From 5bc1cb4a1f11550632d2ffcdb6f714c33e7a91f5 Mon Sep 17 00:00:00 2001
From: Rye Mutt <rye@alchemyviewer.org>
Date: Wed, 30 Dec 2020 12:41:03 -0500
Subject: [PATCH] A few small changes until we're ready to go all-in on c++17
over-aligned allocation
---
indra/llmath/llmatrix4a.h | 26 +++++++++++++++++++---
indra/llmath/llsimdtypes.h | 5 -----
indra/llmath/llvector4a.h | 45 ++++++++++++++++++++++++++++++++++++--
3 files changed, 66 insertions(+), 10 deletions(-)
diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h
index df94d7327e6..7be4e5a066a 100644
--- a/indra/llmath/llmatrix4a.h
+++ b/indra/llmath/llmatrix4a.h
@@ -32,10 +32,10 @@
#include "m4math.h"
#include "m3math.h"
-class alignas(16) LLMatrix4a
+LL_ALIGN_PREFIX(16) class LLMatrix4a
{
public:
- LLVector4a mMatrix[4];
+ LL_ALIGN_PREFIX(16) LLVector4a mMatrix[4] LL_ALIGN_POSTFIX(16);
public:
enum
{
@@ -45,6 +45,26 @@ class alignas(16) LLMatrix4a
ROW_TRANS
};
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void* operator new[](size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
+ void operator delete[](void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
LLMatrix4a() = default;
LLMatrix4a(const LLQuad& q1,const LLQuad& q2,const LLQuad& q3,const LLQuad& q4)
{
@@ -767,7 +787,7 @@ class alignas(16) LLMatrix4a
return _mm_movemask_epi8(_mm_castps_si128(_mm_and_ps(mask1, mask2))) == 0xFFFF;
}
-};
+} LL_ALIGN_POSTFIX(16);
#ifndef SHOW_ASSERT
static_assert(std::is_trivial<LLMatrix4a>::value, "LLMatrix4a must be a trivial type");
diff --git a/indra/llmath/llsimdtypes.h b/indra/llmath/llsimdtypes.h
index ce1e24dc681..715c3563480 100644
--- a/indra/llmath/llsimdtypes.h
+++ b/indra/llmath/llsimdtypes.h
@@ -66,11 +66,6 @@ class LLSimdScalar
mQ = _mm_set_ss(f);
}
- static inline LLSimdScalar getZero()
- {
- return _mm_setzero_ps();
- }
-
inline F32 getF32() const;
inline LLBool32 isApproximatelyEqual(const LLSimdScalar& rhs, F32 tolerance = F_APPROXIMATELY_ZERO) const;
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 0e564889a5e..aa75c5f076c 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -94,6 +94,26 @@ struct alignas(16) LLVector4a
static void memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes);
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void* operator new[](size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
+ void operator delete[](void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
////////////////////////////////////
// CONSTRUCTORS
////////////////////////////////////
@@ -380,7 +400,8 @@ inline std::ostream& operator<<(std::ostream& s, const LLVector4a& v)
return s;
}
-struct alignas(16) LLIVector4a
+LL_ALIGN_PREFIX(16)
+struct LLIVector4a
{
friend struct LLVector4a;
@@ -391,6 +412,26 @@ struct alignas(16) LLIVector4a
return _mm_setzero_si128();
}
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void* operator new[](size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
+ void operator delete[](void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
////////////////////////////////////
// CONSTRUCTORS
////////////////////////////////////
@@ -498,7 +539,7 @@ struct alignas(16) LLIVector4a
}
LLIQuad mQ;
-};
+} LL_ALIGN_POSTFIX(16);
static_assert(std::is_trivial<LLIVector4a>{}, "LLIVector4a must be a trivial type");
static_assert(std::is_standard_layout<LLIVector4a>{}, "LLIVector4a must be a standard layout type");
--
GitLab