ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[Base] Replace GCC workaround (loop opt bug) 

Previous workaround was dangerous, this one is more sane.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801#c3
This commit is contained in:
Joel Linn 2021-09-22 15:04:00 +02:00 committed by Triang3l
parent 36d8b20287
commit 247cb91ac5
1 changed files with 11 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/xenia/base/memory.cc
View File

@ -48,10 +48,10 @@ void copy_128_aligned(void* dest, const void* src, size_t count) {
// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801 // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100801
// TODO(Joel Linn): Remove this when fixed GCC versions are common place. // TODO(Joel Linn): Remove this when fixed GCC versions are common place.
#if XE_COMPILER_GNUC #if XE_COMPILER_GNUC
#define XE_WORKAROUND_LOOP_KILL_MOD(x) \ #define XE_WORKAROUND_CONSTANT_RETURN_IF(x) \
if ((count % (x)) == 0) __builtin_unreachable(); if (__builtin_constant_p(x) && (x)) return;
#else #else
#define XE_WORKAROUND_LOOP_KILL_MOD(x) #define XE_WORKAROUND_CONSTANT_RETURN_IF(x)
#endif #endif
void copy_and_swap_16_aligned(void* dest_ptr, const void* src_ptr, void copy_and_swap_16_aligned(void* dest_ptr, const void* src_ptr,
size_t count) { size_t count) {
@ -70,8 +70,8 @@ void copy_and_swap_16_aligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 8 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(8);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -90,8 +90,8 @@ void copy_and_swap_16_unaligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 8 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(8);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -113,8 +113,8 @@ void copy_and_swap_32_aligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 4 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(4);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -133,8 +133,8 @@ void copy_and_swap_32_unaligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 4 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(4);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -156,8 +156,8 @@ void copy_and_swap_64_aligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 2 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(2);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -176,8 +176,8 @@ void copy_and_swap_64_unaligned(void* dest_ptr, const void* src_ptr,
__m128i output = _mm_shuffle_epi8(input, shufmask); __m128i output = _mm_shuffle_epi8(input, shufmask);
_mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 2 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(2);
dest[i] = byte_swap(src[i]); dest[i] = byte_swap(src[i]);
} }
} }
@ -193,8 +193,8 @@ void copy_and_swap_16_in_32_aligned(void* dest_ptr, const void* src_ptr,
_mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16)); _mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16));
_mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_store_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 4 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(4);
dest[i] = (src[i] >> 16) | (src[i] << 16); dest[i] = (src[i] >> 16) | (src[i] << 16);
} }
} }
@ -210,8 +210,8 @@ void copy_and_swap_16_in_32_unaligned(void* dest_ptr, const void* src_ptr,
_mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16)); _mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16));
_mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output); _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
} }
XE_WORKAROUND_CONSTANT_RETURN_IF(count % 4 == 0);
for (; i < count; ++i) { // handle residual elements for (; i < count; ++i) { // handle residual elements
XE_WORKAROUND_LOOP_KILL_MOD(4);
dest[i] = (src[i] >> 16) | (src[i] << 16); dest[i] = (src[i] >> 16) | (src[i] << 16);
} }
} }