[libc-commits] [libc] a786096 - [libc][NFC] Cleanup and document utils.h
Guillaume Chatelet via libc-commits
libc-commits at lists.llvm.org
Tue Oct 18 07:32:51 PDT 2022
Author: Guillaume Chatelet
Date: 2022-10-18T14:32:39Z
New Revision: a786096f9dd20acf29d8297e706ad96de063f612
URL: https://github.com/llvm/llvm-project/commit/a786096f9dd20acf29d8297e706ad96de063f612
DIFF: https://github.com/llvm/llvm-project/commit/a786096f9dd20acf29d8297e706ad96de063f612.diff
LOG: [libc][NFC] Cleanup and document utils.h
Added:
Modified:
libc/src/string/memory_utils/elements.h
libc/src/string/memory_utils/memset_implementations.h
libc/src/string/memory_utils/utils.h
libc/test/src/string/memory_utils/memory_access_test.cpp
libc/test/src/string/memory_utils/utils_test.cpp
Removed:
################################################################################
diff --git a/libc/src/string/memory_utils/elements.h b/libc/src/string/memory_utils/elements.h
index f5a38308d5273..89a99da823d84 100644
--- a/libc/src/string/memory_utils/elements.h
+++ b/libc/src/string/memory_utils/elements.h
@@ -427,14 +427,14 @@ namespace internal {
template <Arg arg> struct ArgSelector {};
-template <> struct ArgSelector<Arg::_1> {
+template <> struct ArgSelector<Arg::P1> {
template <typename T1, typename T2>
static T1 *__restrict &Select(T1 *__restrict &p1ref, T2 *__restrict &) {
return p1ref;
}
};
-template <> struct ArgSelector<Arg::_2> {
+template <> struct ArgSelector<Arg::P2> {
template <typename T1, typename T2>
static T2 *__restrict &Select(T1 *__restrict &, T2 *__restrict &p2ref) {
return p2ref;
@@ -451,7 +451,7 @@ template <Arg arg, size_t Alignment> struct Align {
static void bump(T1 *__restrict &p1ref, T2 *__restrict &p2ref, size_t &size,
int additional_bumps = 0) {
auto &aligned_ptr = ArgSelector<arg>::Select(p1ref, p2ref);
- auto offset = offset_to_next_aligned<Alignment>(aligned_ptr);
+ auto offset = distance_to_align_up<Alignment>(aligned_ptr);
adjust(offset + additional_bumps * Alignment, p1ref, p2ref, size);
aligned_ptr = assume_aligned<Alignment>(aligned_ptr);
}
@@ -467,7 +467,7 @@ template <Arg arg, size_t Alignment> struct Align {
//
// e.g. A 16-byte Destination Aligned 32-byte Loop Copy can be written as:
// copy<Align<_16, Arg::Dst>::Then<Loop<_32>>>(dst, src, count);
-template <typename AlignmentT, Arg AlignOn = Arg::_1> struct Align {
+template <typename AlignmentT, Arg AlignOn = Arg::P1> struct Align {
private:
static constexpr size_t ALIGNMENT = AlignmentT::SIZE;
static_assert(ALIGNMENT > 1, "Alignment must be more than 1");
@@ -556,7 +556,7 @@ template <typename AlignmentT, Arg AlignOn = Arg::_1> struct Align {
static void splat_set(char *dst, const unsigned char value, size_t size) {
AlignmentT::splat_set(dst, value);
char *dummy = nullptr;
- internal::Align<Arg::_1, ALIGNMENT>::bump(dst, dummy, size);
+ internal::Align<Arg::P1, ALIGNMENT>::bump(dst, dummy, size);
NextT::splat_set(dst, value, size);
}
};
diff --git a/libc/src/string/memory_utils/memset_implementations.h b/libc/src/string/memory_utils/memset_implementations.h
index f1611a32807cb..d58ed3b703306 100644
--- a/libc/src/string/memory_utils/memset_implementations.h
+++ b/libc/src/string/memory_utils/memset_implementations.h
@@ -101,10 +101,10 @@ inline static void inline_memset(char *dst, unsigned char value, size_t count) {
return;
}
if (count >= 448 && value == 0 && hasZva())
- return splat_set<Align<_64, Arg::_1>::Then<Loop<Zva64, _64>>>(dst, 0,
+ return splat_set<Align<_64, Arg::P1>::Then<Loop<Zva64, _64>>>(dst, 0,
count);
else
- return splat_set<Align<_16, Arg::_1>::Then<Loop<_64>>>(dst, value, count);
+ return splat_set<Align<_16, Arg::P1>::Then<Loop<_64>>>(dst, value, count);
#else
/////////////////////////////////////////////////////////////////////////////
// Default
diff --git a/libc/src/string/memory_utils/utils.h b/libc/src/string/memory_utils/utils.h
index d915835e38d82..60e5a19f0e347 100644
--- a/libc/src/string/memory_utils/utils.h
+++ b/libc/src/string/memory_utils/utils.h
@@ -9,19 +9,8 @@
#ifndef LLVM_LIBC_SRC_MEMORY_UTILS_UTILS_H
#define LLVM_LIBC_SRC_MEMORY_UTILS_UTILS_H
-#include "src/__support/architectures.h"
-
-// Cache line sizes for ARM: These values are not strictly correct since
-// cache line sizes depend on implementations, not architectures. There
-// are even implementations with cache line sizes configurable at boot
-// time.
