id/doom3-bfg
0
0
Fork 0
mirror of https://github.com/id-Software/DOOM-3-BFG.git synced 2024-12-02 17:02:17 +00:00
Code Issues Projects Releases Packages Wiki Activity
doom3-bfg/neo/idlib/geometry/RenderMatrix.cpp

3329 lines
142 KiB
C++
Raw Blame History

/*
===========================================================================
Doom 3 BFG Edition GPL Source Code
Copyright (C) 1993-2012 id Software LLC, a ZeniMax Media company.
This file is part of the Doom 3 BFG Edition GPL Source Code ("Doom 3 BFG Edition Source Code").
Doom 3 BFG Edition Source Code is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
Doom 3 BFG Edition Source Code is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with Doom 3 BFG Edition Source Code. If not, see <http://www.gnu.org/licenses/>.
In addition, the Doom 3 BFG Edition Source Code is also subject to certain additional terms. You should have received a copy of these additional terms immediately following the terms and conditions of the GNU General Public License which accompanied the Doom 3 BFG Edition Source Code. If not, please request a copy in writing from id Software at the address below.
If you have questions concerning this license or the applicable additional terms, you may contact in writing id Software LLC, c/o ZeniMax Media Inc., Suite 120, Rockville, Maryland 20850 USA.
===========================================================================
*/
#include "../ParallelJobList_JobHeaders.h"
#include "../math/Math.h"
#include "../math/Vector.h"
#include "../math/Matrix.h"
#include "../math/Rotation.h"
#include "../math/Plane.h"
#include "../bv/Sphere.h"
#include "../bv/Bounds.h"
#include "RenderMatrix.h"
// FIXME: it would be nice if all render matrices were 16-byte aligned
// so there is no need for unaligned loads and stores everywhere
#ifdef _lint
#undef ID_WIN_X86_SSE2_INTRIN
#endif
//lint -e438 // the non-SSE code isn't lint friendly, either
//lint -e550
#define RENDER_MATRIX_INVERSE_EPSILON 1e-16f // JDC: changed from 1e-14f to allow full wasteland parallel light projections to invert
#define RENDER_MATRIX_INFINITY 1e30f // NOTE: cannot initiaize a vec_float4 with idMath::INFINITY on the SPU
#define RENDER_MATRIX_PROJECTION_EPSILON 0.1f
#define CLIP_SPACE_OGL // the OpenGL clip space Z is in the range [-1, 1]
/*
================================================================================================
Constant render matrices
================================================================================================
*/
// identity matrix
ALIGNTYPE16 const idRenderMatrix renderMatrix_identity(
1.0f, 0.0f, 0.0f, 0.0f,
0.0f, 1.0f, 0.0f, 0.0f,
0.0f, 0.0f, 1.0f, 0.0f,
0.0f, 0.0f, 0.0f, 1.0f
);
// convert from our coordinate system (looking down X) to OpenGL's coordinate system (looking down -Z)
ALIGNTYPE16 const idRenderMatrix renderMatrix_flipToOpenGL(
0.0f, -1.0f, 0.0f, 0.0f,
0.0f, 0.0f, 1.0f, 0.0f,
-1.0f, 0.0f, 0.0f, 0.0f,
0.0f, 0.0f, 0.0f, 1.0f
);
// OpenGL -1 to 1.
ALIGNTYPE16 const idRenderMatrix renderMatrix_windowSpaceToClipSpace(
2.0f, 0.0f, 0.0f, -1.0f,
0.0f, 2.0f, 0.0f, -1.0f,
0.0f, 0.0f, 2.0f, -1.0f,
0.0f, 0.0f, 0.0f, 1.0f
);
/*
================================================================================================
SIMD constants
================================================================================================
*/
static const __m128i vector_int_1 = _mm_set1_epi32( 1 );
static const __m128i vector_int_4 = _mm_set1_epi32( 4 );
static const __m128i vector_int_0123 = _mm_set_epi32( 3, 2, 1, 0 );
static const __m128 vector_float_mask0 = __m128c( _mm_set1_epi32( 1 << 0 ) );
static const __m128 vector_float_mask1 = __m128c( _mm_set1_epi32( 1 << 1 ) );
static const __m128 vector_float_mask2 = __m128c( _mm_set1_epi32( 1 << 2 ) );
static const __m128 vector_float_mask3 = __m128c( _mm_set1_epi32( 1 << 3 ) );
static const __m128 vector_float_mask4 = __m128c( _mm_set1_epi32( 1 << 4 ) );
static const __m128 vector_float_mask5 = __m128c( _mm_set1_epi32( 1 << 5 ) );
static const __m128 vector_float_sign_bit = __m128c( _mm_set1_epi32( IEEE_FLT_SIGN_MASK ) );
static const __m128 vector_float_abs_mask = __m128c( _mm_set1_epi32( ~IEEE_FLT_SIGN_MASK ) );
static const __m128 vector_float_keep_last = __m128c( _mm_set_epi32( -1, 0, 0, 0 ) );
static const __m128 vector_float_inverse_epsilon = { RENDER_MATRIX_INVERSE_EPSILON, RENDER_MATRIX_INVERSE_EPSILON, RENDER_MATRIX_INVERSE_EPSILON, RENDER_MATRIX_INVERSE_EPSILON };
static const __m128 vector_float_smallest_non_denorm = { 1.1754944e-038f, 1.1754944e-038f, 1.1754944e-038f, 1.1754944e-038f };
static const __m128 vector_float_pos_infinity = { RENDER_MATRIX_INFINITY, RENDER_MATRIX_INFINITY, RENDER_MATRIX_INFINITY, RENDER_MATRIX_INFINITY };
static const __m128 vector_float_neg_infinity = { -RENDER_MATRIX_INFINITY, -RENDER_MATRIX_INFINITY, -RENDER_MATRIX_INFINITY, -RENDER_MATRIX_INFINITY };
static const __m128 vector_float_zero = { 0.0f, 0.0f, 0.0f, 0.0f };
static const __m128 vector_float_half = { 0.5f, 0.5f, 0.5f, 0.5f };
static const __m128 vector_float_neg_half = { -0.5f, -0.5f, -0.5f, -0.5f };
static const __m128 vector_float_one = { 1.0f, 1.0f, 1.0f, 1.0f };
static const __m128 vector_float_pos_one = { +1.0f, +1.0f, +1.0f, +1.0f };
static const __m128 vector_float_neg_one = { -1.0f, -1.0f, -1.0f, -1.0f };
static const __m128 vector_float_last_one = { 0.0f, 0.0f, 0.0f, 1.0f };
/*
================================================================================================
Box definition
================================================================================================
*/
/*
4----{E}---5
+ /| /|
Z {H} {I} {F} |
- / | / {J}
7--{G}-----6 |
| | | |
{L} 0----|-{A}-1
| / {K} / -
| {D} | {B} Y
|/ |/ +
3---{C}----2
- X +
*/
static const short boxPolygonVertices[6][4] =
{
{ 0, 3, 7, 4 }, // neg-X
{ 0, 1, 5, 4 }, // neg-Y
{ 0, 1, 2, 3 }, // neg-Z
{ 1, 2, 6, 5 }, // pos-X
{ 2, 3, 7, 6 }, // pos-Y
{ 4, 5, 6, 7 } // pos-Z
};
static const short boxEdgeVertices[12][2] =
{
/* A = */ { 0, 1 }, /* B = */ { 1, 2 }, /* C = */ { 2, 3 }, /* D = */ { 3, 0 }, // bottom
/* E = */ { 4, 5 }, /* F = */ { 5, 6 }, /* G = */ { 6, 7 }, /* H = */ { 7, 4 }, // top
/* I = */ { 0, 4 }, /* J = */ { 1, 5 }, /* K = */ { 2, 6 }, /* L = */ { 3, 7 } // sides
};
static int boxEdgePolygons[12][2] =
{
/* A = */ { 1, 2 }, /* B = */ { 3, 2 }, /* C = */ { 4, 2 }, /* D = */ { 0, 2 }, // bottom
/* E = */ { 1, 5 }, /* F = */ { 3, 5 }, /* G = */ { 4, 5 }, /* H = */ { 0, 5 }, // top
/* I = */ { 0, 1 }, /* J = */ { 3, 1 }, /* K = */ { 3, 4 }, /* L = */ { 0, 4 } // sides
};
/*
#include <Windows.h>
class idCreateBoxFrontPolygonsForFrontBits {
public:
idCreateBoxFrontPolygonsForFrontBits() {
for ( int i = 0; i < 64; i++ ) {
int frontPolygons[7] = { 0 };
int numFrontPolygons = 0;
char bits[7] = { 0 };
for ( int j = 0; j < 6; j++ ) {
if ( ( i & ( 1 << j ) ) != 0 ) {
frontPolygons[numFrontPolygons++] = j;
bits[5 - j] = '1';
} else {
bits[5 - j] = '0';
}
}
const char * comment = ( ( i & ( i >> 3 ) & 7 ) != 0 ) ? " invalid" : "";
if ( i == 0 ) {
comment = " inside the box, every polygon is considered front facing";
numFrontPolygons = 6;
for ( int j = 0; j < 6; j++ ) {
frontPolygons[j] = j;
}
}
char buffer[1024];
sprintf( buffer, "{ { %d, %d, %d, %d, %d, %d, %d }, %d }, // %s = %d%s\n",
frontPolygons[0], frontPolygons[1], frontPolygons[2], frontPolygons[3],
frontPolygons[4], frontPolygons[5], frontPolygons[6],
numFrontPolygons, bits, i, comment );
OutputDebugString( buffer );
}
}
} createBoxFrontPolygonsForFrontBits;
*/
// make sure this is a power of two for fast addressing an array of these without integer multiplication
static const struct frontPolygons_t
{
byte indices[7];
byte count;
} boxFrontPolygonsForFrontBits[64] =
{
{ { 0, 1, 2, 3, 4, 5, 0 }, 6 }, // 000000 = 0 inside the box, every polygon is considered front facing
{ { 0, 0, 0, 0, 0, 0, 0 }, 1 }, // 000001 = 1
{ { 1, 0, 0, 0, 0, 0, 0 }, 1 }, // 000010 = 2
{ { 0, 1, 0, 0, 0, 0, 0 }, 2 }, // 000011 = 3
{ { 2, 0, 0, 0, 0, 0, 0 }, 1 }, // 000100 = 4
{ { 0, 2, 0, 0, 0, 0, 0 }, 2 }, // 000101 = 5
{ { 1, 2, 0, 0, 0, 0, 0 }, 2 }, // 000110 = 6
{ { 0, 1, 2, 0, 0, 0, 0 }, 3 }, // 000111 = 7
{ { 3, 0, 0, 0, 0, 0, 0 }, 1 }, // 001000 = 8
{ { 0, 3, 0, 0, 0, 0, 0 }, 2 }, // 001001 = 9 invalid
{ { 1, 3, 0, 0, 0, 0, 0 }, 2 }, // 001010 = 10
{ { 0, 1, 3, 0, 0, 0, 0 }, 3 }, // 001011 = 11 invalid
{ { 2, 3, 0, 0, 0, 0, 0 }, 2 }, // 001100 = 12
{ { 0, 2, 3, 0, 0, 0, 0 }, 3 }, // 001101 = 13 invalid
{ { 1, 2, 3, 0, 0, 0, 0 }, 3 }, // 001110 = 14
{ { 0, 1, 2, 3, 0, 0, 0 }, 4 }, // 001111 = 15 invalid
{ { 4, 0, 0, 0, 0, 0, 0 }, 1 }, // 010000 = 16
{ { 0, 4, 0, 0, 0, 0, 0 }, 2 }, // 010001 = 17
