[Mlir-commits] [mlir] 74c3615 - Add LLVMIR Dialect counterparts of @llvm.maximum and @llvm.minimum.

[Mehdi Amini]

Author: Andrew Pritchard
Date: 2021-02-10T00:57:43Z
New Revision: 74c3615997472b44ef77fedd7914d376f5e45ad7

URL: https://github.com/llvm/llvm-project/commit/74c3615997472b44ef77fedd7914d376f5e45ad7
DIFF: https://github.com/llvm/llvm-project/commit/74c3615997472b44ef77fedd7914d376f5e45ad7.diff

LOG: Add LLVMIR Dialect counterparts of @llvm.maximum and @llvm.minimum.

These are similar to maxnum and minnum, but they're defined to treat -0
as less than +0.  This behavior can't be expressed using float
comparisons and selects, since comparisons are defined to treat
different-signed zeros as equal.  So, the only way to communicate this
behavior into LLVM IR without defining target-specific intrinsics is to
add the corresponding ops.

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D96373

Added: 
    

Modified: 
    mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
    mlir/test/Target/llvmir-intrinsics.mlir

Removed: 
    


################################################################################
diff  --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index 368373519ff4..7a2152b9a481 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -1187,6 +1187,8 @@ def LLVM_BitReverseOp : LLVM_UnaryIntrinsicOp<"bitreverse">;
 def LLVM_CtPopOp : LLVM_UnaryIntrinsicOp<"ctpop">;
 def LLVM_MaxNumOp : LLVM_BinarySameArgsIntrinsicOp<"maxnum">;
 def LLVM_MinNumOp : LLVM_BinarySameArgsIntrinsicOp<"minnum">;
+def LLVM_MaximumOp : LLVM_BinarySameArgsIntrinsicOp<"maximum">;
+def LLVM_MinimumOp : LLVM_BinarySameArgsIntrinsicOp<"minimum">;
 def LLVM_SMaxOp : LLVM_BinarySameArgsIntrinsicOp<"smax">;
 def LLVM_SMinOp : LLVM_BinarySameArgsIntrinsicOp<"smin">;
 

diff  --git a/mlir/test/Target/llvmir-intrinsics.mlir b/mlir/test/Target/llvmir-intrinsics.mlir
index a9fe6f692c68..2a73828ec58c 100644
--- a/mlir/test/Target/llvmir-intrinsics.mlir
+++ b/mlir/test/Target/llvmir-intrinsics.mlir
@@ -144,6 +144,24 @@ llvm.func @ctpop_test(%arg0: i32, %arg1: vector<8xi32>) {
   llvm.return
 }
 
+// CHECK-LABEL: @maximum_test
+llvm.func @maximum_test(%arg0: f32, %arg1: f32, %arg2: vector<8xf32>, %arg3: vector<8xf32>) {
+  // CHECK: call float @llvm.maximum.f32
+  "llvm.intr.maximum"(%arg0, %arg1) : (f32, f32) -> f32
+  // CHECK: call <8 x float> @llvm.maximum.v8f32
+  "llvm.intr.maximum"(%arg2, %arg3) : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
+  llvm.return
+}
+
+// CHECK-LABEL: @minimum_test
+llvm.func @minimum_test(%arg0: f32, %arg1: f32, %arg2: vector<8xf32>, %arg3: vector<8xf32>) {
+  // CHECK: call float @llvm.minimum.f32
+  "llvm.intr.minimum"(%arg0, %arg1) : (f32, f32) -> f32
+  // CHECK: call <8 x float> @llvm.minimum.v8f32
+  "llvm.intr.minimum"(%arg2, %arg3) : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
+  llvm.return
+}
+
 // CHECK-LABEL: @maxnum_test
 llvm.func @maxnum_test(%arg0: f32, %arg1: f32, %arg2: vector<8xf32>, %arg3: vector<8xf32>) {
   // CHECK: call float @llvm.maxnum.f32