Added std.floor operation to match std.ceil

There should be an equivalent std.floor op to std.ceil. This includes
matching lowerings for SPIRV, NVVM, ROCDL, and LLVM.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D85940

diff --git a/mlir/include/mlir/Dialect/StandardOps/IR/Ops.td b/mlir/include/mlir/Dialect/StandardOps/IR/Ops.td
index 088f262..510d485 100644 (file)
--- a/mlir/include/mlir/Dialect/StandardOps/IR/Ops.td
+++ b/mlir/include/mlir/Dialect/StandardOps/IR/Ops.td
@@ -815,6 +815,39 @@ def CeilFOp : FloatUnaryOp<"ceilf"> {
 }
 
 //===----------------------------------------------------------------------===//
+// FloorFOp
+//===----------------------------------------------------------------------===//
+
+def FloorFOp : FloatUnaryOp<"floorf"> {
+  let summary = "floor of the specified value";
+  let description = [{
+    Syntax:
+
+    ```
+    operation ::= ssa-id `=` `std.floorf` ssa-use `:` type
+    ```
+
+    The `floorf` operation computes the floor of a given value. It takes one
+    operand and returns one result of the same type. This type may be a float
+    scalar type, a vector whose element type is float, or a tensor of floats.
+    It has no standard attributes.
+
+    Example:
+
+    ```mlir
+    // Scalar floor value.
+    %a = floorf %b : f64
+
+    // SIMD vector element-wise floor value.
+    %f = floorf %g : vector<4xf32>
+
+    // Tensor element-wise floor value.
+    %x = floorf %y : tensor<4x?xf8>
+    ```
+  }];
+}
+
+//===----------------------------------------------------------------------===//
 // CmpFOp
 //===----------------------------------------------------------------------===//
 

diff --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
index 76c1668..d11cc51 100644 (file)
--- a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
+++ b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -137,7 +137,8 @@ struct LowerGpuOpsToNVVMOpsPass
     LLVMConversionTarget target(getContext());
     target.addIllegalDialect<gpu::GPUDialect>();
     target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
-                        LLVM::LogOp, LLVM::Log10Op, LLVM::Log2Op>();
+                        LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op,
+                        LLVM::Log2Op>();
     target.addIllegalOp<FuncOp>();
     target.addLegalDialect<NVVM::NVVMDialect>();
     // TODO: Remove once we support replacing non-root ops.
@@ -174,6 +175,8 @@ void mlir::populateGpuToNVVMConversionPatterns(
                                                "__nv_cos");
   patterns.insert<OpToFuncCallLowering<ExpOp>>(converter, "__nv_expf",
                                                "__nv_exp");
+  patterns.insert<OpToFuncCallLowering<FloorFOp>>(converter, "__nv_floorf",
+                                                  "__nv_floor");
   patterns.insert<OpToFuncCallLowering<LogOp>>(converter, "__nv_logf",
                                                "__nv_log");
   patterns.insert<OpToFuncCallLowering<Log10Op>>(converter, "__nv_log10f",

diff --git a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
index 697f807..40cf097 100644 (file)
--- a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
+++ b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
@@ -71,7 +71,8 @@ struct LowerGpuOpsToROCDLOpsPass
     LLVMConversionTarget target(getContext());
     target.addIllegalDialect<gpu::GPUDialect>();
     target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
-                        LLVM::LogOp, LLVM::Log10Op, LLVM::Log2Op>();
+                        LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op,
+                        LLVM::Log2Op>();
     target.addIllegalOp<FuncOp>();
     target.addLegalDialect<ROCDL::ROCDLDialect>();
     // TODO: Remove once we support replacing non-root ops.
@@ -104,6 +105,8 @@ void mlir::populateGpuToROCDLConversionPatterns(
                                                "__ocml_cos_f64");
   patterns.insert<OpToFuncCallLowering<ExpOp>>(converter, "__ocml_exp_f32",
                                                "__ocml_exp_f64");
+  patterns.insert<OpToFuncCallLowering<FloorFOp>>(converter, "__ocml_floor_f32",
+                                                  "__ocml_floor_f64");
   patterns.insert<OpToFuncCallLowering<LogOp>>(converter, "__ocml_log_f32",
                                                "__ocml_log_f64");
   patterns.insert<OpToFuncCallLowering<Log10Op>>(converter, "__ocml_log10_f32",