-#if defined(LLVM_LIBC_ARCH_AARCH64) || defined(LLVM_LIBC_ARCH_X86)
-#define LLVM_LIBC_CACHELINE_SIZE 64
-#elif defined(LLVM_LIBC_ARCH_ARM)
-#define LLVM_LIBC_CACHELINE_SIZE 32
-#else
-#error "Unsupported platform for memory functions."
-#endif
+#include "src/__support/CPP/bit.h"
+#include "src/__support/CPP/type_traits.h"
#include <stddef.h> // size_t
#include <stdint.h> // intptr_t / uintptr_t
@@ -62,32 +51,46 @@ static constexpr size_t ge_power2(size_t value) {
return is_power2_or_zero(value) ? value : 1ULL << (log2(value) + 1);
}
-template <size_t alignment> intptr_t offset_from_last_aligned(const void *ptr) {
+// Returns the number of bytes to substract from ptr to get to the previous
+// multiple of alignment. If ptr is already aligned returns 0.
+template <size_t alignment> uintptr_t distance_to_align_down(const void *ptr) {
static_assert(is_power2(alignment), "alignment must be a power of 2");
return reinterpret_cast<uintptr_t>(ptr) & (alignment - 1U);
}
-template <size_t alignment> intptr_t offset_to_next_aligned(const void *ptr) {
+// Returns the number of bytes to add to ptr to get to the next multiple of
+// alignment. If ptr is already aligned returns 0.
+template <size_t alignment> uintptr_t distance_to_align_up(const void *ptr) {
static_assert(is_power2(alignment), "alignment must be a power of 2");
// The logic is not straightforward and involves unsigned modulo arithmetic
// but the generated code is as fast as it can be.
return -reinterpret_cast<uintptr_t>(ptr) & (alignment - 1U);
}
-// Returns the offset from `ptr` to the next cache line.
-static inline intptr_t offset_to_next_cache_line(const void *ptr) {
- return offset_to_next_aligned<LLVM_LIBC_CACHELINE_SIZE>(ptr);
+// Returns the number of bytes to add to ptr to get to the next multiple of
+// alignment. If ptr is already aligned returns alignment.
+template <size_t alignment>
+uintptr_t distance_to_next_aligned(const void *ptr) {
+ return alignment - distance_to_align_down<alignment>(ptr);
}
+// Returns the same pointer but notifies the compiler that it is aligned.
template <size_t alignment, typename T> static T *assume_aligned(T *ptr) {
return reinterpret_cast<T *>(__builtin_assume_aligned(ptr, alignment));
}
+
#if defined __has_builtin
#if __has_builtin(__builtin_memcpy_inline)
#define LLVM_LIBC_HAS_BUILTIN_MEMCPY_INLINE
#endif
#endif
+#if defined __has_builtin
+#if __has_builtin(__builtin_memset_inline)
+#define LLVM_LIBC_HAS_BUILTIN_MEMSET_INLINE
+#endif
+#endif
+
// Performs a constant count copy.
template <size_t Size>
static inline void memcpy_inline(void *__restrict dst,
@@ -103,28 +106,21 @@ static inline void memcpy_inline(void *__restrict dst,
using Ptr = char *; // Pointer to raw data.
using CPtr = const char *; // Const pointer to raw data.
