[libc-commits] [libc] 7c1c07c - [libc][math][c23] Fix X86_Binary80 special cases for canonicalize functions. (#86924)
via libc-commits
libc-commits at lists.llvm.org
Thu Mar 28 13:36:53 PDT 2024
Author: Shourya Goel
Date: 2024-03-28T16:36:49-04:00
New Revision: 7c1c07c92099d3e36e8f362706a5824ab04bdf38
URL: https://github.com/llvm/llvm-project/commit/7c1c07c92099d3e36e8f362706a5824ab04bdf38
DIFF: https://github.com/llvm/llvm-project/commit/7c1c07c92099d3e36e8f362706a5824ab04bdf38.diff
LOG: [libc][math][c23] Fix X86_Binary80 special cases for canonicalize functions. (#86924)
Updates the special case of pseudo infinty as pointed in #85940
Added:
Modified:
libc/src/__support/FPUtil/BasicOperations.h
libc/test/src/math/smoke/CMakeLists.txt
libc/test/src/math/smoke/CanonicalizeTest.h
Removed:
################################################################################
diff --git a/libc/src/__support/FPUtil/BasicOperations.h b/libc/src/__support/FPUtil/BasicOperations.h
index a47931bb33900a..6e4156497618e2 100644
--- a/libc/src/__support/FPUtil/BasicOperations.h
+++ b/libc/src/__support/FPUtil/BasicOperations.h
@@ -185,27 +185,28 @@ LIBC_INLINE int canonicalize(T &cx, const T &x) {
// More precisely :
// Exponent | Significand | Meaning
// | Bits 63-62 | Bits 61-0 |
- // All Ones | 00 | Zero | Pseudo Infinity, Value = Infinty
+ // All Ones | 00 | Zero | Pseudo Infinity, Value = SNaN
// All Ones | 00 | Non-Zero | Pseudo NaN, Value = SNaN
// All Ones | 01 | Anything | Pseudo NaN, Value = SNaN
// | Bit 63 | Bits 62-0 |
// All zeroes | One | Anything | Pseudo Denormal, Value =
// | | | (−1)**s × m × 2**−16382
- // All Other | Zero | Anything | Unnormal, Value =
- // Values | | | (−1)**s × m × 2**−16382
+ // All Other | Zero | Anything | Unnormal, Value = SNaN
+ // Values | | |
bool bit63 = sx.get_implicit_bit();
UInt128 mantissa = sx.get_explicit_mantissa();
bool bit62 = static_cast<bool>((mantissa & (1ULL << 62)) >> 62);
int exponent = sx.get_biased_exponent();
if (exponent == 0x7FFF) {
if (!bit63 && !bit62) {
- if (mantissa == 0)
- cx = FPBits<T>::inf(sx.sign()).get_val();
- else {
+ if (mantissa == 0) {
cx = FPBits<T>::quiet_nan(sx.sign(), mantissa).get_val();
raise_except_if_required(FE_INVALID);
return 1;
}
+ cx = FPBits<T>::quiet_nan(sx.sign(), mantissa).get_val();
+ raise_except_if_required(FE_INVALID);
+ return 1;
} else if (!bit63 && bit62) {
cx = FPBits<T>::quiet_nan(sx.sign(), mantissa).get_val();
raise_except_if_required(FE_INVALID);
@@ -219,9 +220,11 @@ LIBC_INLINE int canonicalize(T &cx, const T &x) {
cx = x;
} else if (exponent == 0 && bit63)
cx = FPBits<T>::make_value(mantissa, 0).get_val();
- else if (exponent != 0 && !bit63)
- cx = FPBits<T>::make_value(mantissa, 0).get_val();
- else if (LIBC_UNLIKELY(sx.is_signaling_nan())) {
+ else if (exponent != 0 && !bit63) {
+ cx = FPBits<T>::quiet_nan(sx.sign(), mantissa).get_val();
+ raise_except_if_required(FE_INVALID);
+ return 1;
+ } else if (LIBC_UNLIKELY(sx.is_signaling_nan())) {
cx =
FPBits<T>::quiet_nan(sx.sign(), sx.get_explicit_mantissa()).get_val();
raise_except_if_required(FE_INVALID);
diff --git a/libc/test/src/math/smoke/CMakeLists.txt b/libc/test/src/math/smoke/CMakeLists.txt
index 3b756127fe21ea..ae2cbad7d5a7d9 100644
--- a/libc/test/src/math/smoke/CMakeLists.txt
+++ b/libc/test/src/math/smoke/CMakeLists.txt
@@ -178,6 +178,7 @@ add_fp_unittest(
libc.src.math.canonicalize
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.integer_literals
)
add_fp_unittest(
@@ -193,6 +194,7 @@ add_fp_unittest(
libc.src.math.canonicalizef
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.integer_literals
)
add_fp_unittest(
@@ -208,6 +210,7 @@ add_fp_unittest(