{ { 1, 4, 0, 0, 0, 0, 0 }, 2 }, // 010010 = 18 invalid
{ { 0, 1, 4, 0, 0, 0, 0 }, 3 }, // 010011 = 19 invalid
{ { 2, 4, 0, 0, 0, 0, 0 }, 2 }, // 010100 = 20
{ { 0, 2, 4, 0, 0, 0, 0 }, 3 }, // 010101 = 21
{ { 1, 2, 4, 0, 0, 0, 0 }, 3 }, // 010110 = 22 invalid
{ { 0, 1, 2, 4, 0, 0, 0 }, 4 }, // 010111 = 23 invalid
{ { 3, 4, 0, 0, 0, 0, 0 }, 2 }, // 011000 = 24
{ { 0, 3, 4, 0, 0, 0, 0 }, 3 }, // 011001 = 25 invalid
{ { 1, 3, 4, 0, 0, 0, 0 }, 3 }, // 011010 = 26 invalid
{ { 0, 1, 3, 4, 0, 0, 0 }, 4 }, // 011011 = 27 invalid
{ { 2, 3, 4, 0, 0, 0, 0 }, 3 }, // 011100 = 28
{ { 0, 2, 3, 4, 0, 0, 0 }, 4 }, // 011101 = 29 invalid
{ { 1, 2, 3, 4, 0, 0, 0 }, 4 }, // 011110 = 30 invalid
{ { 0, 1, 2, 3, 4, 0, 0 }, 5 }, // 011111 = 31 invalid
{ { 5, 0, 0, 0, 0, 0, 0 }, 1 }, // 100000 = 32
{ { 0, 5, 0, 0, 0, 0, 0 }, 2 }, // 100001 = 33
{ { 1, 5, 0, 0, 0, 0, 0 }, 2 }, // 100010 = 34
{ { 0, 1, 5, 0, 0, 0, 0 }, 3 }, // 100011 = 35
{ { 2, 5, 0, 0, 0, 0, 0 }, 2 }, // 100100 = 36 invalid
{ { 0, 2, 5, 0, 0, 0, 0 }, 3 }, // 100101 = 37 invalid
{ { 1, 2, 5, 0, 0, 0, 0 }, 3 }, // 100110 = 38 invalid
{ { 0, 1, 2, 5, 0, 0, 0 }, 4 }, // 100111 = 39 invalid
{ { 3, 5, 0, 0, 0, 0, 0 }, 2 }, // 101000 = 40
{ { 0, 3, 5, 0, 0, 0, 0 }, 3 }, // 101001 = 41 invalid
{ { 1, 3, 5, 0, 0, 0, 0 }, 3 }, // 101010 = 42
{ { 0, 1, 3, 5, 0, 0, 0 }, 4 }, // 101011 = 43 invalid
{ { 2, 3, 5, 0, 0, 0, 0 }, 3 }, // 101100 = 44 invalid
{ { 0, 2, 3, 5, 0, 0, 0 }, 4 }, // 101101 = 45 invalid
{ { 1, 2, 3, 5, 0, 0, 0 }, 4 }, // 101110 = 46 invalid
{ { 0, 1, 2, 3, 5, 0, 0 }, 5 }, // 101111 = 47 invalid
{ { 4, 5, 0, 0, 0, 0, 0 }, 2 }, // 110000 = 48
{ { 0, 4, 5, 0, 0, 0, 0 }, 3 }, // 110001 = 49
{ { 1, 4, 5, 0, 0, 0, 0 }, 3 }, // 110010 = 50 invalid
{ { 0, 1, 4, 5, 0, 0, 0 }, 4 }, // 110011 = 51 invalid
{ { 2, 4, 5, 0, 0, 0, 0 }, 3 }, // 110100 = 52 invalid
{ { 0, 2, 4, 5, 0, 0, 0 }, 4 }, // 110101 = 53 invalid
{ { 1, 2, 4, 5, 0, 0, 0 }, 4 }, // 110110 = 54 invalid
{ { 0, 1, 2, 4, 5, 0, 0 }, 5 }, // 110111 = 55 invalid
{ { 3, 4, 5, 0, 0, 0, 0 }, 3 }, // 111000 = 56
{ { 0, 3, 4, 5, 0, 0, 0 }, 4 }, // 111001 = 57 invalid
{ { 1, 3, 4, 5, 0, 0, 0 }, 4 }, // 111010 = 58 invalid
{ { 0, 1, 3, 4, 5, 0, 0 }, 5 }, // 111011 = 59 invalid
{ { 2, 3, 4, 5, 0, 0, 0 }, 4 }, // 111100 = 60 invalid
{ { 0, 2, 3, 4, 5, 0, 0 }, 5 }, // 111101 = 61 invalid
{ { 1, 2, 3, 4, 5, 0, 0 }, 5 }, // 111110 = 62 invalid
{ { 0, 1, 2, 3, 4, 5, 0 }, 6 }, // 111111 = 63 invalid
};
/*
#include <Windows.h>
class idCreateBoxSilhouetteEdgesForFrontBits {
public:
idCreateBoxSilhouetteEdgesForFrontBits() {
for ( int i = 0; i < 64; i++ ) {
int silhouetteEdges[12] = { 0 };
int numSilhouetteEdges = 0;
for ( int j = 0; j < 12; j++ ) {
if ( i == 0 || ( ( i >> boxEdgePolygons[j][0] ) & 1 ) != ( ( i >> boxEdgePolygons[j][1] ) & 1 ) ) {
silhouetteEdges[numSilhouetteEdges++] = j;
}
}
char bits[7] = { 0 };
for ( int j = 0; j < 6; j++ ) {
if ( ( i & ( 1 << j ) ) != 0 ) {
bits[5 - j] = '1';
} else {
bits[5 - j] = '0';
}
}
const char * comment = ( ( i & ( i >> 3 ) & 7 ) != 0 ) ? " invalid" : "";
if ( i == 0 ) {
comment = " inside the box, every edge is considered part of the silhouette";
}
char buffer[1024];
sprintf( buffer, "{ { %2d, %2d, %2d, %2d, %2d, %2d, %2d, %2d, %2d, %2d, %2d, %2d }, %2d }, // %s = %d%s\n",
silhouetteEdges[0], silhouetteEdges[1], silhouetteEdges[2], silhouetteEdges[3],
silhouetteEdges[4], silhouetteEdges[5], silhouetteEdges[6], silhouetteEdges[7],
silhouetteEdges[8], silhouetteEdges[9], silhouetteEdges[10], silhouetteEdges[11],
numSilhouetteEdges, bits, i, comment );
OutputDebugString( buffer );
}
}
} createBoxSilhouetteEdgesForFrontBits;
*/
// make sure this is a power of two for fast addressing an array of these without integer multiplication
static const struct silhouetteEdges_t
{
byte indices[12];
int32 count;
} boxSilhouetteEdgesForFrontBits[64] =
{
{ { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 }, 12 }, // 000000 = 0 inside the box, every edge is considered part of the silhouette
{ { 3, 7, 8, 11, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 000001 = 1
{ { 0, 4, 8, 9, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 000010 = 2
{ { 0, 3, 4, 7, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 000011 = 3
{ { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 000100 = 4
{ { 0, 1, 2, 7, 8, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 000101 = 5
{ { 1, 2, 3, 4, 8, 9, 0, 0, 0, 0, 0, 0 }, 6 }, // 000110 = 6
{ { 1, 2, 4, 7, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 000111 = 7
{ { 1, 5, 9, 10, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 001000 = 8
{ { 1, 3, 5, 7, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 001001 = 9 invalid
{ { 0, 1, 4, 5, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 001010 = 10
{ { 0, 1, 3, 4, 5, 7, 10, 11, 0, 0, 0, 0 }, 8 }, // 001011 = 11 invalid
{ { 0, 2, 3, 5, 9, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 001100 = 12
{ { 0, 2, 5, 7, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 001101 = 13 invalid
{ { 2, 3, 4, 5, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 001110 = 14
{ { 2, 4, 5, 7, 10, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 001111 = 15 invalid
{ { 2, 6, 10, 11, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 010000 = 16
{ { 2, 3, 6, 7, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 010001 = 17
{ { 0, 2, 4, 6, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 010010 = 18 invalid
{ { 0, 2, 3, 4, 6, 7, 9, 10, 0, 0, 0, 0 }, 8 }, // 010011 = 19 invalid
{ { 0, 1, 3, 6, 10, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 010100 = 20
{ { 0, 1, 6, 7, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 010101 = 21
{ { 1, 3, 4, 6, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 010110 = 22 invalid
{ { 1, 4, 6, 7, 9, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 010111 = 23 invalid
{ { 1, 2, 5, 6, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 011000 = 24
{ { 1, 2, 3, 5, 6, 7, 8, 9, 0, 0, 0, 0 }, 8 }, // 011001 = 25 invalid
{ { 0, 1, 2, 4, 5, 6, 8, 11, 0, 0, 0, 0 }, 8 }, // 011010 = 26 invalid
{ { 0, 1, 2, 3, 4, 5, 6, 7, 0, 0, 0, 0 }, 8 }, // 011011 = 27 invalid
{ { 0, 3, 5, 6, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 011100 = 28
{ { 0, 5, 6, 7, 8, 9, 0, 0, 0, 0, 0, 0 }, 6 }, // 011101 = 29 invalid
{ { 3, 4, 5, 6, 8, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 011110 = 30 invalid
{ { 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 011111 = 31 invalid
{ { 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 100000 = 32
{ { 3, 4, 5, 6, 8, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 100001 = 33
{ { 0, 5, 6, 7, 8, 9, 0, 0, 0, 0, 0, 0 }, 6 }, // 100010 = 34
{ { 0, 3, 5, 6, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 100011 = 35
{ { 0, 1, 2, 3, 4, 5, 6, 7, 0, 0, 0, 0 }, 8 }, // 100100 = 36 invalid
{ { 0, 1, 2, 4, 5, 6, 8, 11, 0, 0, 0, 0 }, 8 }, // 100101 = 37 invalid
{ { 1, 2, 3, 5, 6, 7, 8, 9, 0, 0, 0, 0 }, 8 }, // 100110 = 38 invalid
{ { 1, 2, 5, 6, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 100111 = 39 invalid
{ { 1, 4, 6, 7, 9, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 101000 = 40
{ { 1, 3, 4, 6, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 101001 = 41 invalid
{ { 0, 1, 6, 7, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 101010 = 42
{ { 0, 1, 3, 6, 10, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 101011 = 43 invalid
{ { 0, 2, 3, 4, 6, 7, 9, 10, 0, 0, 0, 0 }, 8 }, // 101100 = 44 invalid
{ { 0, 2, 4, 6, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 101101 = 45 invalid
{ { 2, 3, 6, 7, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 101110 = 46 invalid
{ { 2, 6, 10, 11, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 101111 = 47 invalid
{ { 2, 4, 5, 7, 10, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 110000 = 48
{ { 2, 3, 4, 5, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 110001 = 49
{ { 0, 2, 5, 7, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 110010 = 50 invalid
{ { 0, 2, 3, 5, 9, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 110011 = 51 invalid
{ { 0, 1, 3, 4, 5, 7, 10, 11, 0, 0, 0, 0 }, 8 }, // 110100 = 52 invalid
{ { 0, 1, 4, 5, 8, 10, 0, 0, 0, 0, 0, 0 }, 6 }, // 110101 = 53 invalid
{ { 1, 3, 5, 7, 8, 9, 10, 11, 0, 0, 0, 0 }, 8 }, // 110110 = 54 invalid
{ { 1, 5, 9, 10, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 110111 = 55 invalid
{ { 1, 2, 4, 7, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 111000 = 56
{ { 1, 2, 3, 4, 8, 9, 0, 0, 0, 0, 0, 0 }, 6 }, // 111001 = 57 invalid
{ { 0, 1, 2, 7, 8, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 111010 = 58 invalid
{ { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 111011 = 59 invalid