-// Loads bytes from memory (possibly unaligned) and materializes them as type.
+// Loads bytes from memory (possibly unaligned) and materializes them as
+// type.
template <typename T> static inline T load(CPtr ptr) {
T Out;
memcpy_inline<sizeof(T)>(&Out, ptr);
return Out;
}
-// Stores a value of type T in memory (possibly unaligned)
+// Stores a value of type T in memory (possibly unaligned).
template <typename T> static inline void store(Ptr ptr, T value) {
memcpy_inline<sizeof(T)>(ptr, &value);
}
-// For an operation like memset that operates on a pointer and a count, advances
-// the pointer by offset bytes and decrease count by the same amount.
-static inline void adjust(ptr
diff _t offset, Ptr &ptr, size_t &count) {
- ptr += offset;
- count -= offset;
-}
-
-// For an operation like memcpy or memcmp that operates on two pointers and a
-// count, advances the pointers by offset bytes and decrease count by the same
-// amount.
+// Advances the pointers p1 and p2 by offset bytes and decrease count by the
+// same amount.
template <typename T1, typename T2>
static inline void adjust(ptr
diff _t offset, T1 *__restrict &p1,
T2 *__restrict &p2, size_t &count) {
@@ -133,31 +129,37 @@ static inline void adjust(ptr
diff _t offset, T1 *__restrict &p1,
count -= offset;
}
-// For an operation like memset that operates on a pointer and a count, advances
-// the pointer so it is aligned to SIZE bytes and decrease count by the same
-// amount.
+// Advances p1 and p2 so p1 gets aligned to the next SIZE bytes boundary
+// and decrease count by the same amount.
// We make sure the compiler knows about the adjusted pointer alignment.
-template <size_t SIZE> void align(Ptr &ptr, size_t &count) {
- adjust(offset_to_next_aligned<SIZE>(ptr), ptr, count);
- ptr = assume_aligned<SIZE>(ptr);
+template <size_t SIZE, typename T1, typename T2>
+void align_p1_to_next_boundary(T1 *__restrict &p1, T2 *__restrict &p2,
+ size_t &count) {
+ adjust(distance_to_next_aligned<SIZE>(p1), p1, p2, count);
+ p1 = assume_aligned<SIZE>(p1);
}
-// For an operation like memcpy or memcmp that operates on two pointers and a
-// count, advances the pointers so one of them gets aligned to SIZE bytes and
-// decrease count by the same amount.
-// We make sure the compiler knows about the adjusted pointer alignment.
-enum class Arg { _1, _2, Dst = _1, Src = _2, Lhs = _1, Rhs = _2 };
+// Same as align_p1_to_next_boundary above but with a single pointer instead.
+template <size_t SIZE, typename T1>
+void align_to_next_boundary(T1 *&p1, size_t &count) {
+ CPtr dummy;
+ align_p1_to_next_boundary<SIZE>(p1, dummy, count);
+}
+
+// An enum class that discriminates between the first and second pointer.
+enum class Arg { P1, P2, Dst = P1, Src = P2 };
+
+// Same as align_p1_to_next_boundary but allows for aligning p2 instead of p1.
+// Precondition: &p1 != &p2
template <size_t SIZE, Arg AlignOn, typename T1, typename T2>
-void align(T1 *__restrict &p1, T2 *__restrict &p2, size_t &count) {
- if constexpr (AlignOn == Arg::_1) {
- adjust(offset_to_next_aligned<SIZE>(p1), p1, p2, count);
- p1 = assume_aligned<SIZE>(p1);
- } else if constexpr (AlignOn == Arg::_2) {
- adjust(offset_to_next_aligned<SIZE>(p2), p1, p2, count);
- p2 = assume_aligned<SIZE>(p2);
- } else {
- deferred_static_assert("AlignOn must be either Arg::_1 or Arg::_2");
- }
+void align_to_next_boundary(T1 *__restrict &p1, T2 *__restrict &p2,
+ size_t &count) {
+ if constexpr (AlignOn == Arg::P1)
+ align_p1_to_next_boundary<SIZE>(p1, p2, count);
+ else if constexpr (AlignOn == Arg::P2)
+ align_p1_to_next_boundary<SIZE>(p2, p1, count); // swapping p1 and p2.
