[libc-commits] [libc] 4fde7fc - [libc][math] Add tolerance for test_subnormal_base in powf smoke test. (#229640)

[via libc-commits]

Author: lntue
Date: 2026-10-07T02:50:45Z
New Revision: 4fde7fcfe410d4cd3e97dd8740c2f931c9095f3d

URL: https://github.com/llvm/llvm-project/commit/4fde7fcfe410d4cd3e97dd8740c2f931c9095f3d
DIFF: https://github.com/llvm/llvm-project/commit/4fde7fcfe410d4cd3e97dd8740c2f931c9095f3d.diff

LOG: [libc][math] Add tolerance for test_subnormal_base in powf smoke test. (#229640)

Fix
[openmp-offload-amdgpu-runtime-2](https://lab.llvm.org/buildbot/#/builders/10)

Added: 
    

Modified: 
    libc/test/src/math/smoke/powf_test.cpp

Removed: 
    


################################################################################
diff  --git a/libc/test/src/math/smoke/powf_test.cpp b/libc/test/src/math/smoke/powf_test.cpp
index c20c02ddd8957..bda8d34f7c5f1 100644
--- a/libc/test/src/math/smoke/powf_test.cpp
+++ b/libc/test/src/math/smoke/powf_test.cpp
@@ -223,20 +223,21 @@ class PowfTest : public LIBC_NAMESPACE::testing::FPTest<float> {
   }
 
   void test_subnormal_base(float (*func)(float, float), int tolerance = 0) {
-    EXPECT_FP_EQ(0x1.0p-32f, func(0x1.0p-128f, 0.25f));
-    EXPECT_FP_EQ(0x1.0p96f, func(0x1.0p-128f, -0.75f));
     if (tolerance == 0) {
+      EXPECT_FP_EQ(0x1.0p-32f, func(0x1.0p-128f, 0.25f));
+      EXPECT_FP_EQ(0x1.0p96f, func(0x1.0p-128f, -0.75f));
       EXPECT_FP_EQ(0x1.90a962p-33f, func(0x1.8p-130f, 0.25f));
       EXPECT_FP_EQ(0x1.47238cp+32f, func(0x1.8p-130f, -0.25f));
     } else {
-      uint32_t act1 = FPBits(func(0x1.8p-130f, 0.25f)).uintval();
-      uint32_t exp1 = FPBits(0x1.90a962p-33f).uintval();
-      EXPECT_LE(act1 >= exp1 ? act1 - exp1 : exp1 - act1,
-                static_cast<uint32_t>(tolerance));
-      uint32_t act2 = FPBits(func(0x1.8p-130f, -0.25f)).uintval();
-      uint32_t exp2 = FPBits(0x1.47238cp+32f).uintval();
-      EXPECT_LE(act2 >= exp2 ? act2 - exp2 : exp2 - act2,
-                static_cast<uint32_t>(tolerance));
+      auto check_close = [tolerance](float act, float exp) {
+        uint32_t a = FPBits(act).uintval();
+        uint32_t e = FPBits(exp).uintval();
+        EXPECT_LE(a >= e ? a - e : e - a, static_cast<uint32_t>(tolerance));
+      };
+      check_close(func(0x1.0p-128f, 0.25f), 0x1.0p-32f);
+      check_close(func(0x1.0p-128f, -0.75f), 0x1.0p96f);
+      check_close(func(0x1.8p-130f, 0.25f), 0x1.90a962p-33f);
+      check_close(func(0x1.8p-130f, -0.25f), 0x1.47238cp+32f);
     }
   }