libc.src.math.canonicalizef128
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.integer_literals
)
add_fp_unittest(
@@ -223,6 +226,7 @@ add_fp_unittest(
libc.src.math.canonicalizel
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.integer_literals
)
add_fp_unittest(
diff --git a/libc/test/src/math/smoke/CanonicalizeTest.h b/libc/test/src/math/smoke/CanonicalizeTest.h
index c8af83f1983881..482fc3b153cb2e 100644
--- a/libc/test/src/math/smoke/CanonicalizeTest.h
+++ b/libc/test/src/math/smoke/CanonicalizeTest.h
@@ -10,6 +10,7 @@
#define LLVM_LIBC_TEST_SRC_MATH_SMOKE_CANONICALIZETEST_H
#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/integer_literals.h"
#include "test/UnitTest/FPMatcher.h"
#include "test/UnitTest/Test.h"
@@ -22,6 +23,8 @@
#define TEST_REGULAR(x, y, expected) TEST_SPECIAL(x, y, expected, 0)
+using LIBC_NAMESPACE::operator""_u128;
+
template <typename T>
class CanonicalizeTest : public LIBC_NAMESPACE::testing::Test {
@@ -55,33 +58,31 @@ class CanonicalizeTest : public LIBC_NAMESPACE::testing::Test {
T cx;
// Exponent | Significand | Meaning
// | Bits 63-62 | Bits 61-0 |
- // All Ones | 00 | Zero | Pseudo Infinity, Value = Infinty
-
- FPBits test1((UInt128(0x7FFF) << 64) + UInt128(0x0000000000000000));
+ // All Ones | 00 | Zero | Pseudo Infinity, Value = SNaN
+ FPBits test1(0x00000000'00007FFF'00000000'00000000_u128);
const T test1_val = test1.get_val();
- TEST_SPECIAL(cx, test1_val, 0, 0);
- EXPECT_FP_EQ(cx, inf);
+ TEST_SPECIAL(cx, test1_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
// Exponent | Significand | Meaning
// | Bits 63-62 | Bits 61-0 |
// All Ones | 00 | Non-Zero | Pseudo NaN, Value = SNaN
-
- FPBits test2_1((UInt128(0x7FFF) << 64) + UInt128(0x0000000000000001));
+ FPBits test2_1(0x000000000'00007FFF'00000000'00000001_u128);
const T test2_1_val = test2_1.get_val();
TEST_SPECIAL(cx, test2_1_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test2_2((UInt128(0x7FFF) << 64) + UInt128(0x0000004270000001));
+ FPBits test2_2(0x000000000'00007FFF'00000042'70000001_u128);
const T test2_2_val = test2_2.get_val();
TEST_SPECIAL(cx, test2_2_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test2_3((UInt128(0x7FFF) << 64) + UInt128(0x0000000008261001));
+ FPBits test2_3(0x000000000'00007FFF'00000000'08261001_u128);
const T test2_3_val = test2_3.get_val();
TEST_SPECIAL(cx, test2_3_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test2_4((UInt128(0x7FFF) << 64) + UInt128(0x0000780008261001));
+ FPBits test2_4(0x000000000'00007FFF'00007800'08261001_u128);
const T test2_4_val = test2_4.get_val();
TEST_SPECIAL(cx, test2_4_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
@@ -89,23 +90,22 @@ class CanonicalizeTest : public LIBC_NAMESPACE::testing::Test {
// Exponent | Significand | Meaning
// | Bits 63-62 | Bits 61-0 |
// All Ones | 01 | Anything | Pseudo NaN, Value = SNaN
-
- FPBits test3_1((UInt128(0x7FFF) << 64) + UInt128(0x4000000000000000));
+ FPBits test3_1(0x00000000'00007FFF'40000000'00000000_u128);
const T test3_1_val = test3_1.get_val();
TEST_SPECIAL(cx, test3_1_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test3_2((UInt128(0x7FFF) << 64) + UInt128(0x4000004270000001));
+ FPBits test3_2(0x00000000'00007FFF'40000042'70000001_u128);
const T test3_2_val = test3_2.get_val();
TEST_SPECIAL(cx, test3_2_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test3_3((UInt128(0x7FFF) << 64) + UInt128(0x4000000008261001));
+ FPBits test3_3(0x00000000'00007FFF'40000000'08261001_u128);
const T test3_3_val = test3_3.get_val();
TEST_SPECIAL(cx, test3_3_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
- FPBits test3_4((UInt128(0x7FFF) << 64) + UInt128(0x4007800008261001));
