtsrand(index) -> (!Generator)
+  func.func private @rtrand(!Generator, index) -> (index)
+  func.func private @rtdrand(!Generator) -> ()
+  func.func private @shuffle(index, !Generator) -> (!Array)
+  func.func private @shuffleFree(!Array) -> ()
+
+  //
+  // Main driver.
+  //
+  func.func @entry() {
+    %c0 = arith.constant 0 : index
+    %c1 = arith.constant 1 : index
+    %f0 = arith.constant 0.0 : f64
+    %c99 = arith.constant 99 : index
+    %c100 = arith.constant 100 : index
+
+    // Set up input size and sparsity level.
+    %size = arith.constant 50 : index
+    %sparsity = arith.constant 90 : index
+    %zeros = arith.muli %size, %sparsity : index
+    %nz = arith.floordivsi %zeros, %c100 : index
+    %nse = arith.subi %size, %nz : index
+
+    // Set up an empty vector.
+    %empty = tensor.empty(%size) : tensor<?xf64>
+    %zero_vec = linalg.fill ins(%f0 : f64) outs(%empty : tensor<?xf64>) -> tensor<?xf64>
+
+    // Generate shuffled indices in the range of [0, %size).
+    %g = func.call @rtsrand(%c0) : (index) ->(!Generator)
+    %res = func.call @shuffle(%size, %g) : (index, !Generator) -> !Array
+
+    // Iterate through the number of nse indices to insert values.
+    %output = scf.for %iv = %c0 to %nse step %c1 iter_args(%iter = %zero_vec) -> tensor<?xf64> {
+      // Fetch the index to insert value from shuffled index array.
+      %r = arith.index_cast %iv : index to i64
+      %arr = llvm.getelementptr %res[%r] : (!llvm.ptr, i64) -> !llvm.ptr, i64
+      %val = llvm.load %arr : !Array -> i64
+      %idx = arith.index_cast %val : i64 to index
+      // Generate a random number from 1 to 100.
+      %ri0 = func.call @rtrand(%g, %c99) : (!Generator, index) -> (index)
+      %ri1 = arith.addi %ri0, %c1 : index
+      %r0 = arith.index_cast %ri1 : index to i64
+      %fr = arith.uitofp %r0 : i64 to f64
+      // Insert the random number to current index.
+      %out = tensor.insert %fr into %iter[%idx] : tensor<?xf64>
+      scf.yield %out : tensor<?xf64>
+    }
+
+    %sv = sparse_tensor.convert %output : tensor<?xf64> to tensor<?xf64, #SparseVector>
+    %n0 = sparse_tensor.number_of_entries %sv : tensor<?xf64, #SparseVector>
+
+    // Print the number of non-zeros for verification.
+    //
+    // CHECK: 5
+    vector.print %n0 : index
+
+    // Release the resources.
+    bufferization.dealloc_tensor %sv : tensor<?xf64, #SparseVector>
+    func.call @shuffleFree(%res) : (!Array) -> ()
+    func.call @rtdrand(%g) : (!Generator) -> ()
+
+    return
+  }
+}

llvmbot · 2024-01-05T18:54:22Z

@llvm/pr-subscribers-mlir-execution-engine

Author: Yinying Li (yinying-lisa-li)

Changes

Shuffle can generate an array of unique and random numbers from 0 to size-1. It can be used to generate tensors with specified sparsity level.

Full diff: https://github.com/llvm/llvm-project/pull/77124.diff

3 Files Affected:

(modified) mlir/include/mlir/ExecutionEngine/CRunnerUtils.h (+4)
(modified) mlir/lib/ExecutionEngine/CRunnerUtils.cpp (+16)
(added) mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir (+108)

diff --git a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
index 76b04145b482e4..747e5ca40ca6f6 100644
--- a/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
+++ b/mlir/include/mlir/ExecutionEngine/CRunnerUtils.h
@@ -486,6 +486,10 @@ extern "C" MLIR_CRUNNERUTILS_EXPORT void *rtsrand(uint64_t s);
 extern "C" MLIR_CRUNNERUTILS_EXPORT uint64_t rtrand(void *, uint64_t m);
 // Deletes the random number generator.
 extern "C" MLIR_CRUNNERUTILS_EXPORT void rtdrand(void *);
+// Returns a pointer to an array of random numbers in the range of [0, s).
+extern "C" MLIR_CRUNNERUTILS_EXPORT void *shuffle(uint64_t s, void *g);
+// Deletes the array of random numbers.
+extern "C" MLIR_CRUNNERUTILS_EXPORT void shuffleFree(void *a);
 
 //===----------------------------------------------------------------------===//
 // Runtime support library to allow the use of std::sort in MLIR program.
diff --git a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
index e28e75eb110303..3a3261d1ad4e03 100644
--- a/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
+++ b/mlir/lib/ExecutionEngine/CRunnerUtils.cpp
@@ -160,6 +160,22 @@ extern "C" void mlirAlignedFree(void *ptr) {
 #endif
 }
 