{ { 0, 3, 4, 7, 9, 11, 0, 0, 0, 0, 0, 0 }, 6 }, // 111100 = 60 invalid
{ { 0, 4, 8, 9, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 111101 = 61 invalid
{ { 3, 7, 8, 11, 0, 0, 0, 0, 0, 0, 0, 0 }, 4 }, // 111110 = 62 invalid
{ { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, 0 }, // 111111 = 63 invalid
};
/*
#include <Windows.h>
class idCreateBoxSilhouetteVerticesForFrontBits {
public:
idCreateBoxSilhouetteVerticesForFrontBits() {
for ( int i = 0; i < 64; i++ ) {
int silhouetteEdges[12] = { 0 };
int numSilhouetteEdges = 0;
for ( int j = 0; j < 12; j++ ) {
if ( i == 0 || ( ( i >> boxEdgePolygons[j][0] ) & 1 ) != ( ( i >> boxEdgePolygons[j][1] ) & 1 ) ) {
silhouetteEdges[numSilhouetteEdges++] = j;
}
}
int silhouetteVertices[8] = { 0 };
int numSilhouetteVertices = 0;
int vertex = boxEdgeVertices[silhouetteEdges[0]][0];
for ( int j = 0; j < 7; j++ ) {
int newVertex = -1;
for ( int j = 0; j < numSilhouetteEdges; j++ ) {
if ( silhouetteEdges[j] == -1 ) {
continue;
}
if ( boxEdgeVertices[silhouetteEdges[j]][0] == vertex ) {
newVertex = boxEdgeVertices[silhouetteEdges[j]][1];
silhouetteEdges[j] = -1;
break;
} else if ( boxEdgeVertices[silhouetteEdges[j]][1] == vertex ) {
newVertex = boxEdgeVertices[silhouetteEdges[j]][0];
silhouetteEdges[j] = -1;
break;
}
}
if ( newVertex == -1 ) {
break;
}
silhouetteVertices[numSilhouetteVertices++] = newVertex;
vertex = newVertex;
}
char bits[7] = { 0 };
for ( int j = 0; j < 6; j++ ) {
if ( ( i & ( 1 << j ) ) != 0 ) {
bits[5 - j] = '1';
} else {
bits[5 - j] = '0';
}
}
const char * comment = ( ( i & ( i >> 3 ) & 7 ) != 0 ) ? " invalid" : "";
if ( i == 0 ) {
comment = " inside the box, no silhouette";
}
char buffer[1024];
sprintf( buffer, "{ { %d, %d, %d, %d, %d, %d, %d }, %d }, // %s = %d%s\n",
silhouetteVertices[0], silhouetteVertices[1], silhouetteVertices[2], silhouetteVertices[3],
silhouetteVertices[4], silhouetteVertices[5], silhouetteVertices[6], numSilhouetteVertices, bits, i, comment );
OutputDebugString( buffer );
}
}
} createBoxSilhouetteVerticesForFrontBits;
*/
// make sure this is a power of two for fast addressing an array of these without integer multiplication
static const struct silhouetteVertices_t
{
byte indices[7];
byte count;
} boxSilhouetteVerticesForFrontBits[64] =
{
{ { 1, 2, 3, 0, 4, 5, 6 }, 7 }, // 000000 = 0 inside the box, no vertex is considered part of the silhouette
{ { 0, 4, 7, 3, 0, 0, 0 }, 4 }, // 000001 = 1
{ { 1, 5, 4, 0, 0, 0, 0 }, 4 }, // 000010 = 2
{ { 1, 5, 4, 7, 3, 0, 0 }, 6 }, // 000011 = 3
{ { 1, 2, 3, 0, 0, 0, 0 }, 4 }, // 000100 = 4
{ { 1, 2, 3, 7, 4, 0, 0 }, 6 }, // 000101 = 5
{ { 2, 3, 0, 4, 5, 1, 0 }, 6 }, // 000110 = 6
{ { 2, 3, 7, 4, 5, 1, 0 }, 6 }, // 000111 = 7
{ { 2, 6, 5, 1, 0, 0, 0 }, 4 }, // 001000 = 8
{ { 2, 6, 5, 1, 0, 0, 0 }, 4 }, // 001001 = 9 invalid
{ { 1, 2, 6, 5, 4, 0, 0 }, 6 }, // 001010 = 10
{ { 1, 2, 6, 5, 4, 7, 3 }, 7 }, // 001011 = 11 invalid
{ { 1, 5, 6, 2, 3, 0, 0 }, 6 }, // 001100 = 12
{ { 1, 5, 6, 2, 3, 7, 4 }, 7 }, // 001101 = 13 invalid
{ { 3, 0, 4, 5, 6, 2, 0 }, 6 }, // 001110 = 14
{ { 3, 7, 4, 5, 6, 2, 0 }, 6 }, // 001111 = 15 invalid
{ { 3, 7, 6, 2, 0, 0, 0 }, 4 }, // 010000 = 16
{ { 3, 0, 4, 7, 6, 2, 0 }, 6 }, // 010001 = 17
{ { 1, 5, 4, 0, 0, 0, 0 }, 4 }, // 010010 = 18 invalid
{ { 1, 5, 4, 7, 6, 2, 3 }, 7 }, // 010011 = 19 invalid
{ { 1, 2, 6, 7, 3, 0, 0 }, 6 }, // 010100 = 20
{ { 1, 2, 6, 7, 4, 0, 0 }, 6 }, // 010101 = 21
{ { 2, 6, 7, 3, 0, 4, 5 }, 7 }, // 010110 = 22 invalid
{ { 2, 6, 7, 4, 5, 1, 0 }, 6 }, // 010111 = 23 invalid
{ { 2, 3, 7, 6, 5, 1, 0 }, 6 }, // 011000 = 24
{ { 2, 3, 0, 4, 7, 6, 5 }, 7 }, // 011001 = 25 invalid
{ { 1, 2, 3, 7, 6, 5, 4 }, 7 }, // 011010 = 26 invalid
{ { 1, 2, 3, 0, 0, 0, 0 }, 4 }, // 011011 = 27 invalid
{ { 1, 5, 6, 7, 3, 0, 0 }, 6 }, // 011100 = 28
{ { 1, 5, 6, 7, 4, 0, 0 }, 6 }, // 011101 = 29 invalid
{ { 0, 4, 5, 6, 7, 3, 0 }, 6 }, // 011110 = 30 invalid
{ { 5, 6, 7, 4, 0, 0, 0 }, 4 }, // 011111 = 31 invalid
{ { 5, 6, 7, 4, 0, 0, 0 }, 4 }, // 100000 = 32
{ { 0, 4, 5, 6, 7, 3, 0 }, 6 }, // 100001 = 33
{ { 1, 5, 6, 7, 4, 0, 0 }, 6 }, // 100010 = 34
{ { 1, 5, 6, 7, 3, 0, 0 }, 6 }, // 100011 = 35
{ { 1, 2, 3, 0, 0, 0, 0 }, 4 }, // 100100 = 36 invalid
{ { 1, 2, 3, 7, 6, 5, 4 }, 7 }, // 100101 = 37 invalid
{ { 2, 3, 0, 4, 7, 6, 5 }, 7 }, // 100110 = 38 invalid
{ { 2, 3, 7, 6, 5, 1, 0 }, 6 }, // 100111 = 39 invalid
{ { 2, 6, 7, 4, 5, 1, 0 }, 6 }, // 101000 = 40
{ { 2, 6, 7, 3, 0, 4, 5 }, 7 }, // 101001 = 41 invalid
{ { 1, 2, 6, 7, 4, 0, 0 }, 6 }, // 101010 = 42
{ { 1, 2, 6, 7, 3, 0, 0 }, 6 }, // 101011 = 43 invalid
{ { 1, 5, 4, 7, 6, 2, 3 }, 7 }, // 101100 = 44 invalid
{ { 1, 5, 4, 0, 0, 0, 0 }, 4 }, // 101101 = 45 invalid
{ { 3, 0, 4, 7, 6, 2, 0 }, 6 }, // 101110 = 46 invalid
{ { 3, 7, 6, 2, 0, 0, 0 }, 4 }, // 101111 = 47 invalid
{ { 3, 7, 4, 5, 6, 2, 0 }, 6 }, // 110000 = 48
{ { 3, 0, 4, 5, 6, 2, 0 }, 6 }, // 110001 = 49
{ { 1, 5, 6, 2, 3, 7, 4 }, 7 }, // 110010 = 50 invalid
{ { 1, 5, 6, 2, 3, 0, 0 }, 6 }, // 110011 = 51 invalid
{ { 1, 2, 6, 5, 4, 7, 3 }, 7 }, // 110100 = 52 invalid
{ { 1, 2, 6, 5, 4, 0, 0 }, 6 }, // 110101 = 53 invalid
{ { 2, 6, 5, 1, 0, 0, 0 }, 4 }, // 110110 = 54 invalid
{ { 2, 6, 5, 1, 0, 0, 0 }, 4 }, // 110111 = 55 invalid
{ { 2, 3, 7, 4, 5, 1, 0 }, 6 }, // 111000 = 56
{ { 2, 3, 0, 4, 5, 1, 0 }, 6 }, // 111001 = 57 invalid
{ { 1, 2, 3, 7, 4, 0, 0 }, 6 }, // 111010 = 58 invalid
{ { 1, 2, 3, 0, 0, 0, 0 }, 4 }, // 111011 = 59 invalid
{ { 1, 5, 4, 7, 3, 0, 0 }, 6 }, // 111100 = 60 invalid
{ { 1, 5, 4, 0, 0, 0, 0 }, 4 }, // 111101 = 61 invalid
{ { 0, 4, 7, 3, 0, 0, 0 }, 4 }, // 111110 = 62 invalid
{ { 0, 0, 0, 0, 0, 0, 0 }, 0 }, // 111111 = 63 invalid
};
/*
========================
GetBoxFrontBits
front bits:
bit 0 = neg-X is front facing
bit 1 = neg-Y is front facing
bit 2 = neg-Z is front facing
bit 3 = pos-X is front facing
bit 4 = pos-Y is front facing
bit 5 = pos-Z is front facing
========================
*/
static int GetBoxFrontBits_SSE2( const __m128& b0, const __m128& b1, const __m128& viewOrigin )
{
const __m128 dir0 = _mm_sub_ps( viewOrigin, b0 );
const __m128 dir1 = _mm_sub_ps( b1, viewOrigin );
const __m128 d0 = _mm_cmplt_ps( dir0, _mm_setzero_ps() );
const __m128 d1 = _mm_cmplt_ps( dir1, _mm_setzero_ps() );
int frontBits = _mm_movemask_ps( d0 ) | ( _mm_movemask_ps( d1 ) << 3 );
return frontBits;
}
/*
================================================================================================
idRenderMatrix implementation
================================================================================================
*/
/*
========================
idRenderMatrix::CreateFromOriginAxis
========================
*/
void idRenderMatrix::CreateFromOriginAxis( const idVec3& origin, const idMat3& axis, idRenderMatrix& out )
{
out[0][0] = axis[0][0];
out[0][1] = axis[1][0];
out[0][2] = axis[2][0];
out[0][3] = origin[0];
out[1][0] = axis[0][1];
out[1][1] = axis[1][1];
out[1][2] = axis[2][1];
out[1][3] = origin[1];
out[2][0] = axis[0][2];
out[2][1] = axis[1][2];
out[2][2] = axis[2][2];
out[2][3] = origin[2];
out[3][0] = 0.0f;
out[3][1] = 0.0f;
out[3][2] = 0.0f;
out[3][3] = 1.0f;
}
/*
========================
idRenderMatrix::CreateFromOriginAxisScale
========================
*/
void idRenderMatrix::CreateFromOriginAxisScale( const idVec3& origin, const idMat3& axis, const idVec3& scale, idRenderMatrix& out )
{
out[0][0] = axis[0][0] * scale[0];
out[0][1] = axis[1][0] * scale[1];
out[0][2] = axis[2][0] * scale[2];
out[0][3] = origin[0];
out[1][0] = axis[0][1] * scale[0];
out[1][1] = axis[1][1] * scale[1];
out[1][2] = axis[2][1] * scale[2];
out[1][3] = origin[1];
out[2][0] = axis[0][2] * scale[0];
out[2][1] = axis[1][2] * scale[1];
out[2][2] = axis[2][2] * scale[2];
out[2][3] = origin[2];
out[3][0] = 0.0f;
out[3][1] = 0.0f;
out[3][2] = 0.0f;
out[3][3] = 1.0f;
}
/*
========================
idRenderMatrix::CreateViewMatrix
Our axis looks down positive +X, render matrix looks down -Z.
========================
*/
void idRenderMatrix::CreateViewMatrix( const idVec3& origin, const idMat3& axis, idRenderMatrix& out )
{
out[0][0] = -axis[1][0];
out[0][1] = -axis[1][1];
out[0][2] = -axis[1][2];
out[0][3] = origin[0] * axis[1][0] + origin[1] * axis[1][1] + origin[2] * axis[1][2];
out[1][0] = axis[2][0];
out[1][1] = axis[2][1];
out[1][2] = axis[2][2];
out[1][3] = -( origin[0] * axis[2][0] + origin[1] * axis[2][1] + origin[2] * axis[2][2] );
out[2][0] = -axis[0][0];
out[2][1] = -axis[0][1];
out[2][2] = -axis[0][2];
out[2][3] = origin[0] * axis[0][0] + origin[1] * axis[0][1] + origin[2] * axis[0][2];
out[3][0] = 0.0f;
out[3][1] = 0.0f;
out[3][2] = 0.0f;
out[3][3] = 1.0f;
}
/*
========================
idRenderMatrix::CreateProjectionMatrix
If zFar == 0, an infinite far plane will be used.
