TorqueGameEngines/Torque3D
1
0
Fork 0
mirror of https://github.com/TorqueGameEngines/Torque3D.git synced 2026-07-11 22:54:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

ISA backends float3 and float4 - cleanup history squash

Browse source 
working for both neon32 and neon64

Update math_backend.cpp

further sse simd additions

avx2 float3 added
added normalize_magnitude
added divide fast to float3 may copy to float4

move static spheremesh to drawSphere (initialize on first use) so platform has a chance to load the math backend

all float3 and float4 functions and isas

completed all options of float3 and float4 functions in isas and math_c
neon still to be done but that will be on mac.

Update math_backend.cpp

mac isa neon update

added float3
restructured the classes to look more like the final version of the x86 classes

linux required changes

Update build-macos-clang.yml

Update build-macos-clang.yml

Revert "Update build-macos-clang.yml"

This reverts commit 29dfc567f40f20d2400a9967a35bbdb823182e2d.

Revert "Update build-macos-clang.yml"

This reverts commit 2abad2b4ca4de717c5f4278708f289dd1bb22561.

Update CMakeLists.txt

fix macs stupid build

remove god awful rolling average from frame time tracker....

use intrinsic headers instead

each isa implementation now uses a header for that isa's intrinsic functions these are then used in the impl files. This will make it easier for matrix functions when those are implemented.

fixed comment saying 256 when it should be 512 for avx512

consolidated initializers for function tables

Update neon_intrinsics.h

fixes for some neon intrinsics no idea if this is the best way to do these but they work at least

v_cross is especially messy at the moment we basically just do it as a c math function need to look into getting this done correctly
This commit is contained in:
marauder2k7 2026-02-26 16:45:13 +00:00
parent 73ed502ac9
commit 67f12311d4
36 changed files with 1481 additions and 419 deletions
Download patch file Download diff file Expand all files Collapse all files

25
Engine/source/math/isa/neon/float3.cpp Normal file
View file

@ -0,0 +1,25 @@
#include "neon_intrinsics.h"
#include "float3_dispatch.h"
#include "float3_impl.inl"
namespace math_backend::float3::dispatch
{
// Install NEON backend
void install_neon()
{
gFloat3.add = float3_add_impl;
gFloat3.sub = float3_sub_impl;
gFloat3.mul = float3_mul_impl;
gFloat3.mul_scalar = float3_mul_scalar_impl;
gFloat3.div = float3_div_impl;
gFloat3.div_scalar = float3_div_scalar_impl;
gFloat3.dot = float3_dot_impl;
gFloat3.length = float3_length_impl;
gFloat3.lengthSquared = float3_length_squared_impl;
gFloat3.normalize = float3_normalize_impl;
gFloat3.normalize_mag = float3_normalize_mag_impl;
gFloat3.lerp = float3_lerp_impl;
gFloat3.cross = float3_cross_impl;
}
}

55
Engine/source/math/isa/neon/float4.cpp
View file

@ -1,50 +1,25 @@
#include "neon_intrinsics.h"
#include "float4_dispatch.h"
#include <arm_neon.h>
namespace
{
typedef float32x4_t f32x4;
inline f32x4 v_load(const float* p) { return vld1q_f32(p); }
inline void v_store(float* dst, f32x4 v) { vst1q_f32(dst, v); }
inline f32x4 v_set1(float s) { return vdupq_n_f32(s); }
inline f32x4 v_mul(f32x4 a, f32x4 b) { return vmulq_f32(a, b); }
inline f32x4 v_add(f32x4 a, f32x4 b) { return vaddq_f32(a, b); }
inline f32x4 v_sub(f32x4 a, f32x4 b) { return vsubq_f32(a, b); }
// AArch64 native divide
inline f32x4 v_div(f32x4 a, f32x4 b)
{
return vdivq_f32(a, b);
}
inline float v_hadd4(f32x4 a)
{
float32x2_t low = vget_low_f32(a);
float32x2_t high = vget_high_f32(a);
float32x2_t sum = vadd_f32(low, high);
sum = vpadd_f32(sum, sum);
return vget_lane_f32(sum, 0);
}
}
#include "../../impl/float4_impl.inl"
#include "float4_impl.inl"
namespace math_backend::float4::dispatch
{
// Install NEON64 backend
void install_neon()
{
gFloat4.add = float4_add_impl;
gFloat4.sub = float4_sub_impl;
gFloat4.mul = float4_mul_impl;
gFloat4.mul_scalar = float4_mul_scalar_impl;
gFloat4.div = float4_div_impl;
gFloat4.div_scalar = float4_div_scalar_impl;
gFloat4.dot = float4_dot_impl;
gFloat4.length = float4_length_impl;
gFloat4.add = float4_add_impl;
gFloat4.sub = float4_sub_impl;
gFloat4.mul = float4_mul_impl;
gFloat4.mul_scalar = float4_mul_scalar_impl;
gFloat4.div = float4_div_impl;
gFloat4.div_scalar = float4_div_scalar_impl;
gFloat4.dot = float4_dot_impl;
gFloat4.length = float4_length_impl;
gFloat4.lengthSquared = float4_length_squared_impl;
gFloat4.normalize = float4_normalize_impl;
gFloat4.lerp = float4_lerp_impl;
gFloat4.normalize = float4_normalize_impl;
gFloat4.normalize_mag = float4_normalize_mag_impl;
gFloat4.lerp = float4_lerp_impl;
gFloat4.cross = float4_cross_impl;
}
}

