[libc][bazel] Add Bazel rules for rand/srand functions.

[vonosmas]

These functions are unlikely to be used in the overlay mode (since they are stateful), but internal llvm-libc implementation is used in tests for FMA family of functions, so we need Bazel support for them to ensure proper test coverage.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Alexey Samsonov (vonosmas)

Changes

These functions are unlikely to be used in the overlay mode (since they are stateful), but internal llvm-libc implementation is used in tests for FMA family of functions, so we need Bazel support for them to ensure proper test coverage.

---

Full diff: https://github.com/llvm/llvm-project/pull/159617.diff

2 Files Affected:

• (modified) utils/bazel/llvm-project-overlay/libc/BUILD.bazel (+43)
• (modified) utils/bazel/llvm-project-overlay/libc/test/src/stdlib/BUILD.bazel (+9)

diff --git a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
index cee92f2cc56c4..67c292f08be73 100644
--- a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
@@ -278,6 +278,14 @@ libc_support_library(
     hdrs = ["hdr/stdio_overlay.h"],
 )
 
+libc_support_library(
+    name = "hdr_stdlib_macros",
+    hdrs = ["hdr/stdlib_macros.h"],
+    deps = [
+        ":hdr_stdlib_overlay",
+    ],
+)
+
 libc_support_library(
     name = "hdr_stdlib_overlay",
     hdrs = ["hdr/stdlib_overlay.h"],
@@ -5066,6 +5074,41 @@ libc_function(
     ],
 )
 
+libc_support_library(
+    name = "rand_util",
+    srcs = ["src/stdlib/rand_util.cpp"],
+    hdrs = ["src/stdlib/rand_util.h"],
+    deps = [
+        ":__support_cpp_atomic",
+        ":__support_macros_attributes",
+        ":__support_macros_config",
+    ],
+)
+
+libc_function(
+    name = "rand",
+    srcs = ["src/stdlib/rand.cpp"],
+    hdrs = ["src/stdlib/rand.h"],
+    deps = [
+        ":__support_common",
+        ":__support_macros_config",
+        ":__support_threads_sleep",
+        ":hdr_stdlib_macros",
+        ":rand_util",
+    ],
+)
+
+libc_function(
+    name = "srand",
+    srcs = ["src/stdlib/srand.cpp"],
+    hdrs = ["src/stdlib/srand.h"],
+    deps = [
+        ":__support_common",
+        ":__support_macros_config",
+        ":rand_util",
+    ],
+)
+
 libc_support_library(
     name = "str_from_util",
     hdrs = ["src/stdlib/str_from_util.h"],
diff --git a/utils/bazel/llvm-project-overlay/libc/test/src/stdlib/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/test/src/stdlib/BUILD.bazel
index ce92ca8caa6f4..fd4389e8c0579 100644
--- a/utils/bazel/llvm-project-overlay/libc/test/src/stdlib/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/libc/test/src/stdlib/BUILD.bazel
@@ -159,6 +159,15 @@ libc_test(
     ],
 )
 
+libc_test(
+    name = "rand_test",
+    srcs = ["rand_test.cpp"],
+    deps = [
+        "//libc:rand",
+        "//libc:srand",
+    ],
+)
+
 libc_test_library(
     name = "strfrom_test_helper",
     hdrs = ["StrfromTest.h"],

[jhuber6]

If this is only for internal it would probably be better to make an internal randomness function that takes the state as an argument instead of it being a global. I already do that in a few places.

[vonosmas]

If this is only for internal it would probably be better to make an internal randomness function that takes the state as an argument instead of it being a global. I already do that in a few places.

Good point. Yes, the file in question is libc/test/src/math/FmaTest.h, and I agree it's probably worth migrating it to internal RNG rather than use LIBC_NAMESPACE::(s)rand directly (you even had to fix it on GPU in 71ffc1f). I'll check w/ @lntue , but this can be done independently or rand/srand Bazel support.

[lntue]

If this is only for internal it would probably be better to make an internal randomness function that takes the state as an argument instead of it being a global. I already do that in a few places.

Good point. Yes, the file in question is libc/test/src/math/FmaTest.h, and I agree it's probably worth migrating it to internal RNG rather than use LIBC_NAMESPACE::(s)rand directly (you even had to fix it on GPU in 71ffc1f). I'll check w/ @lntue , but this can be done independently or rand/srand Bazel support.

Yes, we should definitely remove the rand from the tests' dependency.

[vonosmas]

If this is only for internal it would probably be better to make an internal randomness function that takes the state as an argument instead of it being a global. I already do that in a few places.

Good point. Yes, the file in question is libc/test/src/math/FmaTest.h, and I agree it's probably worth migrating it to internal RNG rather than use LIBC_NAMESPACE::(s)rand directly (you even had to fix it on GPU in 71ffc1f). I'll check w/ @lntue , but this can be done independently or rand/srand Bazel support.

Yes, we should definitely remove the rand from the tests' dependency.

Do you think we should just move the current rand() implementation to somewhere like libc/src/__support/random.h , so that we can use it both in rand() implementation and in unit tests that need randomness?

[jhuber6]

Do you think we should just move the current rand() implementation to somewhere like libc/src/__support/random.h , so that we can use it both in rand() implementation and in unit tests that need randomness?

That's what I would assume, we already have a bunch of these floating around so we could consolidate. I have used both 32-bit and 64-bit versions of xorshift* for most of my PRNG needs. You could extract those to a common header and replace the uses there. Off the top of my head, the GPU benchmarks, allocator, and allocator test. Potentially you could replace the implementation of rand and just make it pass in the global state as a reference.