========================
*/
void idRenderMatrix::CreateProjectionMatrix( float xMin, float xMax, float yMin, float yMax, float zNear, float zFar, idRenderMatrix& out )
{
const float width = xMax - xMin;
const float height = yMax - yMin;
out[0][0] = 2.0f * zNear / width;
out[0][1] = 0.0f;
out[0][2] = ( xMax + xMin ) / width; // normally 0
out[0][3] = 0.0f;
out[1][0] = 0.0f;
out[1][1] = 2.0f * zNear / height;
out[1][2] = ( yMax + yMin ) / height; // normally 0
out[1][3] = 0.0f;
if( zFar <= zNear )
{
// this is the far-plane-at-infinity formulation
out[2][0] = 0.0f;
out[2][1] = 0.0f;
out[2][2] = -1.0f;
#if defined( CLIP_SPACE_D3D )
// the D3D clip space Z is in range [0,1] instead of [-1,1]
out[2][3] = -zNear;
#else
out[2][3] = -2.0f * zNear;
#endif
}
else
{
out[2][0] = 0.0f;
out[2][1] = 0.0f;
#if defined( CLIP_SPACE_D3D )
// the D3D clip space Z is in range [0,1] instead of [-1,1]
out[2][2] = -( zFar ) / ( zFar - zNear );
out[2][3] = -( zFar * zNear ) / ( zFar - zNear );
#else
out[2][2] = -( zFar + zNear ) / ( zFar - zNear );
out[2][3] = -( 2.0f * zFar * zNear ) / ( zFar - zNear );
#endif
}
out[3][0] = 0.0f;
out[3][1] = 0.0f;
out[3][2] = -1.0f;
out[3][3] = 0.0f;
}
/*
========================
idRenderMatrix::CreateProjectionMatrixFov
xOffset and yOffset should be in the -1 to 1 range for sub-pixel accumulation jitter.
xOffset can also be used for eye separation when rendering stereo.
========================
*/
void idRenderMatrix::CreateProjectionMatrixFov( float xFovDegrees, float yFovDegrees, float zNear, float zFar, float xOffset, float yOffset, idRenderMatrix& out )
{
float xMax = zNear * idMath::Tan( DEG2RAD( xFovDegrees ) * 0.5f );
float xMin = -xMax;
float yMax = zNear * idMath::Tan( DEG2RAD( yFovDegrees ) * 0.5f );
float yMin = -yMax;
xMin += xOffset;
xMax += xOffset;
yMin += yOffset;
yMax += yOffset;
CreateProjectionMatrix( xMin, xMax, yMin, yMax, zNear, zFar, out );
}
/*
========================
idRenderMatrix::OffsetScaleForBounds
Add the offset to the center of the bounds and scale for the width of the bounds.
The result matrix will transform the unit-cube to exactly cover the bounds.
========================
*/
void idRenderMatrix::OffsetScaleForBounds( const idRenderMatrix& src, const idBounds& bounds, idRenderMatrix& out )
{
assert( &src != &out );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
__m128 offset = _mm_mul_ps( _mm_add_ps( b1, b0 ), vector_float_half );
__m128 scale = _mm_mul_ps( _mm_sub_ps( b1, b0 ), vector_float_half );
scale = _mm_or_ps( scale, vector_float_last_one );
__m128 a0 = _mm_loadu_ps( src.m + 0 * 4 );
__m128 a1 = _mm_loadu_ps( src.m + 1 * 4 );
__m128 a2 = _mm_loadu_ps( src.m + 2 * 4 );
__m128 a3 = _mm_loadu_ps( src.m + 3 * 4 );
__m128 d0 = _mm_mul_ps( a0, offset );
__m128 d1 = _mm_mul_ps( a1, offset );
__m128 d2 = _mm_mul_ps( a2, offset );
__m128 d3 = _mm_mul_ps( a3, offset );
__m128 s0 = _mm_unpacklo_ps( d0, d2 ); // a0, c0, a1, c1
__m128 s1 = _mm_unpackhi_ps( d0, d2 ); // a2, c2, a3, c3
__m128 s2 = _mm_unpacklo_ps( d1, d3 ); // b0, d0, b1, d1
__m128 s3 = _mm_unpackhi_ps( d1, d3 ); // b2, d2, b3, d3
__m128 t0 = _mm_unpacklo_ps( s0, s2 ); // a0, b0, c0, d0
__m128 t1 = _mm_unpackhi_ps( s0, s2 ); // a1, b1, c1, d1
__m128 t2 = _mm_unpacklo_ps( s1, s3 ); // a2, b2, c2, d2
t0 = _mm_add_ps( t0, t1 );
t0 = _mm_add_ps( t0, t2 );
__m128 n0 = _mm_and_ps( _mm_splat_ps( t0, 0 ), vector_float_keep_last );
__m128 n1 = _mm_and_ps( _mm_splat_ps( t0, 1 ), vector_float_keep_last );
__m128 n2 = _mm_and_ps( _mm_splat_ps( t0, 2 ), vector_float_keep_last );
__m128 n3 = _mm_and_ps( _mm_splat_ps( t0, 3 ), vector_float_keep_last );
a0 = _mm_madd_ps( a0, scale, n0 );
a1 = _mm_madd_ps( a1, scale, n1 );
a2 = _mm_madd_ps( a2, scale, n2 );
a3 = _mm_madd_ps( a3, scale, n3 );
_mm_storeu_ps( out.m + 0 * 4, a0 );
_mm_storeu_ps( out.m + 1 * 4, a1 );
_mm_storeu_ps( out.m + 2 * 4, a2 );
_mm_storeu_ps( out.m + 3 * 4, a3 );
}
/*
========================
idRenderMatrix::InverseOffsetScaleForBounds
Subtract the offset to the center of the bounds and inverse scale for the width of the bounds.
The result matrix will transform the bounds to exactly cover the unit-cube.
========================
*/
void idRenderMatrix::InverseOffsetScaleForBounds( const idRenderMatrix& src, const idBounds& bounds, idRenderMatrix& out )
{
assert( &src != &out );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
__m128 offset = _mm_mul_ps( _mm_add_ps( b1, b0 ), vector_float_neg_half );
__m128 scale = _mm_mul_ps( _mm_sub_ps( b0, b1 ), vector_float_neg_half );
scale = _mm_max_ps( scale, vector_float_smallest_non_denorm );
__m128 rscale = _mm_rcp32_ps( scale );
offset = _mm_mul_ps( offset, rscale );
__m128 d0 = _mm_and_ps( _mm_splat_ps( offset, 0 ), vector_float_keep_last );
__m128 d1 = _mm_and_ps( _mm_splat_ps( offset, 1 ), vector_float_keep_last );
__m128 d2 = _mm_and_ps( _mm_splat_ps( offset, 2 ), vector_float_keep_last );
__m128 a0 = _mm_loadu_ps( src.m + 0 * 4 );
__m128 a1 = _mm_loadu_ps( src.m + 1 * 4 );
__m128 a2 = _mm_loadu_ps( src.m + 2 * 4 );
__m128 a3 = _mm_loadu_ps( src.m + 3 * 4 );
a0 = _mm_madd_ps( a0, _mm_splat_ps( rscale, 0 ), d0 );
a1 = _mm_madd_ps( a1, _mm_splat_ps( rscale, 1 ), d1 );
a2 = _mm_madd_ps( a2, _mm_splat_ps( rscale, 2 ), d2 );
_mm_storeu_ps( out.m + 0 * 4, a0 );
_mm_storeu_ps( out.m + 1 * 4, a1 );
_mm_storeu_ps( out.m + 2 * 4, a2 );
_mm_storeu_ps( out.m + 3 * 4, a3 );
}
/*
========================
idRenderMatrix::Transpose
========================
*/
void idRenderMatrix::Transpose( const idRenderMatrix& src, idRenderMatrix& out )
{
assert( &src != &out );
const __m128 a0 = _mm_loadu_ps( src.m + 0 * 4 );
const __m128 a1 = _mm_loadu_ps( src.m + 1 * 4 );
const __m128 a2 = _mm_loadu_ps( src.m + 2 * 4 );
const __m128 a3 = _mm_loadu_ps( src.m + 3 * 4 );
const __m128 r0 = _mm_unpacklo_ps( a0, a2 );
const __m128 r1 = _mm_unpackhi_ps( a0, a2 );
const __m128 r2 = _mm_unpacklo_ps( a1, a3 );
const __m128 r3 = _mm_unpackhi_ps( a1, a3 );
const __m128 t0 = _mm_unpacklo_ps( r0, r2 );
const __m128 t1 = _mm_unpackhi_ps( r0, r2 );
const __m128 t2 = _mm_unpacklo_ps( r1, r3 );
const __m128 t3 = _mm_unpackhi_ps( r1, r3 );
_mm_storeu_ps( out.m + 0 * 4, t0 );
_mm_storeu_ps( out.m + 1 * 4, t1 );
_mm_storeu_ps( out.m + 2 * 4, t2 );
_mm_storeu_ps( out.m + 3 * 4, t3 );
}
/*
========================
idRenderMatrix::Multiply
========================
*/
void idRenderMatrix::Multiply( const idRenderMatrix& a, const idRenderMatrix& b, idRenderMatrix& out )
{
__m128 a0 = _mm_loadu_ps( a.m + 0 * 4 );
__m128 a1 = _mm_loadu_ps( a.m + 1 * 4 );
__m128 a2 = _mm_loadu_ps( a.m + 2 * 4 );
__m128 a3 = _mm_loadu_ps( a.m + 3 * 4 );
__m128 b0 = _mm_loadu_ps( b.m + 0 * 4 );
__m128 b1 = _mm_loadu_ps( b.m + 1 * 4 );
__m128 b2 = _mm_loadu_ps( b.m + 2 * 4 );
__m128 b3 = _mm_loadu_ps( b.m + 3 * 4 );
__m128 t0 = _mm_mul_ps( _mm_splat_ps( a0, 0 ), b0 );
__m128 t1 = _mm_mul_ps( _mm_splat_ps( a1, 0 ), b0 );
__m128 t2 = _mm_mul_ps( _mm_splat_ps( a2, 0 ), b0 );
__m128 t3 = _mm_mul_ps( _mm_splat_ps( a3, 0 ), b0 );
t0 = _mm_madd_ps( _mm_splat_ps( a0, 1 ), b1, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( a1, 1 ), b1, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( a2, 1 ), b1, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( a3, 1 ), b1, t3 );
t0 = _mm_madd_ps( _mm_splat_ps( a0, 2 ), b2, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( a1, 2 ), b2, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( a2, 2 ), b2, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( a3, 2 ), b2, t3 );
t0 = _mm_madd_ps( _mm_splat_ps( a0, 3 ), b3, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( a1, 3 ), b3, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( a2, 3 ), b3, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( a3, 3 ), b3, t3 );
_mm_storeu_ps( out.m + 0 * 4, t0 );
_mm_storeu_ps( out.m + 1 * 4, t1 );
_mm_storeu_ps( out.m + 2 * 4, t2 );
_mm_storeu_ps( out.m + 3 * 4, t3 );
}
/*
========================
idRenderMatrix::Inverse
inverse( M ) = ( 1 / determinant( M ) ) * transpose( cofactor( M ) )
This code is based on the code written by C<>dric Lallain, published on "Cell Performance"
(by Mike Acton) and released under the BSD 3-Clause ("BSD New" or "BSD Simplified") license.
https://code.google.com/p/cellperformance-snippets/
Note that large parallel lights can have very small values in the projection matrix,
scaling tens of thousands of world units down to a 0-1 range, so the determinants
can get really, really small.
