[SYCL] Remove direct initialization constructor from bfloat16 class (#…

…4989) This constructor is being removed to resolve ambiguity when user trying initialize bfloat16 with int, like `auto a = bfloat16(0);` Add conversion operator for sycl::half Conversion operator for storage_t is replaced by 'raw()' method Signed-off-by: Alexey Sotkin <[email protected]>
intel · Dec 17, 2021 · 81154ec · 81154ec
1 parent 756c2e8
commit 81154ec
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 4 deletions.
diff --git a/sycl/include/sycl/ext/intel/experimental/bfloat16.hpp b/sycl/include/sycl/ext/intel/experimental/bfloat16.hpp
@@ -9,6 +9,7 @@
 #pragma once
 
 #include <CL/__spirv/spirv_ops.hpp>
+#include <CL/sycl/half_type.hpp>
 
 __SYCL_INLINE_NAMESPACE(cl) {
 namespace sycl {
@@ -43,8 +44,11 @@ class [[sycl_detail::uses_aspects(ext_intel_bf16_conversion)]] bfloat16 {
 #endif
   }
 
-  // Direct initialization
-  bfloat16(const storage_t &a) : value(a) {}
+  static bfloat16 from_bits(const storage_t &a) {
+    bfloat16 res;
+    res.value = a;
+    return res;
+  }
 
   // Implicit conversion from float to bfloat16
   bfloat16(const float &a) { value = from_float(a); }
@@ -56,9 +60,10 @@ class [[sycl_detail::uses_aspects(ext_intel_bf16_conversion)]] bfloat16 {
 
   // Implicit conversion from bfloat16 to float
   operator float() const { return to_float(value); }
+  operator sycl::half() const { return to_float(value); }
 
   // Get raw bits representation of bfloat16
-  operator storage_t() const { return value; }
+  storage_t raw() const { return value; }
 
   // Logical operators (!,||,&&) are covered if we can cast to bool
   explicit operator bool() { return to_float(value) != 0.0f; }

diff --git a/sycl/test/extensions/bfloat16.cpp b/sycl/test/extensions/bfloat16.cpp
@@ -8,6 +8,7 @@
 using sycl::ext::intel::experimental::bfloat16;
 
 SYCL_EXTERNAL uint16_t some_bf16_intrinsic(uint16_t x, uint16_t y);
+SYCL_EXTERNAL void foo(long x, sycl::half y);
 
 __attribute__((noinline)) float op(float a, float b) {
   // CHECK: define {{.*}} spir_func float @_Z2opff(float [[a:%.*]], float [[b:%.*]])
@@ -27,11 +28,22 @@ __attribute__((noinline)) float op(float a, float b) {
   // CHECK-NOT: uitofp
   // CHECK-NOT: fptoui
 
-  bfloat16 D = some_bf16_intrinsic(A, C);
+  bfloat16 D = bfloat16::from_bits(some_bf16_intrinsic(A.raw(), C.raw()));
   // CHECK: [[D:%.*]] = tail call spir_func zeroext i16 @_Z19some_bf16_intrinsictt(i16 zeroext [[A]], i16 zeroext [[C]])
   // CHECK-NOT: uitofp
   // CHECK-NOT: fptoui
 
+  long L = bfloat16(3.14f);
+  // CHECK: [[L_bfloat16:%.*]] = tail call spir_func zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float 0x40091EB860000000)
+  // CHECK: [[L_float:%.*]] = tail call spir_func float @_Z27__spirv_ConvertBF16ToFINTELt(i16 zeroext [[L_bfloat16]])
+  // CHECK: [[L:%.*]] = fptosi float [[L_float]] to i{{32|64}}
+
+  sycl::half H = bfloat16(2.71f);
+  // CHECK: [[H_bfloat16:%.*]] = tail call spir_func zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float 0x4005AE1480000000)
+  // CHECK: [[H_float:%.*]] = tail call spir_func float @_Z27__spirv_ConvertBF16ToFINTELt(i16 zeroext [[H_bfloat16]])
+  // CHECK: [[H:%.*]] = fptrunc float [[H_float]] to half
+  foo(L, H);
+
   return D;
   // CHECK: [[RetVal:%.*]] = tail call spir_func float @_Z27__spirv_ConvertBF16ToFINTELt(i16 zeroext [[D]])
   // CHECK: ret float [[RetVal]]