diff --git a/mlir/integration_test/Dialect/LLVMIR/CPU/test-vector-reductions-fp.mlir b/mlir/integration_test/Dialect/LLVMIR/CPU/test-vector-reductions-fp.mlir --- a/mlir/integration_test/Dialect/LLVMIR/CPU/test-vector-reductions-fp.mlir +++ b/mlir/integration_test/Dialect/LLVMIR/CPU/test-vector-reductions-fp.mlir @@ -42,24 +42,48 @@ llvm.call @print_newline() : () -> () // CHECK: 11 + %add1r = "llvm.intr.experimental.vector.reduce.v2.fadd"(%0, %v) + {reassoc = true} : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float + llvm.call @print_f32(%add1r) : (!llvm.float) -> () + llvm.call @print_newline() : () -> () + // CHECK: 11 + %add2 = "llvm.intr.experimental.vector.reduce.v2.fadd"(%1, %v) : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float llvm.call @print_f32(%add2) : (!llvm.float) -> () llvm.call @print_newline() : () -> () // CHECK: 12 + %add2r = "llvm.intr.experimental.vector.reduce.v2.fadd"(%1, %v) + {reassoc = true} : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float + llvm.call @print_f32(%add2r) : (!llvm.float) -> () + llvm.call @print_newline() : () -> () + // CHECK: 12 + %mul1 = "llvm.intr.experimental.vector.reduce.v2.fmul"(%0, %v) : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float llvm.call @print_f32(%mul1) : (!llvm.float) -> () llvm.call @print_newline() : () -> () // CHECK: 24 + %mul1r = "llvm.intr.experimental.vector.reduce.v2.fmul"(%0, %v) + {reassoc = true} : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float + llvm.call @print_f32(%mul1r) : (!llvm.float) -> () + llvm.call @print_newline() : () -> () + // CHECK: 24 + %mul2 = "llvm.intr.experimental.vector.reduce.v2.fmul"(%1, %v) : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float llvm.call @print_f32(%mul2) : (!llvm.float) -> () llvm.call @print_newline() : () -> () // CHECK: 48 + %mul2r = "llvm.intr.experimental.vector.reduce.v2.fmul"(%1, %v) + {reassoc = true} : (!llvm.float, !llvm<"<4 x float>">) -> !llvm.float + llvm.call @print_f32(%mul2r) : (!llvm.float) -> () + llvm.call @print_newline() : () -> () + // CHECK: 48 + llvm.return } } diff --git a/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f32-reassoc.mlir b/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f32-reassoc.mlir new file mode 100644 --- /dev/null +++ b/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f32-reassoc.mlir @@ -0,0 +1,38 @@ +// RUN: mlir-opt %s -convert-scf-to-std \ +// RUN: -convert-vector-to-llvm='reassociate-fp-reductions' \ +// RUN: -convert-std-to-llvm | \ +// RUN: mlir-cpu-runner -e entry -entry-point-result=void \ +// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ +// RUN: FileCheck %s + +func @entry() { + // Construct test vector, numerically very stable. + %f1 = constant 1.0: f32 + %f2 = constant 2.0: f32 + %f3 = constant 3.0: f32 + %v0 = vector.broadcast %f1 : f32 to vector<64xf32> + %v1 = vector.insert %f2, %v0[11] : f32 into vector<64xf32> + %v2 = vector.insert %f3, %v1[52] : f32 into vector<64xf32> + vector.print %v2 : vector<64xf32> + // + // test vector: + // + // CHECK: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ) + + // Various vector reductions. Not full functional unit tests, but + // a simple integration test to see if the code runs end-to-end. + %0 = vector.reduction "add", %v2 : vector<64xf32> into f32 + vector.print %0 : f32 + // CHECK: 67 + %1 = vector.reduction "mul", %v2 : vector<64xf32> into f32 + vector.print %1 : f32 + // CHECK: 6 + %2 = vector.reduction "min", %v2 : vector<64xf32> into f32 + vector.print %2 : f32 + // CHECK: 1 + %3 = vector.reduction "max", %v2 : vector<64xf32> into f32 + vector.print %3 : f32 + // CHECK: 3 + + return +} diff --git a/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f64-reassoc.mlir b/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f64-reassoc.mlir new file mode 100644 --- /dev/null +++ b/mlir/integration_test/Dialect/Vector/CPU/test-reductions-f64-reassoc.mlir @@ -0,0 +1,38 @@ +// RUN: mlir-opt %s -convert-scf-to-std \ +// RUN: -convert-vector-to-llvm='reassociate-fp-reductions' \ +// RUN: -convert-std-to-llvm | \ +// RUN: mlir-cpu-runner -e entry -entry-point-result=void \ +// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ +// RUN: FileCheck %s + +func @entry() { + // Construct test vector, numerically very stable. + %f1 = constant 1.0: f64 + %f2 = constant 2.0: f64 + %f3 = constant 3.0: f64 + %v0 = vector.broadcast %f1 : f64 to vector<64xf64> + %v1 = vector.insert %f2, %v0[11] : f64 into vector<64xf64> + %v2 = vector.insert %f3, %v1[52] : f64 into vector<64xf64> + vector.print %v2 : vector<64xf64> + // + // test vector: + // + // CHECK: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ) + + // Various vector reductions. Not full functional unit tests, but + // a simple integration test to see if the code runs end-to-end. + %0 = vector.reduction "add", %v2 : vector<64xf64> into f64 + vector.print %0 : f64 + // CHECK: 67 + %1 = vector.reduction "mul", %v2 : vector<64xf64> into f64 + vector.print %1 : f64 + // CHECK: 6 + %2 = vector.reduction "min", %v2 : vector<64xf64> into f64 + vector.print %2 : f64 + // CHECK: 1 + %3 = vector.reduction "max", %v2 : vector<64xf64> into f64 + vector.print %3 : f64 + // CHECK: 3 + + return +}