========================
*/
bool idRenderMatrix::Inverse( const idRenderMatrix& src, idRenderMatrix& out )
{
const __m128 r0 = _mm_loadu_ps( src.m + 0 * 4 );
const __m128 r1 = _mm_loadu_ps( src.m + 1 * 4 );
const __m128 r2 = _mm_loadu_ps( src.m + 2 * 4 );
const __m128 r3 = _mm_loadu_ps( src.m + 3 * 4 );
// rXuY = row X rotated up by Y floats.
const __m128 r0u1 = _mm_perm_ps( r0, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r0u2 = _mm_perm_ps( r0, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r0u3 = _mm_perm_ps( r0, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 r1u1 = _mm_perm_ps( r1, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r1u2 = _mm_perm_ps( r1, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r1u3 = _mm_perm_ps( r1, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 r2u1 = _mm_perm_ps( r2, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r2u2 = _mm_perm_ps( r2, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r2u3 = _mm_perm_ps( r2, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 r3u1 = _mm_perm_ps( r3, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r3u2 = _mm_perm_ps( r3, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r3u3 = _mm_perm_ps( r3, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 m_r2u2_r3u3 = _mm_mul_ps( r2u2, r3u3 );
const __m128 m_r1u1_r2u2_r3u3 = _mm_mul_ps( r1u1, m_r2u2_r3u3 );
const __m128 m_r2u3_r3u1 = _mm_mul_ps( r2u3, r3u1 );
const __m128 a_m_r1u2_r2u3_r3u1_m_r1u1_r2u2_r3u3 = _mm_madd_ps( r1u2, m_r2u3_r3u1, m_r1u1_r2u2_r3u3 );
const __m128 m_r2u1_r3u2 = _mm_perm_ps( m_r2u2_r3u3, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 pos_part_det3x3_r0 = _mm_madd_ps( r1u3, m_r2u1_r3u2, a_m_r1u2_r2u3_r3u1_m_r1u1_r2u2_r3u3 );
const __m128 m_r2u3_r3u2 = _mm_mul_ps( r2u3, r3u2 );
const __m128 m_r1u1_r2u3_r3u2 = _mm_mul_ps( r1u1, m_r2u3_r3u2 );
const __m128 m_r2u1_r3u3 = _mm_perm_ps( m_r2u3_r3u1, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 a_m_r1u2_r2u1_r3u3_m_r1u1_r2u3_r3u2 = _mm_madd_ps( r1u2, m_r2u1_r3u3, m_r1u1_r2u3_r3u2 );
const __m128 m_r2u2_r3u1 = _mm_perm_ps( m_r2u3_r3u2, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 neg_part_det3x3_r0 = _mm_madd_ps( r1u3, m_r2u2_r3u1, a_m_r1u2_r2u1_r3u3_m_r1u1_r2u3_r3u2 );
const __m128 det3x3_r0 = _mm_sub_ps( pos_part_det3x3_r0, neg_part_det3x3_r0 );
const __m128 m_r0u1_r2u2_r3u3 = _mm_mul_ps( r0u1, m_r2u2_r3u3 );
const __m128 a_m_r0u2_r2u3_r3u1_m_r0u1_r2u2_r3u3 = _mm_madd_ps( r0u2, m_r2u3_r3u1, m_r0u1_r2u2_r3u3 );
const __m128 pos_part_det3x3_r1 = _mm_madd_ps( r0u3, m_r2u1_r3u2, a_m_r0u2_r2u3_r3u1_m_r0u1_r2u2_r3u3 );
const __m128 m_r0u1_r2u3_r3u2 = _mm_mul_ps( r0u1, m_r2u3_r3u2 );
const __m128 a_m_r0u2_r2u1_r3u3_m_r0u1_r2u3_r3u2 = _mm_madd_ps( r0u2, m_r2u1_r3u3, m_r0u1_r2u3_r3u2 );
const __m128 neg_part_det3x3_r1 = _mm_madd_ps( r0u3, m_r2u2_r3u1, a_m_r0u2_r2u1_r3u3_m_r0u1_r2u3_r3u2 );
const __m128 det3x3_r1 = _mm_sub_ps( pos_part_det3x3_r1, neg_part_det3x3_r1 );
const __m128 m_r0u1_r1u2 = _mm_mul_ps( r0u1, r1u2 );
const __m128 m_r0u1_r1u2_r2u3 = _mm_mul_ps( m_r0u1_r1u2, r2u3 );
const __m128 m_r0u2_r1u3 = _mm_perm_ps( m_r0u1_r1u2, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 a_m_r0u2_r1u3_r2u1_m_r0u1_r1u2_r2u3 = _mm_madd_ps( m_r0u2_r1u3, r2u1, m_r0u1_r1u2_r2u3 );
const __m128 m_r0u3_r1u1 = _mm_mul_ps( r0u3, r1u1 );
const __m128 pos_part_det3x3_r3 = _mm_madd_ps( m_r0u3_r1u1, r2u2, a_m_r0u2_r1u3_r2u1_m_r0u1_r1u2_r2u3 );
const __m128 m_r0u1_r1u3 = _mm_perm_ps( m_r0u3_r1u1, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 m_r0u1_r1u3_r2u2 = _mm_mul_ps( m_r0u1_r1u3, r2u2 );
const __m128 m_r0u2_r1u1 = _mm_mul_ps( r0u2, r1u1 );
const __m128 a_m_r0u2_r1u1_r2u3_m_r0u1_r1u3_r2u2 = _mm_madd_ps( m_r0u2_r1u1, r2u3, m_r0u1_r1u3_r2u2 );
const __m128 m_r0u3_r1u2 = _mm_perm_ps( m_r0u2_r1u1, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 neg_part_det3x3_r3 = _mm_madd_ps( m_r0u3_r1u2, r2u1, a_m_r0u2_r1u1_r2u3_m_r0u1_r1u3_r2u2 );
const __m128 det3x3_r3 = _mm_sub_ps( pos_part_det3x3_r3, neg_part_det3x3_r3 );
const __m128 m_r0u1_r1u2_r3u3 = _mm_mul_ps( m_r0u1_r1u2, r3u3 );
const __m128 a_m_r0u2_r1u3_r3u1_m_r0u1_r1u2_r3u3 = _mm_madd_ps( m_r0u2_r1u3, r3u1, m_r0u1_r1u2_r3u3 );
const __m128 pos_part_det3x3_r2 = _mm_madd_ps( m_r0u3_r1u1, r3u2, a_m_r0u2_r1u3_r3u1_m_r0u1_r1u2_r3u3 );
const __m128 m_r0u1_r1u3_r3u2 = _mm_mul_ps( m_r0u1_r1u3, r3u2 );
const __m128 a_m_r0u2_r1u1_r3u3_m_r0u1_r1u3_r3u2 = _mm_madd_ps( m_r0u2_r1u1, r3u3, m_r0u1_r1u3_r3u2 );
const __m128 neg_part_det3x3_r2 = _mm_madd_ps( m_r0u3_r1u2, r3u1, a_m_r0u2_r1u1_r3u3_m_r0u1_r1u3_r3u2 );
const __m128 det3x3_r2 = _mm_sub_ps( pos_part_det3x3_r2, neg_part_det3x3_r2 );
const __m128 c_zero = _mm_setzero_ps();
const __m128 c_mask = _mm_cmpeq_ps( c_zero, c_zero );
const __m128 c_signmask = _mm_castsi128_ps( _mm_slli_epi32( _mm_castps_si128( c_mask ), 31 ) );
const __m128 c_znzn = _mm_unpacklo_ps( c_zero, c_signmask );
const __m128 c_nznz = _mm_unpacklo_ps( c_signmask, c_zero );
const __m128 cofactor_r0 = _mm_xor_ps( det3x3_r0, c_znzn );
const __m128 cofactor_r1 = _mm_xor_ps( det3x3_r1, c_nznz );
const __m128 cofactor_r2 = _mm_xor_ps( det3x3_r2, c_znzn );
const __m128 cofactor_r3 = _mm_xor_ps( det3x3_r3, c_nznz );
const __m128 dot0 = _mm_mul_ps( r0, cofactor_r0 );
const __m128 dot1 = _mm_add_ps( dot0, _mm_perm_ps( dot0, _MM_SHUFFLE( 2, 1, 0, 3 ) ) );
const __m128 det = _mm_add_ps( dot1, _mm_perm_ps( dot1, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
const __m128 absDet = _mm_andnot_ps( c_signmask, det );
if( _mm_movemask_ps( _mm_cmplt_ps( absDet, vector_float_inverse_epsilon ) ) & 15 )
{
return false;
}
const __m128 rcpDet = _mm_rcp32_ps( det );
const __m128 hi_part_r0_r2 = _mm_unpacklo_ps( cofactor_r0, cofactor_r2 );
const __m128 lo_part_r0_r2 = _mm_unpackhi_ps( cofactor_r0, cofactor_r2 );
const __m128 hi_part_r1_r3 = _mm_unpacklo_ps( cofactor_r1, cofactor_r3 );
const __m128 lo_part_r1_r3 = _mm_unpackhi_ps( cofactor_r1, cofactor_r3 );
const __m128 adjoint_r0 = _mm_unpacklo_ps( hi_part_r0_r2, hi_part_r1_r3 );
const __m128 adjoint_r1 = _mm_unpackhi_ps( hi_part_r0_r2, hi_part_r1_r3 );
const __m128 adjoint_r2 = _mm_unpacklo_ps( lo_part_r0_r2, lo_part_r1_r3 );
const __m128 adjoint_r3 = _mm_unpackhi_ps( lo_part_r0_r2, lo_part_r1_r3 );
_mm_storeu_ps( out.m + 0 * 4, _mm_mul_ps( adjoint_r0, rcpDet ) );
_mm_storeu_ps( out.m + 1 * 4, _mm_mul_ps( adjoint_r1, rcpDet ) );
_mm_storeu_ps( out.m + 2 * 4, _mm_mul_ps( adjoint_r2, rcpDet ) );
_mm_storeu_ps( out.m + 3 * 4, _mm_mul_ps( adjoint_r3, rcpDet ) );
return true;
}
/*
========================
idRenderMatrix::InverseByTranspose
========================
*/
void idRenderMatrix::InverseByTranspose( const idRenderMatrix& src, idRenderMatrix& out )
{
assert( &src != &out );
assert( src.IsAffineTransform( 0.01f ) );
out[0][0] = src[0][0];
out[1][0] = src[0][1];
out[2][0] = src[0][2];
out[3][0] = 0.0f;
out[0][1] = src[1][0];
out[1][1] = src[1][1];
out[2][1] = src[1][2];
out[3][1] = 0.0f;
out[0][2] = src[2][0];
out[1][2] = src[2][1];
out[2][2] = src[2][2];
out[3][2] = 0.0f;
out[0][3] = -( src[0][0] * src[0][3] + src[1][0] * src[1][3] + src[2][0] * src[2][3] );
out[1][3] = -( src[0][1] * src[0][3] + src[1][1] * src[1][3] + src[2][1] * src[2][3] );
out[2][3] = -( src[0][2] * src[0][3] + src[1][2] * src[1][3] + src[2][2] * src[2][3] );
out[3][3] = 1.0f;
}
/*
========================
idRenderMatrix::InverseByDoubles
This should never be used at run-time.
This is only for tools where more precision is needed.
