[libc][NFC] Split str_to_float tests and reduce type clutter (#75471)

This is a follow up on #75353.

Author: gchatelet

libc/test/src/__support/CMakeLists.txt

@@ -50,6 +50,8 @@ add_libc_test(
     libc-support-tests
   SRCS
     str_to_float_test.cpp
+    str_to_double_test.cpp
+    str_to_long_double_test.cpp
   DEPENDS
     libc.src.__support.str_to_float
     libc.src.__support.uint128

libc/test/src/__support/str_to_double_test.cpp

@@ -0,0 +1,105 @@
+#include "str_to_fp_test.h"
+
+namespace LIBC_NAMESPACE {
+using LlvmLibcStrToDblTest = LlvmLibcStrToFloatTest<double>;
+
+TEST_F(LlvmLibcStrToDblTest, ClingerFastPathFloat64Simple) {
+  clinger_fast_path_test(123, 0, 0xEC00000000000, 1029);
+  clinger_fast_path_test(1234567890123456, 1, 0x5ee2a2eb5a5c0, 1076);
+  clinger_fast_path_test(1234567890, -10, 0xf9add3739635f, 1019);
+}
+
+TEST_F(LlvmLibcStrToDblTest, ClingerFastPathFloat64ExtendedExp) {
+  clinger_fast_path_test(1, 30, 0x93e5939a08cea, 1122);
+  clinger_fast_path_test(1, 37, 0xe17b84357691b, 1145);
+  clinger_fast_path_fails_test(10, 37);
+  clinger_fast_path_fails_test(1, 100);
+}
+
+TEST_F(LlvmLibcStrToDblTest, ClingerFastPathFloat64NegativeExp) {
+  clinger_fast_path_test(1, -10, 0xb7cdfd9d7bdbb, 989);
+  clinger_fast_path_test(1, -20, 0x79ca10c924223, 956);
+  clinger_fast_path_fails_test(1, -25);
+}
+
+TEST_F(LlvmLibcStrToDblTest, EiselLemireFloat64Simple) {
+  eisel_lemire_test(12345678901234567890u, 1, 0x1AC53A7E04BCDA, 1089);
+  eisel_lemire_test(123, 0, 0x1EC00000000000, 1029);
+  eisel_lemire_test(12345678901234568192u, 0, 0x156A95319D63E2, 1086);
+}
+
+TEST_F(LlvmLibcStrToDblTest, EiselLemireFloat64SpecificFailures) {
+  // These test cases have caused failures in the past.
+  eisel_lemire_test(358416272, -33, 0x1BBB2A68C9D0B9, 941);
+  eisel_lemire_test(2166568064000000238u, -9, 0x10246690000000, 1054);
+  eisel_lemire_test(2794967654709307187u, 1, 0x183e132bc608c8, 1087);
+  eisel_lemire_test(2794967654709307188u, 1, 0x183e132bc608c9, 1087);
+}
+
+// Check the fallback states for the algorithm:
+TEST_F(LlvmLibcStrToDblTest, EiselLemireFallbackStates) {
+  // This number can't be evaluated by Eisel-Lemire since it's exactly 1024 away
+  // from both of its closest floating point approximations
+  // (12345678901234548736 and 12345678901234550784)
+  ASSERT_FALSE(
+      internal::eisel_lemire<double>({12345678901234549760u, 0}).has_value());
+}
+
+TEST_F(LlvmLibcStrToDblTest, SimpleDecimalConversion64BasicWholeNumbers) {
+  simple_decimal_conversion_test("123456789012345678900", 0x1AC53A7E04BCDA,
+                                 1089);
+  simple_decimal_conversion_test("123", 0x1EC00000000000, 1029);
+  simple_decimal_conversion_test("12345678901234549760", 0x156A95319D63D8,
+                                 1086);
+}
+
+TEST_F(LlvmLibcStrToDblTest, SimpleDecimalConversion64BasicDecimals) {
+  simple_decimal_conversion_test("1.2345", 0x13c083126e978d, 1023);
+  simple_decimal_conversion_test(".2345", 0x1e04189374bc6a, 1020);
+  simple_decimal_conversion_test(".299792458", 0x132fccb4aca314, 1021);
+}
+
+TEST_F(LlvmLibcStrToDblTest, SimpleDecimalConversion64BasicExponents) {
+  simple_decimal_conversion_test("1e10", 0x12a05f20000000, 1056);
+  simple_decimal_conversion_test("1e-10", 0x1b7cdfd9d7bdbb, 989);
+  simple_decimal_conversion_test("1e300", 0x17e43c8800759c, 2019);
+  simple_decimal_conversion_test("1e-300", 0x156e1fc2f8f359, 26);
+}
+
+TEST_F(LlvmLibcStrToDblTest, SimpleDecimalConversion64BasicSubnormals) {
+  simple_decimal_conversion_test("1e-320", 0x7e8, 0, ERANGE);
+  simple_decimal_conversion_test("1e-308", 0x730d67819e8d2, 0, ERANGE);
+  simple_decimal_conversion_test("2.9e-308", 0x14da6df5e4bcc8, 1);
+}
+
+TEST_F(LlvmLibcStrToDblTest, SimpleDecimalConversion64SubnormalRounding) {
+
+  // Technically you can keep adding digits until you hit the truncation limit,
+  // but this is the shortest string that results in the maximum subnormal that
+  // I found.
+  simple_decimal_conversion_test("2.225073858507201e-308", 0xfffffffffffff, 0,
+                                 ERANGE);
+
+  // Same here, if you were to extend the max subnormal out for another 800
+  // digits, incrementing any one of those digits would create a normal number.
+  simple_decimal_conversion_test("2.2250738585072012e-308", 0x10000000000000,
+                                 1);
+}
+
+TEST(LlvmLibcStrToDblTest, SimpleDecimalConversionExtraTypes) {
+  uint64_t double_output_mantissa = 0;
+  uint32_t output_exp2 = 0;
+
+  libc_errno = 0;
+  auto double_result =
+      internal::simple_decimal_conversion<double>("123456789012345678900");
+
+  double_output_mantissa = double_result.num.mantissa;
+  output_exp2 = double_result.num.exponent;
+
+  EXPECT_EQ(double_output_mantissa, uint64_t(0x1AC53A7E04BCDA));
+  EXPECT_EQ(output_exp2, uint32_t(1089));
+  EXPECT_EQ(double_result.error, 0);
+}
+
+} // namespace LIBC_NAMESPACE