130
Engine/source/math/isa/neon/neon_intrinsics.h Normal file
View file

@ -0,0 +1,130 @@
#pragma once
#include <arm_neon.h>
namespace
{
typedef float32x4_t f32x4;
//------------------------------------------------------
// Load / Store
//------------------------------------------------------
inline f32x4 v_load(const float* p) { return vld1q_f32(p); }
inline void v_store(float* dst, f32x4 v) { vst1q_f32(dst, v); }
inline f32x4 v_set1(float s) { return vdupq_n_f32(s); }
inline f32x4 v_zero() { return vdupq_n_f32(0.0f); }
inline float v_extract0(f32x4 v) { return vgetq_lane_f32(v, 0); }
//------------------------------------------------------
// Mask helpers
//------------------------------------------------------
inline f32x4 v_mask_xyz()
{
// equivalent to [1,1,1,0]
float32x4_t mask = {1.0f, 1.0f, 1.0f, 0.0f};
return mask;
}
inline f32x4 v_preserve_w(f32x4 newv, f32x4 original)
{
float32x4_t mask = {0.0f, 0.0f, 0.0f, 1.0f};
return vbslq_f32(vreinterpretq_u32_f32(mask), original, newv);
}
//------------------------------------------------------
// Float3 helpers
//------------------------------------------------------
inline f32x4 v_load3_vec(const float* p) // w = 0
{
float tmp[4] = { p[0], p[1], p[2], 0.0f };
return vld1q_f32(tmp);
}
inline f32x4 v_load3_pos(const float* p) // w = 1
{
float tmp[4] = { p[0], p[1], p[2], 1.0f };
return vld1q_f32(tmp);
}
inline void v_store3(float* dst, f32x4 v)
{
float tmp[4];
vst1q_f32(tmp, v);
dst[0] = tmp[0];
dst[1] = tmp[1];
dst[2] = tmp[2];
}
//------------------------------------------------------
// Simple Arithmetic
//------------------------------------------------------
inline f32x4 v_mul(f32x4 a, f32x4 b) { return vmulq_f32(a, b); }
inline f32x4 v_div_exact(f32x4 a, f32x4 b) { return vdivq_f32(a, b); } // only NEON64
inline f32x4 v_add(f32x4 a, f32x4 b) { return vaddq_f32(a, b); }
inline f32x4 v_sub(f32x4 a, f32x4 b) { return vsubq_f32(a, b); }
//------------------------------------------------------
// Fast recip
//------------------------------------------------------
inline f32x4 v_rcp_nr(f32x4 b)
{
f32x4 r = vrecpeq_f32(b);
r = vmulq_f32(r, vrecpsq_f32(b, r)); // Newton-Raphson
r = vmulq_f32(r, vrecpsq_f32(b, r));
return r;
}
inline f32x4 v_div(f32x4 a, f32x4 b)
{
return vmulq_f32(a, v_rcp_nr(b));
}
inline f32x4 v_rsqrt_nr(f32x4 x)
{
f32x4 r = vrsqrteq_f32(x);
r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(r,r), x)); // refine
r = vmulq_f32(r, vrsqrtsq_f32(vmulq_f32(r,r), x));
return r;
}
//------------------------------------------------------
// Vector intrinsic functions
//------------------------------------------------------
inline f32x4 v_dot4(f32x4 a, f32x4 b)
{
f32x4 mul = vmulq_f32(a, b);
float32x2_t sum2 = vpadd_f32(vget_low_f32(mul), vget_high_f32(mul));
float sum = vget_lane_f32(sum2, 0) + vget_lane_f32(sum2, 1);
return vdupq_n_f32(sum);
}
inline f32x4 v_dot3(f32x4 a, f32x4 b)
{
float32x4_t mask = {1.0f, 1.0f, 1.0f, 0.0f};
f32x4 mul = vmulq_f32(a, b);
mul = vmulq_f32(mul, mask);
float32x2_t sum2 = vpadd_f32(vget_low_f32(mul), vget_high_f32(mul));
float sum = vget_lane_f32(sum2, 0) + vget_lane_f32(sum2, 1);
return vdupq_n_f32(sum);
}
inline f32x4 v_cross(f32x4 a, f32x4 b)
{
float32x4_t a_yzx = { vgetq_lane_f32(a,1), vgetq_lane_f32(a,2), vgetq_lane_f32(a,0), 0 };
float32x4_t b_yzx = { vgetq_lane_f32(b,1), vgetq_lane_f32(b,2), vgetq_lane_f32(b,0), 0 };
float32x4_t c = vsubq_f32(vmulq_f32(a, b_yzx), vmulq_f32(a_yzx, b));
return (float32x4_t){ vgetq_lane_f32(c,2), vgetq_lane_f32(c,0), vgetq_lane_f32(c,1), 0 };
}
inline f32x4 v_normalize3(f32x4 v)
{
f32x4 inv = v_rsqrt_nr(v_dot3(v,v));
return vmulq_f32(v, inv);
}
inline f32x4 v_hadd4(f32x4 a)
{
float32x2_t sum2 = vpadd_f32(vget_low_f32(a), vget_high_f32(a));
float sum = vget_lane_f32(sum2,0) + vget_lane_f32(sum2,1);
return vdupq_n_f32(sum);
}
}