========================
*/
bool idRenderMatrix::InverseByDoubles( const idRenderMatrix& src, idRenderMatrix& out )
{
const int FRL = 4;
// 84+4+16 = 104 multiplications
// 1 division
// 2x2 sub-determinants required to calculate 4x4 determinant
const double det2_01_01 = ( double )src.m[0 * FRL + 0] * ( double )src.m[1 * FRL + 1] - ( double )src.m[0 * FRL + 1] * ( double )src.m[1 * FRL + 0];
const double det2_01_02 = ( double )src.m[0 * FRL + 0] * ( double )src.m[1 * FRL + 2] - ( double )src.m[0 * FRL + 2] * ( double )src.m[1 * FRL + 0];
const double det2_01_03 = ( double )src.m[0 * FRL + 0] * ( double )src.m[1 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[1 * FRL + 0];
const double det2_01_12 = ( double )src.m[0 * FRL + 1] * ( double )src.m[1 * FRL + 2] - ( double )src.m[0 * FRL + 2] * ( double )src.m[1 * FRL + 1];
const double det2_01_13 = ( double )src.m[0 * FRL + 1] * ( double )src.m[1 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[1 * FRL + 1];
const double det2_01_23 = ( double )src.m[0 * FRL + 2] * ( double )src.m[1 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[1 * FRL + 2];
// 3x3 sub-determinants required to calculate 4x4 determinant
const double det3_201_012 = ( double )src.m[2 * FRL + 0] * det2_01_12 - ( double )src.m[2 * FRL + 1] * det2_01_02 + ( double )src.m[2 * FRL + 2] * det2_01_01;
const double det3_201_013 = ( double )src.m[2 * FRL + 0] * det2_01_13 - ( double )src.m[2 * FRL + 1] * det2_01_03 + ( double )src.m[2 * FRL + 3] * det2_01_01;
const double det3_201_023 = ( double )src.m[2 * FRL + 0] * det2_01_23 - ( double )src.m[2 * FRL + 2] * det2_01_03 + ( double )src.m[2 * FRL + 3] * det2_01_02;
const double det3_201_123 = ( double )src.m[2 * FRL + 1] * det2_01_23 - ( double )src.m[2 * FRL + 2] * det2_01_13 + ( double )src.m[2 * FRL + 3] * det2_01_12;
const double det = ( - det3_201_123 * ( double )src.m[3 * FRL + 0] + det3_201_023 * ( double )src.m[3 * FRL + 1] - det3_201_013 * ( double )src.m[3 * FRL + 2] + det3_201_012 * ( double )src.m[3 * FRL + 3] );
const double rcpDet = 1.0f / det;
// remaining 2x2 sub-determinants
const double det2_03_01 = ( double )src.m[0 * FRL + 0] * ( double )src.m[3 * FRL + 1] - ( double )src.m[0 * FRL + 1] * ( double )src.m[3 * FRL + 0];
const double det2_03_02 = ( double )src.m[0 * FRL + 0] * ( double )src.m[3 * FRL + 2] - ( double )src.m[0 * FRL + 2] * ( double )src.m[3 * FRL + 0];
const double det2_03_03 = ( double )src.m[0 * FRL + 0] * ( double )src.m[3 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[3 * FRL + 0];
const double det2_03_12 = ( double )src.m[0 * FRL + 1] * ( double )src.m[3 * FRL + 2] - ( double )src.m[0 * FRL + 2] * ( double )src.m[3 * FRL + 1];
const double det2_03_13 = ( double )src.m[0 * FRL + 1] * ( double )src.m[3 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[3 * FRL + 1];
const double det2_03_23 = ( double )src.m[0 * FRL + 2] * ( double )src.m[3 * FRL + 3] - ( double )src.m[0 * FRL + 3] * ( double )src.m[3 * FRL + 2];
const double det2_13_01 = ( double )src.m[1 * FRL + 0] * ( double )src.m[3 * FRL + 1] - ( double )src.m[1 * FRL + 1] * ( double )src.m[3 * FRL + 0];
const double det2_13_02 = ( double )src.m[1 * FRL + 0] * ( double )src.m[3 * FRL + 2] - ( double )src.m[1 * FRL + 2] * ( double )src.m[3 * FRL + 0];
const double det2_13_03 = ( double )src.m[1 * FRL + 0] * ( double )src.m[3 * FRL + 3] - ( double )src.m[1 * FRL + 3] * ( double )src.m[3 * FRL + 0];
const double det2_13_12 = ( double )src.m[1 * FRL + 1] * ( double )src.m[3 * FRL + 2] - ( double )src.m[1 * FRL + 2] * ( double )src.m[3 * FRL + 1];
const double det2_13_13 = ( double )src.m[1 * FRL + 1] * ( double )src.m[3 * FRL + 3] - ( double )src.m[1 * FRL + 3] * ( double )src.m[3 * FRL + 1];
const double det2_13_23 = ( double )src.m[1 * FRL + 2] * ( double )src.m[3 * FRL + 3] - ( double )src.m[1 * FRL + 3] * ( double )src.m[3 * FRL + 2];
// remaining 3x3 sub-determinants
const double det3_203_012 = ( double )src.m[2 * FRL + 0] * det2_03_12 - ( double )src.m[2 * FRL + 1] * det2_03_02 + ( double )src.m[2 * FRL + 2] * det2_03_01;
const double det3_203_013 = ( double )src.m[2 * FRL + 0] * det2_03_13 - ( double )src.m[2 * FRL + 1] * det2_03_03 + ( double )src.m[2 * FRL + 3] * det2_03_01;
const double det3_203_023 = ( double )src.m[2 * FRL + 0] * det2_03_23 - ( double )src.m[2 * FRL + 2] * det2_03_03 + ( double )src.m[2 * FRL + 3] * det2_03_02;
const double det3_203_123 = ( double )src.m[2 * FRL + 1] * det2_03_23 - ( double )src.m[2 * FRL + 2] * det2_03_13 + ( double )src.m[2 * FRL + 3] * det2_03_12;
const double det3_213_012 = ( double )src.m[2 * FRL + 0] * det2_13_12 - ( double )src.m[2 * FRL + 1] * det2_13_02 + ( double )src.m[2 * FRL + 2] * det2_13_01;
const double det3_213_013 = ( double )src.m[2 * FRL + 0] * det2_13_13 - ( double )src.m[2 * FRL + 1] * det2_13_03 + ( double )src.m[2 * FRL + 3] * det2_13_01;
const double det3_213_023 = ( double )src.m[2 * FRL + 0] * det2_13_23 - ( double )src.m[2 * FRL + 2] * det2_13_03 + ( double )src.m[2 * FRL + 3] * det2_13_02;
const double det3_213_123 = ( double )src.m[2 * FRL + 1] * det2_13_23 - ( double )src.m[2 * FRL + 2] * det2_13_13 + ( double )src.m[2 * FRL + 3] * det2_13_12;
const double det3_301_012 = ( double )src.m[3 * FRL + 0] * det2_01_12 - ( double )src.m[3 * FRL + 1] * det2_01_02 + ( double )src.m[3 * FRL + 2] * det2_01_01;
const double det3_301_013 = ( double )src.m[3 * FRL + 0] * det2_01_13 - ( double )src.m[3 * FRL + 1] * det2_01_03 + ( double )src.m[3 * FRL + 3] * det2_01_01;
const double det3_301_023 = ( double )src.m[3 * FRL + 0] * det2_01_23 - ( double )src.m[3 * FRL + 2] * det2_01_03 + ( double )src.m[3 * FRL + 3] * det2_01_02;
const double det3_301_123 = ( double )src.m[3 * FRL + 1] * det2_01_23 - ( double )src.m[3 * FRL + 2] * det2_01_13 + ( double )src.m[3 * FRL + 3] * det2_01_12;
out.m[0 * FRL + 0] = ( float )( - det3_213_123 * rcpDet );
out.m[1 * FRL + 0] = ( float )( + det3_213_023 * rcpDet );
out.m[2 * FRL + 0] = ( float )( - det3_213_013 * rcpDet );
out.m[3 * FRL + 0] = ( float )( + det3_213_012 * rcpDet );
out.m[0 * FRL + 1] = ( float )( + det3_203_123 * rcpDet );
out.m[1 * FRL + 1] = ( float )( - det3_203_023 * rcpDet );
out.m[2 * FRL + 1] = ( float )( + det3_203_013 * rcpDet );
out.m[3 * FRL + 1] = ( float )( - det3_203_012 * rcpDet );
out.m[0 * FRL + 2] = ( float )( + det3_301_123 * rcpDet );
out.m[1 * FRL + 2] = ( float )( - det3_301_023 * rcpDet );
out.m[2 * FRL + 2] = ( float )( + det3_301_013 * rcpDet );
out.m[3 * FRL + 2] = ( float )( - det3_301_012 * rcpDet );
out.m[0 * FRL + 3] = ( float )( - det3_201_123 * rcpDet );
out.m[1 * FRL + 3] = ( float )( + det3_201_023 * rcpDet );
out.m[2 * FRL + 3] = ( float )( - det3_201_013 * rcpDet );
out.m[3 * FRL + 3] = ( float )( + det3_201_012 * rcpDet );
return true;
}
/*
========================
DeterminantIsNegative
========================
*/
void DeterminantIsNegative( bool& negativeDeterminant, const __m128& r0, const __m128& r1, const __m128& r2, const __m128& r3 )
{
const __m128 r1u1 = _mm_perm_ps( r1, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r1u2 = _mm_perm_ps( r1, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r1u3 = _mm_perm_ps( r1, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 r2u2 = _mm_perm_ps( r2, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r2u3 = _mm_perm_ps( r2, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 r3u1 = _mm_perm_ps( r3, _MM_SHUFFLE( 2, 1, 0, 3 ) );
const __m128 r3u2 = _mm_perm_ps( r3, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 r3u3 = _mm_perm_ps( r3, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 m_r2u2_r3u3 = _mm_mul_ps( r2u2, r3u3 );
const __m128 m_r1u1_r2u2_r3u3 = _mm_mul_ps( r1u1, m_r2u2_r3u3 );
const __m128 m_r2u3_r3u1 = _mm_mul_ps( r2u3, r3u1 );
const __m128 a_m_r1u2_r2u3_r3u1_m_r1u1_r2u2_r3u3 = _mm_madd_ps( r1u2, m_r2u3_r3u1, m_r1u1_r2u2_r3u3 );
const __m128 m_r2u1_r3u2 = _mm_perm_ps( m_r2u2_r3u3, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 pos_part_det3x3_r0 = _mm_madd_ps( r1u3, m_r2u1_r3u2, a_m_r1u2_r2u3_r3u1_m_r1u1_r2u2_r3u3 );
const __m128 m_r2u3_r3u2 = _mm_mul_ps( r2u3, r3u2 );
const __m128 m_r1u1_r2u3_r3u2 = _mm_mul_ps( r1u1, m_r2u3_r3u2 );
const __m128 m_r2u1_r3u3 = _mm_perm_ps( m_r2u3_r3u1, _MM_SHUFFLE( 1, 0, 3, 2 ) );
const __m128 a_m_r1u2_r2u1_r3u3_m_r1u1_r2u3_r3u2 = _mm_madd_ps( r1u2, m_r2u1_r3u3, m_r1u1_r2u3_r3u2 );
const __m128 m_r2u2_r3u1 = _mm_perm_ps( m_r2u3_r3u2, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 neg_part_det3x3_r0 = _mm_madd_ps( r1u3, m_r2u2_r3u1, a_m_r1u2_r2u1_r3u3_m_r1u1_r2u3_r3u2 );
const __m128 det3x3_r0 = _mm_sub_ps( pos_part_det3x3_r0, neg_part_det3x3_r0 );
const __m128 c_zero = _mm_setzero_ps();
const __m128 c_mask = _mm_cmpeq_ps( c_zero, c_zero );
const __m128 c_signmask = _mm_castsi128_ps( _mm_slli_epi32( _mm_castps_si128( c_mask ), 31 ) );
const __m128 c_znzn = _mm_unpacklo_ps( c_zero, c_signmask );
const __m128 cofactor_r0 = _mm_xor_ps( det3x3_r0, c_znzn );
const __m128 dot0 = _mm_mul_ps( r0, cofactor_r0 );
const __m128 dot1 = _mm_add_ps( dot0, _mm_perm_ps( dot0, _MM_SHUFFLE( 2, 1, 0, 3 ) ) );
const __m128 det = _mm_add_ps( dot1, _mm_perm_ps( dot1, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
const __m128 result = _mm_cmpgt_ps( c_zero, det );
negativeDeterminant = _mm_movemask_ps( result ) & 1;
}
/*
========================
idRenderMatrix::CopyMatrix
========================
*/
void idRenderMatrix::CopyMatrix( const idRenderMatrix& matrix, idVec4& row0, idVec4& row1, idVec4& row2, idVec4& row3 )
{
assert_16_byte_aligned( row0.ToFloatPtr() );
assert_16_byte_aligned( row1.ToFloatPtr() );
assert_16_byte_aligned( row2.ToFloatPtr() );
assert_16_byte_aligned( row3.ToFloatPtr() );
const __m128 r0 = _mm_loadu_ps( matrix.m + 0 * 4 );
const __m128 r1 = _mm_loadu_ps( matrix.m + 1 * 4 );
const __m128 r2 = _mm_loadu_ps( matrix.m + 2 * 4 );
const __m128 r3 = _mm_loadu_ps( matrix.m + 3 * 4 );
_mm_store_ps( row0.ToFloatPtr(), r0 );
_mm_store_ps( row1.ToFloatPtr(), r1 );
_mm_store_ps( row2.ToFloatPtr(), r2 );
_mm_store_ps( row3.ToFloatPtr(), r3 );
}
/*
========================
idRenderMatrix::SetMVP
========================
*/
void idRenderMatrix::SetMVP( const idRenderMatrix& mvp, idVec4& row0, idVec4& row1, idVec4& row2, idVec4& row3, bool& negativeDeterminant )
{
assert_16_byte_aligned( row0.ToFloatPtr() );
assert_16_byte_aligned( row1.ToFloatPtr() );
