================
@@ -9860,6 +9912,342 @@ TEST(APFloatTest, ConvertDoubleToE8M0FNU) {
EXPECT_EQ(status, APFloat::opUnderflow | APFloat::opInexact);
}
+TEST(APFloatTest, Float8E5M3FNUValues) {
+ // High end of the range
+ auto test = APFloat(APFloat::Float8E5M3FNU(), "0x1.c0p16");
+ EXPECT_EQ(0x1.c0p16, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.c0p15");
+ EXPECT_EQ(0x1.cp15, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p14");
+ EXPECT_EQ(0x1.0p14, test.convertToDouble());
+
+ // tests the fix in makeLargest()
+ test = APFloat::getLargest(APFloat::Float8E5M3FNU());
+ EXPECT_EQ(0x1.cp16, test.convertToDouble());
+
+ // tests overflow to nan
+ APFloat nan = APFloat(APFloat::Float8E5M3FNU(), "nan");
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.e0p+16");
+ EXPECT_TRUE(test.bitwiseIsEqual(nan));
+
+ // Mid of the range
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p0");
+ EXPECT_EQ(1.0, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p1");
+ EXPECT_EQ(2.0, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p2");
+ EXPECT_EQ(4.0, test.convertToDouble());
+
+ // Low end of the range
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p-12");
+ EXPECT_EQ(0x1.0p-12, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p-13");
+ EXPECT_EQ(0x1.0p-13, test.convertToDouble());
+
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p-14");
+ EXPECT_EQ(0x1.0p-14, test.convertToDouble());
+ EXPECT_TRUE(test.isSmallestNormalized());
+
+ // Smallest value
+ test = APFloat::getSmallest(APFloat::Float8E5M3FNU());
+ EXPECT_EQ(0x1.0p-17, test.convertToDouble());
+
+ // Value below the smallest, but clamped to the smallest
+ test = APFloat(APFloat::Float8E5M3FNU(), "0x1.0p-18");
+ EXPECT_EQ(0, test.convertToDouble());
+}
+
+TEST(APFloatTest, Float8E5M3FNUFromString) {
+ // Exactly representable
+ EXPECT_EQ(64, APFloat(APFloat::Float8E5M3FNU(), "64").convertToDouble());
+ // Overflow to NaN
+ EXPECT_TRUE(APFloat(APFloat::Float8E5M3FNU(), "0x1.0p17").isNaN());
+ // Inf converted to NaN
+ EXPECT_TRUE(APFloat(APFloat::Float8E5M3FNU(), "inf").isNaN());
+ // NaN converted to NaN
+ EXPECT_TRUE(APFloat(APFloat::Float8E5M3FNU(), "nan").isNaN());
+}
+
+TEST(APFloatTest, Float8E5M3FNUDivideByZero) {
+ APFloat x(APFloat::Float8E5M3FNU(), "1");
+ APFloat zero(APFloat::Float8E5M3FNU(), "0");
+ EXPECT_EQ(x.divide(zero, APFloat::rmNearestTiesToEven),
APFloat::opDivByZero);
+ EXPECT_TRUE(x.isNaN());
+}
+
+TEST(APFloatTest, Float8E5M3FNUGetSignedValues) {
+#ifdef GTEST_HAS_DEATH_TEST
+#ifndef NDEBUG
+ EXPECT_DEATH(APFloat(APFloat::Float8E5M3FNU(), "-64"),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat(APFloat::Float8E5M3FNU(), "-0x1.0p17"),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat(APFloat::Float8E5M3FNU(), "-inf"),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat::getNaN(APFloat::Float8E5M3FNU(), true),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat::getInf(APFloat::Float8E5M3FNU(), true),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat::getSmallest(APFloat::Float8E5M3FNU(), true),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat::getSmallestNormalized(APFloat::Float8E5M3FNU(), true),
+ "This floating point format does not support signed values");
+ EXPECT_DEATH(APFloat::getLargest(APFloat::Float8E5M3FNU(), true),
+ "This floating point format does not support signed values");
+ APFloat x = APFloat(APFloat::Float8E5M3FNU(), "4");
+ APFloat y = APFloat(APFloat::Float8E5M3FNU(), "8");
+ EXPECT_DEATH(x.subtract(y, APFloat::rmNearestTiesToEven),
+ "This floating point format does not support signed values");
+#endif // NDEBUG
+#endif // GTEST_HAS_DEATH_TEST
+}
+
+TEST(APFloatTest, Float8E5M3FNUGetInf) {
+ // The Float8E5M3FNU format does not support infinity and the all ones
+ // representation is treated as NaN.
+ APFloat t = APFloat::getInf(APFloat::Float8E5M3FNU());
+ EXPECT_TRUE(t.isNaN());
+ EXPECT_FALSE(t.isInfinity());
+}
+
+TEST(APFloatTest, Float8E5M3FNUSmallest) {
+ APFloat test(APFloat::getSmallest(APFloat::Float8E5M3FNU()));
+ EXPECT_EQ(0x1.0p-17, test.convertToDouble());
+
+ EXPECT_TRUE(test.isSmallest());
+ EXPECT_EQ(fcPosSubnormal, test.classify());
+
+ test = APFloat::getAllOnesValue(APFloat::Float8E5M3FNU());
+ EXPECT_TRUE(test.isNaN());
+}
+
+TEST(APFloatTest, Float8E5M3FNUExhaustivePair) {
+ // Test each pair of 8-bit values for Float8E5M3FNU format.
+ // This format is unsigned, so subtraction is only tested when the result
+ // is non-negative (which corresponds to i >= j since the bit-pattern
+ // ordering matches the value ordering). IEEE remainder can produce
+ // negative results, so it is only tested when the reference result is
+ // non-negative.
+ APFloat::Semantics Sem = APFloat::S_Float8E5M3FNU;
+ const llvm::fltSemantics &S = APFloat::EnumToSemantics(Sem);
+ for (int i = 0; i < 256; i++) {
+ for (int j = 0; j < 256; j++) {
----------------
schwarzschild-radius wrote:
Fixed it. Updated it to pre-increment
https://github.com/llvm/llvm-project/pull/210720
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