+ else
+ deferred_static_assert("AlignOn must be either Arg::P1 or Arg::P2");
}
} // namespace __llvm_libc
diff --git a/libc/test/src/string/memory_utils/memory_access_test.cpp b/libc/test/src/string/memory_utils/memory_access_test.cpp
index b81700f0eb255..252cb6e518ef3 100644
--- a/libc/test/src/string/memory_utils/memory_access_test.cpp
+++ b/libc/test/src/string/memory_utils/memory_access_test.cpp
@@ -217,9 +217,9 @@ struct LlvmLibcTestAccessAlignedAccess : public LlvmLibcTestAccessBase {
expected.Touch(Offset + Size - ParamType::SIZE, ParamType::SIZE);
checkMaxAccess(expected, 3);
- checkOperations<Align<AlignmentT, Arg::_1>::Then<HeadTail<ParamType>>, Size,
+ checkOperations<Align<AlignmentT, Arg::P1>::Then<HeadTail<ParamType>>, Size,
Offset>(expected);
- checkOperations<Align<AlignmentT, Arg::_2>::Then<HeadTail<ParamType>>, Size,
+ checkOperations<Align<AlignmentT, Arg::P2>::Then<HeadTail<ParamType>>, Size,
Offset>(expected);
}
};
diff --git a/libc/test/src/string/memory_utils/utils_test.cpp b/libc/test/src/string/memory_utils/utils_test.cpp
index a20c0900b7234..5c7920c4960b6 100644
--- a/libc/test/src/string/memory_utils/utils_test.cpp
+++ b/libc/test/src/string/memory_utils/utils_test.cpp
@@ -72,55 +72,41 @@ TEST(LlvmLibcUtilsTest, GEPowerOf2) {
EXPECT_EQ(ge_power2(i), kExpectedValues[i]);
}
-using I = intptr_t;
+using UINT = uintptr_t;
// Converts an offset into a pointer.
const void *forge(size_t offset) {
return reinterpret_cast<const void *>(offset);
}
-TEST(LlvmLibcUtilsTest, OffsetToNextAligned) {
- EXPECT_EQ(offset_to_next_aligned<16>(forge(0)), I(0));
- EXPECT_EQ(offset_to_next_aligned<16>(forge(1)), I(15));
- EXPECT_EQ(offset_to_next_aligned<16>(forge(16)), I(0));
- EXPECT_EQ(offset_to_next_aligned<16>(forge(15)), I(1));
- EXPECT_EQ(offset_to_next_aligned<32>(forge(16)), I(16));
+TEST(LlvmLibcUtilsTest, DistanceToNextAligned) {
+ EXPECT_EQ(distance_to_next_aligned<16>(forge(0)), UINT(16));
+ EXPECT_EQ(distance_to_next_aligned<16>(forge(1)), UINT(15));
+ EXPECT_EQ(distance_to_next_aligned<16>(forge(16)), UINT(16));
+ EXPECT_EQ(distance_to_next_aligned<16>(forge(15)), UINT(1));
+ EXPECT_EQ(distance_to_next_aligned<32>(forge(16)), UINT(16));
}
-TEST(LlvmLibcUtilsTest, OffsetFromLastAligned) {
- EXPECT_EQ(offset_from_last_aligned<16>(forge(0)), I(0));
- EXPECT_EQ(offset_from_last_aligned<16>(forge(1)), I(1));
- EXPECT_EQ(offset_from_last_aligned<16>(forge(16)), I(0));
- EXPECT_EQ(offset_from_last_aligned<16>(forge(15)), I(15));
- EXPECT_EQ(offset_from_last_aligned<32>(forge(16)), I(16));
+TEST(LlvmLibcUtilsTest, DistanceToAlignUp) {
+ EXPECT_EQ(distance_to_align_up<16>(forge(0)), UINT(0));
+ EXPECT_EQ(distance_to_align_up<16>(forge(1)), UINT(15));