diff --git a/mlir/lib/Conversion/StandardToLLVM/StandardToLLVM.cpp b/mlir/lib/Conversion/StandardToLLVM/StandardToLLVM.cpp
index 4a06196..0ee1166 100644 (file)
--- a/mlir/lib/Conversion/StandardToLLVM/StandardToLLVM.cpp
+++ b/mlir/lib/Conversion/StandardToLLVM/StandardToLLVM.cpp
@@ -1418,6 +1418,7 @@ using CosOpLowering = VectorConvertToLLVMPattern<CosOp, LLVM::CosOp>;
 using DivFOpLowering = VectorConvertToLLVMPattern<DivFOp, LLVM::FDivOp>;
 using ExpOpLowering = VectorConvertToLLVMPattern<ExpOp, LLVM::ExpOp>;
 using Exp2OpLowering = VectorConvertToLLVMPattern<Exp2Op, LLVM::Exp2Op>;
+using FloorFOpLowering = VectorConvertToLLVMPattern<FloorFOp, LLVM::FFloorOp>;
 using Log10OpLowering = VectorConvertToLLVMPattern<Log10Op, LLVM::Log10Op>;
 using Log2OpLowering = VectorConvertToLLVMPattern<Log2Op, LLVM::Log2Op>;
 using LogOpLowering = VectorConvertToLLVMPattern<LogOp, LLVM::LogOp>;
@@ -3285,6 +3286,7 @@ void mlir::populateStdToLLVMNonMemoryConversionPatterns(
       DivFOpLowering,
       ExpOpLowering,
       Exp2OpLowering,
+      FloorFOpLowering,
       GenericAtomicRMWOpLowering,
       LogOpLowering,
       Log10OpLowering,

diff --git a/mlir/lib/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.cpp b/mlir/lib/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.cpp
index 268139f..6ae17c3 100644 (file)
--- a/mlir/lib/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.cpp
+++ b/mlir/lib/Conversion/StandardToSPIRV/ConvertStandardToSPIRV.cpp
@@ -1076,6 +1076,7 @@ void populateStandardToSPIRVPatterns(MLIRContext *context,
       UnaryAndBinaryOpPattern<CosOp, spirv::GLSLCosOp>,
       UnaryAndBinaryOpPattern<DivFOp, spirv::FDivOp>,
       UnaryAndBinaryOpPattern<ExpOp, spirv::GLSLExpOp>,
+      UnaryAndBinaryOpPattern<FloorFOp, spirv::GLSLFloorOp>,
       UnaryAndBinaryOpPattern<LogOp, spirv::GLSLLogOp>,
       UnaryAndBinaryOpPattern<MulFOp, spirv::FMulOp>,
       UnaryAndBinaryOpPattern<MulIOp, spirv::IMulOp>,

diff --git a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
index df38df1..6b071a0 100644 (file)
--- a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
+++ b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
@@ -173,6 +173,21 @@ gpu.module @test_module {
 // -----
 
 gpu.module @test_module {
+  // CHECK: llvm.func @__nv_floorf(!llvm.float) -> !llvm.float
+  // CHECK: llvm.func @__nv_floor(!llvm.double) -> !llvm.double
+  // CHECK-LABEL: func @gpu_floor
+  func @gpu_floor(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = std.floorf %arg_f32 : f32
+    // CHECK: llvm.call @__nv_floorf(%{{.*}}) : (!llvm.float) -> !llvm.float
+    %result64 = std.floorf %arg_f64 : f64
+    // CHECK: llvm.call @__nv_floor(%{{.*}}) : (!llvm.double) -> !llvm.double
+    std.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
   // CHECK: llvm.func @__nv_cosf(!llvm.float) -> !llvm.float
   // CHECK: llvm.func @__nv_cos(!llvm.double) -> !llvm.double
   // CHECK-LABEL: func @gpu_cos

diff --git a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
index a7565bb..b17d75f 100644 (file)
--- a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
+++ b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
@@ -115,6 +115,21 @@ gpu.module @test_module {
 // -----
 