+ FPBits test3_4(0x00000000'00007FFF'40007800'08261001_u128);
const T test3_4_val = test3_4.get_val();
TEST_SPECIAL(cx, test3_4_val, 1, FE_INVALID);
EXPECT_FP_EQ(cx, aNaN);
@@ -114,20 +114,18 @@ class CanonicalizeTest : public LIBC_NAMESPACE::testing::Test {
// | Bit 63 | Bits 62-0 |
// All zeroes | One | Anything | Pseudo Denormal, Value =
// | | | (−1)**s × m × 2**−16382
-
- FPBits test4_1((UInt128(0x0000) << 64) + UInt128(0x8000000000000000));
+ FPBits test4_1(0x00000000'00000000'80000000'00000000_u128);
const T test4_1_val = test4_1.get_val();
TEST_SPECIAL(cx, test4_1_val, 0, 0);
EXPECT_FP_EQ(
cx, FPBits::make_value(test4_1.get_explicit_mantissa(), 0).get_val());
- FPBits test4_2((UInt128(0x0000) << 64) + UInt128(0x8000004270000001));
+ FPBits test4_2(0x00000000'00000000'80000042'70000001_u128);
const T test4_2_val = test4_2.get_val();
TEST_SPECIAL(cx, test4_2_val, 0, 0);
EXPECT_FP_EQ(
cx, FPBits::make_value(test4_2.get_explicit_mantissa(), 0).get_val());
-
- FPBits test4_3((UInt128(0x0000) << 64) + UInt128(0x8000000008261001));
+ FPBits test4_3(0x00000000'00000000'80000000'08261001_u128);
const T test4_3_val = test4_3.get_val();
TEST_SPECIAL(cx, test4_3_val, 0, 0);
EXPECT_FP_EQ(
@@ -135,44 +133,37 @@ class CanonicalizeTest : public LIBC_NAMESPACE::testing::Test {
// Exponent | Significand | Meaning
// | Bit 63 | Bits 62-0 |
- // All Other | Zero | Anything | Unnormal, Value =
- // Values | | | (−1)**s × m × 2**−16382
-
- FPBits test5_1(UInt128(0x0000000000000001));
+ // All Other | Zero | Anything | Unnormal, Value = SNaN
+ // Values | | |
+ FPBits test5_1(0x00000000'00000040'00000000'00000001_u128);
const T test5_1_val = test5_1.get_val();
- TEST_SPECIAL(cx, test5_1_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_1.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_1_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
- FPBits test5_2(UInt128(0x0000004270000001));
+ FPBits test5_2(0x00000000'00000230'00000042'70000001_u128);
const T test5_2_val = test5_2.get_val();
- TEST_SPECIAL(cx, test5_2_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_2.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_2_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
- FPBits test5_3(UInt128(0x0000000008261001));
+ FPBits test5_3(0x00000000'00000560'00000000'08261001_u128);
const T test5_3_val = test5_3.get_val();
- TEST_SPECIAL(cx, test5_3_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_3.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_3_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
- FPBits test5_4(UInt128(0x0000002816000000));
+ FPBits test5_4(0x00000000'00000780'00000028'16000000_u128);
const T test5_4_val = test5_4.get_val();
- TEST_SPECIAL(cx, test5_4_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_4.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_4_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
- FPBits test5_5(UInt128(0x0000004270000001));
+ FPBits test5_5(0x00000000'00000900'00000042'70000001_u128);
const T test5_5_val = test5_5.get_val();
- TEST_SPECIAL(cx, test5_5_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_5.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_5_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
- FPBits test5_6(UInt128(0x0000000008261001));
+ FPBits test5_6(0x00000000'00000AB0'00000000'08261001_u128);
const T test5_6_val = test5_6.get_val();
- TEST_SPECIAL(cx, test5_6_val, 0, 0);
- EXPECT_FP_EQ(
- cx, FPBits::make_value(test5_6.get_explicit_mantissa(), 0).get_val());
+ TEST_SPECIAL(cx, test5_6_val, 1, FE_INVALID);
+ EXPECT_FP_EQ(cx, aNaN);
}
}
More information about the libc-commits
mailing list