+/// Generates an array with unique and random numbers from 0 to s-1.
+extern "C" void *shuffle(uint64_t s, void *g) {
+  std::mt19937 *generator = static_cast<std::mt19937 *>(g);
+  uint64_t *output = new uint64_t[s];
+  std::vector<uint64_t> arr(s);
+  std::iota(arr.begin(), arr.end(), 0);
+  std::shuffle(arr.begin(), arr.end(), *generator);
+  std::copy(arr.begin(), arr.end(), output);
+  return output;
+}
+
+extern "C" void shuffleFree(void *a) {
+  uint64_t *arr = static_cast<uint64_t *>(a);
+  delete[] arr;
+}
+
 extern "C" void *rtsrand(uint64_t s) {
   // Standard mersenne_twister_engine seeded with s.
   return new std::mt19937(s);
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
new file mode 100644
index 00000000000000..b2bf3eb74e8f52
--- /dev/null
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir
@@ -0,0 +1,108 @@
+//--------------------------------------------------------------------------------------------------
+// WHEN CREATING A NEW TEST, PLEASE JUST COPY & PASTE WITHOUT EDITS.
+//
+// Set-up that's shared across all tests in this directory. In principle, this
+// config could be moved to lit.local.cfg. However, there are downstream users that
+//  do not use these LIT config files. Hence why this is kept inline.
+//
+// DEFINE: %{sparsifier_opts} = enable-runtime-library=true
+// DEFINE: %{sparsifier_opts_sve} = enable-arm-sve=true %{sparsifier_opts}
+// DEFINE: %{compile} = mlir-opt %s --sparsifier="%{sparsifier_opts}"
+// DEFINE: %{compile_sve} = mlir-opt %s --sparsifier="%{sparsifier_opts_sve}"
+// DEFINE: %{run_libs} = -shared-libs=%mlir_c_runner_utils,%mlir_runner_utils
+// DEFINE: %{run_opts} = -e entry -entry-point-result=void
+// DEFINE: %{run} = mlir-cpu-runner %{run_opts} %{run_libs}
+// DEFINE: %{run_sve} = %mcr_aarch64_cmd --march=aarch64 --mattr="+sve" %{run_opts} %{run_libs}
+//
+// DEFINE: %{env} =
+//--------------------------------------------------------------------------------------------------
+
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and vectorization.
+// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true
+// RUN: %{compile} | %{run} | FileCheck %s
+//
+// Do the same run, but now with direct IR generation and VLA vectorization.
+// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}
+
+//
+// Integration test that generates a tensor with specified sparsity level.
+//
+
+!Generator = !llvm.ptr
+!Array = !llvm.ptr
+
+#SparseVector = #sparse_tensor.encoding<{
+  map = (d0) -> (d0 : compressed)
+}>
+
+module {
+  func.func private @rtsrand(index) -> (!Generator)
+  func.func private @rtrand(!Generator, index) -> (index)
+  func.func private @rtdrand(!Generator) -> ()
+  func.func private @shuffle(index, !Generator) -> (!Array)
+  func.func private @shuffleFree(!Array) -> ()
+
+  //
+  // Main driver.
+  //
+  func.func @entry() {
+    %c0 = arith.constant 0 : index
+    %c1 = arith.constant 1 : index
+    %f0 = arith.constant 0.0 : f64
+    %c99 = arith.constant 99 : index
+    %c100 = arith.constant 100 : index
+
+    // Set up input size and sparsity level.
+    %size = arith.constant 50 : index
+    %sparsity = arith.constant 90 : index
+    %zeros = arith.muli %size, %sparsity : index
+    %nz = arith.floordivsi %zeros, %c100 : index
+    %nse = arith.subi %size, %nz : index
+
+    // Set up an empty vector.
+    %empty = tensor.empty(%size) : tensor<?xf64>
+    %zero_vec = linalg.fill ins(%f0 : f64) outs(%empty : tensor<?xf64>) -> tensor<?xf64>
+
+    // Generate shuffled indices in the range of [0, %size).
+    %g = func.call @rtsrand(%c0) : (index) ->(!Generator)
+    %res = func.call @shuffle(%size, %g) : (index, !Generator) -> !Array
+
+    // Iterate through the number of nse indices to insert values.
+    %output = scf.for %iv = %c0 to %nse step %c1 iter_args(%iter = %zero_vec) -> tensor<?xf64> {
+      // Fetch the index to insert value from shuffled index array.
+      %r = arith.index_cast %iv : index to i64
+      %arr = llvm.getelementptr %res[%r] : (!llvm.ptr, i64) -> !llvm.ptr, i64
+      %val = llvm.load %arr : !Array -> i64
+      %idx = arith.index_cast %val : i64 to index
+      // Generate a random number from 1 to 100.
+      %ri0 = func.call @rtrand(%g, %c99) : (!Generator, index) -> (index)
+      %ri1 = arith.addi %ri0, %c1 : index
+      %r0 = arith.index_cast %ri1 : index to i64
+      %fr = arith.uitofp %r0 : i64 to f64
+      // Insert the random number to current index.
+      %out = tensor.insert %fr into %iter[%idx] : tensor<?xf64>
+      scf.yield %out : tensor<?xf64>
+    }
+
+    %sv = sparse_tensor.convert %output : tensor<?xf64> to tensor<?xf64, #SparseVector>
+    %n0 = sparse_tensor.number_of_entries %sv : tensor<?xf64, #SparseVector>
+
+    // Print the number of non-zeros for verification.
+    //
+    // CHECK: 5
+    vector.print %n0 : index
+
+    // Release the resources.
+    bufferization.dealloc_tensor %sv : tensor<?xf64, #SparseVector>
+    func.call @shuffleFree(%res) : (!Array) -> ()
+    func.call @rtdrand(%g) : (!Generator) -> ()
+
+    return
+  }
+}