 gpu.module @test_module {
+  // CHECK: llvm.func @__ocml_floor_f32(!llvm.float) -> !llvm.float
+  // CHECK: llvm.func @__ocml_floor_f64(!llvm.double) -> !llvm.double
+  // CHECK-LABEL: func @gpu_floor
+  func @gpu_floor(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = std.floorf %arg_f32 : f32
+    // CHECK: llvm.call @__ocml_floor_f32(%{{.*}}) : (!llvm.float) -> !llvm.float
+    %result64 = std.floorf %arg_f64 : f64
+    // CHECK: llvm.call @__ocml_floor_f64(%{{.*}}) : (!llvm.double) -> !llvm.double
+    std.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
   // CHECK: llvm.func @__ocml_cos_f32(!llvm.float) -> !llvm.float
   // CHECK: llvm.func @__ocml_cos_f64(!llvm.double) -> !llvm.double
   // CHECK-LABEL: func @gpu_cos

diff --git a/mlir/test/Conversion/StandardToLLVM/standard-to-llvm.mlir b/mlir/test/Conversion/StandardToLLVM/standard-to-llvm.mlir
index c55950a..c736308 100644 (file)
--- a/mlir/test/Conversion/StandardToLLVM/standard-to-llvm.mlir
+++ b/mlir/test/Conversion/StandardToLLVM/standard-to-llvm.mlir
@@ -40,6 +40,27 @@ func @sine(%arg0 : f32) {
 
 // -----
 
+// CHECK-LABEL: func @ceilf(
+// CHECK-SAME: !llvm.float
+func @ceilf(%arg0 : f32) {
+  // CHECK: "llvm.intr.ceil"(%arg0) : (!llvm.float) -> !llvm.float
+  %0 = ceilf %arg0 : f32
+  std.return
+}
+
+// -----
+
+// CHECK-LABEL: func @floorf(
+// CHECK-SAME: !llvm.float
+func @floorf(%arg0 : f32) {
+  // CHECK: "llvm.intr.floor"(%arg0) : (!llvm.float) -> !llvm.float
+  %0 = floorf %arg0 : f32
+  std.return
+}
+
+// -----
+
+
 // CHECK-LABEL: func @rsqrt_double(
 // CHECK-SAME: !llvm.double
 func @rsqrt_double(%arg0 : f64) {

diff --git a/mlir/test/Conversion/StandardToSPIRV/std-ops-to-spirv.mlir b/mlir/test/Conversion/StandardToSPIRV/std-ops-to-spirv.mlir
index e85f78f..1b83af1 100644 (file)
--- a/mlir/test/Conversion/StandardToSPIRV/std-ops-to-spirv.mlir
+++ b/mlir/test/Conversion/StandardToSPIRV/std-ops-to-spirv.mlir
@@ -65,6 +65,8 @@ func @float32_unary_scalar(%arg0: f32) {
   %8 = tanh %arg0 : f32
   // CHECK: spv.GLSL.Sin %{{.*}}: f32
   %9 = sin %arg0 : f32
+  // CHECK: spv.GLSL.Floor %{{.*}}: f32
+  %10 = floorf %arg0 : f32
   return
 }
 

diff --git a/mlir/test/IR/core-ops.mlir b/mlir/test/IR/core-ops.mlir
index 7447071..69e974b 100644 (file)
--- a/mlir/test/IR/core-ops.mlir
+++ b/mlir/test/IR/core-ops.mlir
@@ -554,6 +554,18 @@ func @standard_instrs(tensor<4x4x?xf32>, f32, i32, index, i64, f16) {
   // CHECK: = fptosi {{.*}} : f16 to i64
   %162 = fptosi %half : f16 to i64
 
+  // CHECK: floorf %arg1 : f32
+  %163 = "std.floorf"(%f) : (f32) -> f32
+
+  // CHECK: %{{[0-9]+}} = floorf %arg1 : f32
+  %164 = floorf %f : f32
+
+  // CHECK: %{{[0-9]+}} = floorf %cst_8 : vector<4xf32>
+  %165 = floorf %vcf32 : vector<4xf32>
+
+  // CHECK: %{{[0-9]+}} = floorf %arg0 : tensor<4x4x?xf32>
+  %166 = floorf %t : tensor<4x4x?xf32>
+
   return
 }