[Mlir-commits] [mlir] b312776 - [mlir][sparse] Preserve NaNs when converting float to bfloat

[Benjamin Kramer]

Author: Benjamin Kramer
Date: 2022-06-21T15:22:35+02:00
New Revision: b3127769b1ae39928a76b0a0469152c0f826e424

URL: https://github.com/llvm/llvm-project/commit/b3127769b1ae39928a76b0a0469152c0f826e424
DIFF: https://github.com/llvm/llvm-project/commit/b3127769b1ae39928a76b0a0469152c0f826e424.diff

LOG: [mlir][sparse] Preserve NaNs when converting float to bfloat

Added: 
    

Modified: 
    mlir/lib/ExecutionEngine/Float16bits.cpp

Removed: 
    


################################################################################
diff  --git a/mlir/lib/ExecutionEngine/Float16bits.cpp b/mlir/lib/ExecutionEngine/Float16bits.cpp
index 289eaeab2ccc..96eed35fef26 100644
--- a/mlir/lib/ExecutionEngine/Float16bits.cpp
+++ b/mlir/lib/ExecutionEngine/Float16bits.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "mlir/ExecutionEngine/Float16bits.h"
+#include <cmath>
 
 namespace {
 
@@ -106,6 +107,9 @@ const uint32_t kF32BfMantiBitDiff = 16;
 // Constructs the 16 bit representation for a bfloat value from a float value.
 // This implementation is adapted from Eigen.
 uint16_t float2bfloat(float floatValue) {
+  if (std::isnan(floatValue))
+    return std::signbit(floatValue) ? 0xFFC0 : 0x7FC0;
+
   Float32Bits floatBits;
   floatBits.f = floatValue;
   uint16_t bfloatBits;