mlir/lib/ExecutionEngine/CRunnerUtils.cpp

mlir/include/mlir/ExecutionEngine/CRunnerUtils.h

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_generate.mlir

github-actions · 2024-01-07T21:59:57Z

✅ With the latest revision this PR passed the C/C++ code formatter.

mlir/lib/ExecutionEngine/CRunnerUtils.cpp

mlir/include/mlir/ExecutionEngine/CRunnerUtils.h

mlir/lib/ExecutionEngine/CRunnerUtils.cpp

…ils to generate unique and random numbers It's helpful for generating tensor with specified sparsity level.

aartbik

Nice work!

Shuffle can generate an array of unique and random numbers from 0 to size-1. It can be used to generate tensors with specified sparsity level.

yinying-lisa-li marked this pull request as ready for review January 5, 2024 18:53

llvmbot added mlir:sparse Sparse compiler in MLIR mlir mlir:execution-engine labels Jan 5, 2024

yinying-lisa-li requested review from PeimingLiu and aartbik January 5, 2024 18:54

aartbik reviewed Jan 5, 2024

View reviewed changes

yinying-lisa-li force-pushed the shuffle branch from ce45261 to bd08525 Compare January 7, 2024 21:57

yinying-lisa-li requested a review from aartbik January 8, 2024 17:15

PeimingLiu reviewed Jan 8, 2024

View reviewed changes

mlir/lib/ExecutionEngine/CRunnerUtils.cpp Outdated Show resolved Hide resolved

aartbik reviewed Jan 8, 2024

View reviewed changes

mlir/include/mlir/ExecutionEngine/CRunnerUtils.h Outdated Show resolved Hide resolved

mlir/include/mlir/ExecutionEngine/CRunnerUtils.h Outdated Show resolved Hide resolved

mlir/lib/ExecutionEngine/CRunnerUtils.cpp Outdated Show resolved Hide resolved

yinying-lisa-li force-pushed the shuffle branch from ca45add to 3f7b0f9 Compare January 8, 2024 18:33

yinying-lisa-li requested review from PeimingLiu and aartbik January 8, 2024 18:41

PeimingLiu approved these changes Jan 9, 2024

View reviewed changes

aartbik reviewed Jan 9, 2024

View reviewed changes

mlir/lib/ExecutionEngine/CRunnerUtils.cpp Show resolved Hide resolved

yinying-lisa-li added 7 commits January 9, 2024 19:57

[mlir][sparse][CRunnerUtils] Add shuffle and shuffleFree in CRunnerUt…

a1f9b09

…ils to generate unique and random numbers It's helpful for generating tensor with specified sparsity level.

address comments and use memref instead of pointer

dfea486

format

a49feea

fix issue for windows

acf7ca2

header

0c2b978

remove copy

b93ee7e

mref check

a3c71f7

yinying-lisa-li force-pushed the shuffle branch from 3bd628a to a3c71f7 Compare January 9, 2024 20:10

yinying-lisa-li requested a review from aartbik January 9, 2024 20:17

aartbik approved these changes Jan 9, 2024

View reviewed changes

yinying-lisa-li merged commit 412d784 into llvm:main Jan 10, 2024

yinying-lisa-li deleted the shuffle branch January 10, 2024 00:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[mlir][sparse][CRunnerUtils] Add shuffle in CRunnerUtils #77124

[mlir][sparse][CRunnerUtils] Add shuffle in CRunnerUtils #77124

Uh oh!

yinying-lisa-li commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Jan 7, 2024 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aartbik left a comment

Uh oh!

Uh oh!

[mlir][sparse][CRunnerUtils] Add shuffle in CRunnerUtils #77124

[mlir][sparse][CRunnerUtils] Add shuffle in CRunnerUtils #77124

Uh oh!

Conversation

yinying-lisa-li commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

llvmbot commented Jan 5, 2024

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Jan 7, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aartbik left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Jan 7, 2024 •

edited

Loading