+ EXPECT_EQ(distance_to_align_up<16>(forge(16)), UINT(0));
+ EXPECT_EQ(distance_to_align_up<16>(forge(15)), UINT(1));
+ EXPECT_EQ(distance_to_align_up<32>(forge(16)), UINT(16));
}
-TEST(LlvmLibcUtilsTest, OffsetToNextCacheLine) {
- EXPECT_GT(LLVM_LIBC_CACHELINE_SIZE, 0);
- EXPECT_EQ(offset_to_next_cache_line(forge(0)), I(0));
- EXPECT_EQ(offset_to_next_cache_line(forge(1)),
- I(LLVM_LIBC_CACHELINE_SIZE - 1));
- EXPECT_EQ(offset_to_next_cache_line(forge(LLVM_LIBC_CACHELINE_SIZE)), I(0));
- EXPECT_EQ(offset_to_next_cache_line(forge(LLVM_LIBC_CACHELINE_SIZE - 1)),
- I(1));
-}
-
-TEST(LlvmLibcUtilsTest, Adjust1) {
- char a;
- const size_t base_size = 10;
- for (size_t I = -2; I < 2; ++I) {
- auto *ptr = &a;
- size_t size = base_size;
- adjust(I, ptr, size);
- EXPECT_EQ(intptr_t(ptr), intptr_t(&a + I));
- EXPECT_EQ(size, base_size - I);
- }
+TEST(LlvmLibcUtilsTest, DistanceToAlignDown) {
+ EXPECT_EQ(distance_to_align_down<16>(forge(0)), UINT(0));
+ EXPECT_EQ(distance_to_align_down<16>(forge(1)), UINT(1));
+ EXPECT_EQ(distance_to_align_down<16>(forge(16)), UINT(0));
+ EXPECT_EQ(distance_to_align_down<16>(forge(15)), UINT(15));
+ EXPECT_EQ(distance_to_align_down<32>(forge(16)), UINT(16));
}
TEST(LlvmLibcUtilsTest, Adjust2) {
char a, b;
const size_t base_size = 10;
- for (size_t I = -2; I < 2; ++I) {
+ for (ptr
diff _t I = -2; I < 2; ++I) {
auto *p1 = &a;
auto *p2 = &b;
size_t size = base_size;
@@ -131,19 +117,6 @@ TEST(LlvmLibcUtilsTest, Adjust2) {
}
}
-TEST(LlvmLibcUtilsTest, Align1) {
- char a;
- const size_t base_size = 10;
- {
- auto *ptr = &a;
- size_t size = base_size;
- align<128>(ptr, size);
- EXPECT_TRUE(uintptr_t(ptr) % 128 == 0);
- EXPECT_GE(ptr, &a);
- EXPECT_EQ(size_t(ptr - &a), base_size - size);
- }
-}
-
TEST(LlvmLibcUtilsTest, Align2) {
char a, b;
const size_t base_size = 10;
@@ -151,10 +124,10 @@ TEST(LlvmLibcUtilsTest, Align2) {
auto *p1 = &a;
auto *p2 = &b;
size_t size = base_size;
- align<128, Arg::_1>(p1, p2, size);
+ align_to_next_boundary<128, Arg::P1>(p1, p2, size);
EXPECT_TRUE(uintptr_t(p1) % 128 == 0);
- EXPECT_GE(p1, &a);
- EXPECT_GE(p2, &b);
+ EXPECT_GT(p1, &a);
+ EXPECT_GT(p2, &b);
EXPECT_EQ(size_t(p1 - &a), base_size - size);
EXPECT_EQ(size_t(p2 - &b), base_size - size);
}
@@ -162,10 +135,10 @@ TEST(LlvmLibcUtilsTest, Align2) {
auto *p1 = &a;
auto *p2 = &b;
size_t size = base_size;
- align<128, Arg::_2>(p1, p2, size);
+ align_to_next_boundary<128, Arg::P2>(p1, p2, size);
EXPECT_TRUE(uintptr_t(p2) % 128 == 0);
- EXPECT_GE(p1, &a);
- EXPECT_GE(p2, &b);
+ EXPECT_GT(p1, &a);
+ EXPECT_GT(p2, &b);
EXPECT_EQ(size_t(p1 - &a), base_size - size);
EXPECT_EQ(size_t(p2 - &b), base_size - size);
}
More information about the libc-commits
mailing list