assert_16_byte_aligned( row2.ToFloatPtr() );
assert_16_byte_aligned( row3.ToFloatPtr() );
const __m128 r0 = _mm_loadu_ps( mvp.m + 0 * 4 );
const __m128 r1 = _mm_loadu_ps( mvp.m + 1 * 4 );
const __m128 r2 = _mm_loadu_ps( mvp.m + 2 * 4 );
const __m128 r3 = _mm_loadu_ps( mvp.m + 3 * 4 );
_mm_store_ps( row0.ToFloatPtr(), r0 );
_mm_store_ps( row1.ToFloatPtr(), r1 );
_mm_store_ps( row2.ToFloatPtr(), r2 );
_mm_store_ps( row3.ToFloatPtr(), r3 );
DeterminantIsNegative( negativeDeterminant, r0, r1, r2, r3 );
}
/*
========================
idRenderMatrix::SetMVPForBounds
========================
*/
void idRenderMatrix::SetMVPForBounds( const idRenderMatrix& mvp, const idBounds& bounds, idVec4& row0, idVec4& row1, idVec4& row2, idVec4& row3, bool& negativeDeterminant )
{
assert_16_byte_aligned( row0.ToFloatPtr() );
assert_16_byte_aligned( row1.ToFloatPtr() );
assert_16_byte_aligned( row2.ToFloatPtr() );
assert_16_byte_aligned( row3.ToFloatPtr() );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
__m128 offset = _mm_mul_ps( _mm_add_ps( b1, b0 ), vector_float_half );
__m128 scale = _mm_mul_ps( _mm_sub_ps( b1, b0 ), vector_float_half );
scale = _mm_or_ps( scale, vector_float_last_one );
__m128 r0 = _mm_loadu_ps( mvp.m + 0 * 4 );
__m128 r1 = _mm_loadu_ps( mvp.m + 1 * 4 );
__m128 r2 = _mm_loadu_ps( mvp.m + 2 * 4 );
__m128 r3 = _mm_loadu_ps( mvp.m + 3 * 4 );
__m128 d0 = _mm_mul_ps( r0, offset );
__m128 d1 = _mm_mul_ps( r1, offset );
__m128 d2 = _mm_mul_ps( r2, offset );
__m128 d3 = _mm_mul_ps( r3, offset );
__m128 s0 = _mm_unpacklo_ps( d0, d2 ); // a0, c0, a1, c1
__m128 s1 = _mm_unpackhi_ps( d0, d2 ); // a2, c2, a3, c3
__m128 s2 = _mm_unpacklo_ps( d1, d3 ); // b0, d0, b1, d1
__m128 s3 = _mm_unpackhi_ps( d1, d3 ); // b2, d2, b3, d3
__m128 t0 = _mm_unpacklo_ps( s0, s2 ); // a0, b0, c0, d0
__m128 t1 = _mm_unpackhi_ps( s0, s2 ); // a1, b1, c1, d1
__m128 t2 = _mm_unpacklo_ps( s1, s3 ); // a2, b2, c2, d2
t0 = _mm_add_ps( t0, t1 );
t0 = _mm_add_ps( t0, t2 );
__m128 n0 = _mm_and_ps( _mm_splat_ps( t0, 0 ), vector_float_keep_last );
__m128 n1 = _mm_and_ps( _mm_splat_ps( t0, 1 ), vector_float_keep_last );
__m128 n2 = _mm_and_ps( _mm_splat_ps( t0, 2 ), vector_float_keep_last );
__m128 n3 = _mm_and_ps( _mm_splat_ps( t0, 3 ), vector_float_keep_last );
r0 = _mm_madd_ps( r0, scale, n0 );
r1 = _mm_madd_ps( r1, scale, n1 );
r2 = _mm_madd_ps( r2, scale, n2 );
r3 = _mm_madd_ps( r3, scale, n3 );
_mm_store_ps( row0.ToFloatPtr(), r0 );
_mm_store_ps( row1.ToFloatPtr(), r1 );
_mm_store_ps( row2.ToFloatPtr(), r2 );
_mm_store_ps( row3.ToFloatPtr(), r3 );
DeterminantIsNegative( negativeDeterminant, r0, r1, r2, r3 );
}
/*
========================
idRenderMatrix::SetMVPForInverseProject
========================
*/
void idRenderMatrix::SetMVPForInverseProject( const idRenderMatrix& mvp, const idRenderMatrix& inverseProject, idVec4& row0, idVec4& row1, idVec4& row2, idVec4& row3, bool& negativeDeterminant )
{
assert_16_byte_aligned( row0.ToFloatPtr() );
assert_16_byte_aligned( row1.ToFloatPtr() );
assert_16_byte_aligned( row2.ToFloatPtr() );
assert_16_byte_aligned( row3.ToFloatPtr() );
__m128 r0 = _mm_loadu_ps( mvp.m + 0 * 4 );
__m128 r1 = _mm_loadu_ps( mvp.m + 1 * 4 );
__m128 r2 = _mm_loadu_ps( mvp.m + 2 * 4 );
__m128 r3 = _mm_loadu_ps( mvp.m + 3 * 4 );
__m128 p0 = _mm_loadu_ps( inverseProject.m + 0 * 4 );
__m128 p1 = _mm_loadu_ps( inverseProject.m + 1 * 4 );
__m128 p2 = _mm_loadu_ps( inverseProject.m + 2 * 4 );
__m128 p3 = _mm_loadu_ps( inverseProject.m + 3 * 4 );
__m128 t0 = _mm_mul_ps( _mm_splat_ps( r0, 0 ), p0 );
__m128 t1 = _mm_mul_ps( _mm_splat_ps( r1, 0 ), p0 );
__m128 t2 = _mm_mul_ps( _mm_splat_ps( r2, 0 ), p0 );
__m128 t3 = _mm_mul_ps( _mm_splat_ps( r3, 0 ), p0 );
t0 = _mm_madd_ps( _mm_splat_ps( r0, 1 ), p1, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( r1, 1 ), p1, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( r2, 1 ), p1, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( r3, 1 ), p1, t3 );
t0 = _mm_madd_ps( _mm_splat_ps( r0, 2 ), p2, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( r1, 2 ), p2, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( r2, 2 ), p2, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( r3, 2 ), p2, t3 );
t0 = _mm_madd_ps( _mm_splat_ps( r0, 3 ), p3, t0 );
t1 = _mm_madd_ps( _mm_splat_ps( r1, 3 ), p3, t1 );
t2 = _mm_madd_ps( _mm_splat_ps( r2, 3 ), p3, t2 );
t3 = _mm_madd_ps( _mm_splat_ps( r3, 3 ), p3, t3 );
_mm_store_ps( row0.ToFloatPtr(), t0 );
_mm_store_ps( row1.ToFloatPtr(), t1 );
_mm_store_ps( row2.ToFloatPtr(), t2 );
_mm_store_ps( row3.ToFloatPtr(), t3 );
DeterminantIsNegative( negativeDeterminant, t0, t1, t2, t3 );
}
/*
========================
idRenderMatrix::CullPointToMVPbits
Returns true if the point transformed by the given Model View Projection (MVP) matrix is
outside the clip space.
Normally the clip space extends from -1.0 to 1.0 on each axis, but by setting 'zeroToOne'
to true, the clip space will extend from 0.0 to 1.0 on each axis for a light projection matrix.
========================
*/
bool idRenderMatrix::CullPointToMVPbits( const idRenderMatrix& mvp, const idVec3& p, byte* outBits, bool zeroToOne )
{
idVec4 c;
for( int i = 0; i < 4; i++ )
{
c[i] = p[0] * mvp[i][0] + p[1] * mvp[i][1] + p[2] * mvp[i][2] + mvp[i][3];
}
const float minW = zeroToOne ? 0.0f : -c[3];
const float maxW = c[3];
#if defined( CLIP_SPACE_D3D ) // the D3D clip space Z is in the range [0,1] so always compare Z vs zero whether 'zeroToOne' is true or false
const float minZ = 0.0f;
#else
const float minZ = minW;
#endif
int bits = 0;
if( c[0] > minW )
{
bits |= ( 1 << 0 );
}
if( c[0] < maxW )
{
bits |= ( 1 << 1 );
}
if( c[1] > minW )
{
bits |= ( 1 << 2 );
}
if( c[1] < maxW )
{
bits |= ( 1 << 3 );
}
if( c[2] > minZ )
{
bits |= ( 1 << 4 ); // NOTE: using minZ
}
if( c[2] < maxW )
{
bits |= ( 1 << 5 );
}
// store out a bit set for each side where the point is outside the clip space
*outBits = ( byte )( bits ^ 63 );
// if any bits weren't set, the point is completely off one side of the frustum
return ( bits != 63 );
}
/*
========================
idRenderMatrix::CullBoundsToMVPbits
Returns true if nothing contained in the bounds is transformed by the given
Model View Projection (MVP) matrix to anything inside the clip space.
Normally the clip space extends from -1.0 to 1.0 on each axis, but by setting 'zeroToOne'
to true, the clip space will extend from 0.0 to 1.0 on each axis for a light projection matrix.
When all the corners of the bounding box are behind one of the six frustum planes, the box is
culled. This is conservative, because some boxes may "cross corners" and can be in front of a
frustum plane, but only while also being behind another one.
========================
*/
bool idRenderMatrix::CullBoundsToMVPbits( const idRenderMatrix& mvp, const idBounds& bounds, byte* outBits, bool zeroToOne )
{
__m128 mvp0 = _mm_loadu_ps( mvp[0] );
__m128 mvp1 = _mm_loadu_ps( mvp[1] );
__m128 mvp2 = _mm_loadu_ps( mvp[2] );
__m128 mvp3 = _mm_loadu_ps( mvp[3] );
__m128 minMul = zeroToOne ? vector_float_zero : vector_float_neg_one;
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
// compute four partial X,Y,Z,W values
__m128 parx = _mm_splat_ps( mvp0, 3 );
__m128 pary = _mm_splat_ps( mvp1, 3 );
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parx = _mm_madd_ps( vx, _mm_splat_ps( mvp0, 0 ), parx );
pary = _mm_madd_ps( vx, _mm_splat_ps( mvp1, 0 ), pary );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parx = _mm_madd_ps( vy, _mm_splat_ps( mvp0, 1 ), parx );
pary = _mm_madd_ps( vy, _mm_splat_ps( mvp1, 1 ), pary );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
// compute full X,Y,Z,W values
__m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
__m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
__m128 x0 = _mm_madd_ps( vz0, mvp0Z, parx );
__m128 y0 = _mm_madd_ps( vz0, mvp1Z, pary );
__m128 z0 = _mm_madd_ps( vz0, mvp2Z, parz );
__m128 w0 = _mm_madd_ps( vz0, mvp3Z, parw );
__m128 x1 = _mm_madd_ps( vz1, mvp0Z, parx );
__m128 y1 = _mm_madd_ps( vz1, mvp1Z, pary );
__m128 z1 = _mm_madd_ps( vz1, mvp2Z, parz );
__m128 w1 = _mm_madd_ps( vz1, mvp3Z, parw );
__m128 maxW0 = w0;
__m128 maxW1 = w1;
__m128 minW0 = _mm_mul_ps( w0, minMul );
__m128 minW1 = _mm_mul_ps( w1, minMul );
#if defined( CLIP_SPACE_D3D ) // the D3D clip space Z is in the range [0,1] so always compare Z vs zero whether 'zeroToOne' is true or false
__m128 minZ0 = vector_float_zero;
__m128 minZ1 = vector_float_zero;
#else
__m128 minZ0 = minW0;
__m128 minZ1 = minW1;
#endif
__m128 cullBits0 = _mm_cmpgt_ps( x0, minW0 );
__m128 cullBits1 = _mm_cmpgt_ps( maxW0, x0 );
__m128 cullBits2 = _mm_cmpgt_ps( y0, minW0 );
__m128 cullBits3 = _mm_cmpgt_ps( maxW0, y0 );
__m128 cullBits4 = _mm_cmpgt_ps( z0, minZ0 ); // NOTE: using minZ0
__m128 cullBits5 = _mm_cmpgt_ps( maxW0, z0 );
cullBits0 = _mm_or_ps( cullBits0, _mm_cmpgt_ps( x1, minW1 ) );
cullBits1 = _mm_or_ps( cullBits1, _mm_cmpgt_ps( maxW1, x1 ) );
cullBits2 = _mm_or_ps( cullBits2, _mm_cmpgt_ps( y1, minW1 ) );
cullBits3 = _mm_or_ps( cullBits3, _mm_cmpgt_ps( maxW1, y1 ) );
cullBits4 = _mm_or_ps( cullBits4, _mm_cmpgt_ps( z1, minZ1 ) ); // NOTE: using minZ1
cullBits5 = _mm_or_ps( cullBits5, _mm_cmpgt_ps( maxW1, z1 ) );
cullBits0 = _mm_and_ps( cullBits0, vector_float_mask0 );
cullBits1 = _mm_and_ps( cullBits1, vector_float_mask1 );
cullBits2 = _mm_and_ps( cullBits2, vector_float_mask2 );
cullBits3 = _mm_and_ps( cullBits3, vector_float_mask3 );
cullBits4 = _mm_and_ps( cullBits4, vector_float_mask4 );
cullBits5 = _mm_and_ps( cullBits5, vector_float_mask5 );
cullBits0 = _mm_or_ps( cullBits0, cullBits1 );
cullBits2 = _mm_or_ps( cullBits2, cullBits3 );
cullBits4 = _mm_or_ps( cullBits4, cullBits5 );
cullBits0 = _mm_or_ps( cullBits0, cullBits2 );
cullBits0 = _mm_or_ps( cullBits0, cullBits4 );
cullBits0 = _mm_or_ps( cullBits0, _mm_perm_ps( cullBits0, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
cullBits0 = _mm_or_ps( cullBits0, _mm_perm_ps( cullBits0, _MM_SHUFFLE( 0, 1, 0, 1 ) ) );
int bits = _mm_cvtsi128_si32( ( const __m128i& )cullBits0 );
*outBits = ( byte )( bits ^ 63 );
return ( bits != 63 );
}
/*
========================
idRenderMatrix::CullExtrudedBoundsToMVPbits
Returns true if nothing contained in the extruded bounds is transformed by the
given Model View Projection (MVP) matrix to anything inside the clip space.
The given bounds is extruded in the 'extrudeDirection' up to the 'clipPlane'.
Normally the clip space extends from -1.0 to 1.0 on each axis, but by setting 'zeroToOne'
to true, the clip space will extend from 0.0 to 1.0 on each axis for a light projection matrix.
When all the corners of the bounding box are behind one of the six frustum planes, the box is
culled. This is conservative, because some boxes may "cross corners" and can be in front of a
frustum plane, but only while also being behind another one.
========================
*/
bool idRenderMatrix::CullExtrudedBoundsToMVPbits( const idRenderMatrix& mvp, const idBounds& bounds, const idVec3& extrudeDirection, const idPlane& clipPlane, byte* outBits, bool zeroToOne )
{
assert( idMath::Fabs( extrudeDirection * clipPlane.Normal() ) >= idMath::FLT_SMALLEST_NON_DENORMAL );
__m128 mvp0 = _mm_loadu_ps( mvp[0] );
__m128 mvp1 = _mm_loadu_ps( mvp[1] );
__m128 mvp2 = _mm_loadu_ps( mvp[2] );
__m128 mvp3 = _mm_loadu_ps( mvp[3] );
__m128 minMul = zeroToOne ? vector_float_zero : vector_float_neg_one;
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
__m128 cullBits0;
__m128 cullBits1;
__m128 cullBits2;
__m128 cullBits3;
__m128 cullBits4;
__m128 cullBits5;
// calculate the cull bits for the bounding box corners
{
// compute four partial X,Y,Z,W values
__m128 parx = _mm_splat_ps( mvp0, 3 );
__m128 pary = _mm_splat_ps( mvp1, 3 );
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parx = _mm_madd_ps( vx, _mm_splat_ps( mvp0, 0 ), parx );
pary = _mm_madd_ps( vx, _mm_splat_ps( mvp1, 0 ), pary );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parx = _mm_madd_ps( vy, _mm_splat_ps( mvp0, 1 ), parx );
pary = _mm_madd_ps( vy, _mm_splat_ps( mvp1, 1 ), pary );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
// compute full X,Y,Z,W values
__m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
__m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
__m128 x0 = _mm_madd_ps( vz0, mvp0Z, parx );
__m128 y0 = _mm_madd_ps( vz0, mvp1Z, pary );
__m128 z0 = _mm_madd_ps( vz0, mvp2Z, parz );
__m128 w0 = _mm_madd_ps( vz0, mvp3Z, parw );
__m128 x1 = _mm_madd_ps( vz1, mvp0Z, parx );
__m128 y1 = _mm_madd_ps( vz1, mvp1Z, pary );
__m128 z1 = _mm_madd_ps( vz1, mvp2Z, parz );
__m128 w1 = _mm_madd_ps( vz1, mvp3Z, parw );
__m128 maxW0 = w0;
__m128 maxW1 = w1;
__m128 minW0 = _mm_mul_ps( w0, minMul );
__m128 minW1 = _mm_mul_ps( w1, minMul );
#if defined( CLIP_SPACE_D3D ) // the D3D clip space Z is in the range [0,1] so always compare Z vs zero whether 'zeroToOne' is true or false
__m128 minZ0 = vector_float_zero;
__m128 minZ1 = vector_float_zero;
#else
__m128 minZ0 = minW0;
__m128 minZ1 = minW1;
#endif
cullBits0 = _mm_cmpgt_ps( x0, minW0 );
cullBits1 = _mm_cmpgt_ps( maxW0, x0 );
cullBits2 = _mm_cmpgt_ps( y0, minW0 );
cullBits3 = _mm_cmpgt_ps( maxW0, y0 );
cullBits4 = _mm_cmpgt_ps( z0, minZ0 ); // NOTE: using minZ0
cullBits5 = _mm_cmpgt_ps( maxW0, z0 );
cullBits0 = _mm_or_ps( cullBits0, _mm_cmpgt_ps( x1, minW1 ) );
cullBits1 = _mm_or_ps( cullBits1, _mm_cmpgt_ps( maxW1, x1 ) );
cullBits2 = _mm_or_ps( cullBits2, _mm_cmpgt_ps( y1, minW1 ) );
cullBits3 = _mm_or_ps( cullBits3, _mm_cmpgt_ps( maxW1, y1 ) );
cullBits4 = _mm_or_ps( cullBits4, _mm_cmpgt_ps( z1, minZ1 ) ); // NOTE: using minZ1
cullBits5 = _mm_or_ps( cullBits5, _mm_cmpgt_ps( maxW1, z1 ) );
}
// calculate and include the cull bits for the extruded bounding box corners
{
__m128 clipX = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 0 ), 0 );
__m128 clipY = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 1 ), 0 );
__m128 clipZ = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 2 ), 0 );
__m128 clipW = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 3 ), 0 );
__m128 extrudeX = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 0 ), 0 );
__m128 extrudeY = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 1 ), 0 );
__m128 extrudeZ = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 2 ), 0 );
__m128 closing = _mm_madd_ps( clipX, extrudeX, _mm_madd_ps( clipY, extrudeY, _mm_mul_ps( clipZ, extrudeZ ) ) );
__m128 invClosing = _mm_rcp32_ps( closing );
invClosing = _mm_xor_ps( invClosing, vector_float_sign_bit );
__m128 dt = _mm_madd_ps( clipX, vx, _mm_madd_ps( clipY, vy, clipW ) );
__m128 d0 = _mm_madd_ps( clipZ, vz0, dt );
__m128 d1 = _mm_madd_ps( clipZ, vz1, dt );
d0 = _mm_mul_ps( d0, invClosing );
d1 = _mm_mul_ps( d1, invClosing );
__m128 vx0 = _mm_madd_ps( extrudeX, d0, vx );
__m128 vx1 = _mm_madd_ps( extrudeX, d1, vx );
__m128 vy0 = _mm_madd_ps( extrudeY, d0, vy );
__m128 vy1 = _mm_madd_ps( extrudeY, d1, vy );
vz0 = _mm_madd_ps( extrudeZ, d0, vz0 );
vz1 = _mm_madd_ps( extrudeZ, d1, vz1 );
__m128 mvp0X = _mm_splat_ps( mvp0, 0 );
__m128 mvp1X = _mm_splat_ps( mvp1, 0 );
__m128 mvp2X = _mm_splat_ps( mvp2, 0 );
__m128 mvp3X = _mm_splat_ps( mvp3, 0 );
__m128 mvp0W = _mm_splat_ps( mvp0, 3 );
__m128 mvp1W = _mm_splat_ps( mvp1, 3 );
__m128 mvp2W = _mm_splat_ps( mvp2, 3 );
__m128 mvp3W = _mm_splat_ps( mvp3, 3 );
__m128 x0 = _mm_madd_ps( vx0, mvp0X, mvp0W );
__m128 y0 = _mm_madd_ps( vx0, mvp1X, mvp1W );
__m128 z0 = _mm_madd_ps( vx0, mvp2X, mvp2W );
__m128 w0 = _mm_madd_ps( vx0, mvp3X, mvp3W );
__m128 x1 = _mm_madd_ps( vx1, mvp0X, mvp0W );
__m128 y1 = _mm_madd_ps( vx1, mvp1X, mvp1W );
__m128 z1 = _mm_madd_ps( vx1, mvp2X, mvp2W );
__m128 w1 = _mm_madd_ps( vx1, mvp3X, mvp3W );
__m128 mvp0Y = _mm_splat_ps( mvp0, 1 );
__m128 mvp1Y = _mm_splat_ps( mvp1, 1 );
__m128 mvp2Y = _mm_splat_ps( mvp2, 1 );
__m128 mvp3Y = _mm_splat_ps( mvp3, 1 );
x0 = _mm_madd_ps( vy0, mvp0Y, x0 ); //-V537
y0 = _mm_madd_ps( vy0, mvp1Y, y0 );
z0 = _mm_madd_ps( vy0, mvp2Y, z0 ); //-V537
w0 = _mm_madd_ps( vy0, mvp3Y, w0 );
x1 = _mm_madd_ps( vy1, mvp0Y, x1 ); //-V537
y1 = _mm_madd_ps( vy1, mvp1Y, y1 );
z1 = _mm_madd_ps( vy1, mvp2Y, z1 ); //-V537
w1 = _mm_madd_ps( vy1, mvp3Y, w1 );
__m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
__m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
x0 = _mm_madd_ps( vz0, mvp0Z, x0 );
y0 = _mm_madd_ps( vz0, mvp1Z, y0 ); //-V537
z0 = _mm_madd_ps( vz0, mvp2Z, z0 );
w0 = _mm_madd_ps( vz0, mvp3Z, w0 );
x1 = _mm_madd_ps( vz1, mvp0Z, x1 );
y1 = _mm_madd_ps( vz1, mvp1Z, y1 ); //-V537
z1 = _mm_madd_ps( vz1, mvp2Z, z1 );
w1 = _mm_madd_ps( vz1, mvp3Z, w1 );
__m128 maxW0 = w0;
__m128 maxW1 = w1;
__m128 minW0 = _mm_mul_ps( w0, minMul );
__m128 minW1 = _mm_mul_ps( w1, minMul );
#if defined( CLIP_SPACE_D3D ) // the D3D clip space Z is in the range [0,1] so always compare Z vs zero whether 'zeroToOne' is true or false
__m128 minZ0 = vector_float_zero;
__m128 minZ1 = vector_float_zero;
#else
__m128 minZ0 = minW0;
__m128 minZ1 = minW1;
#endif
cullBits0 = _mm_or_ps( cullBits0, _mm_cmpgt_ps( x0, minW0 ) );
cullBits1 = _mm_or_ps( cullBits1, _mm_cmpgt_ps( maxW0, x0 ) );
cullBits2 = _mm_or_ps( cullBits2, _mm_cmpgt_ps( y0, minW0 ) );
cullBits3 = _mm_or_ps( cullBits3, _mm_cmpgt_ps( maxW0, y0 ) );
cullBits4 = _mm_or_ps( cullBits4, _mm_cmpgt_ps( z0, minZ0 ) ); // NOTE: using minZ0
cullBits5 = _mm_or_ps( cullBits5, _mm_cmpgt_ps( maxW0, z0 ) );
cullBits0 = _mm_or_ps( cullBits0, _mm_cmpgt_ps( x1, minW1 ) );
cullBits1 = _mm_or_ps( cullBits1, _mm_cmpgt_ps( maxW1, x1 ) );
cullBits2 = _mm_or_ps( cullBits2, _mm_cmpgt_ps( y1, minW1 ) );
cullBits3 = _mm_or_ps( cullBits3, _mm_cmpgt_ps( maxW1, y1 ) );
cullBits4 = _mm_or_ps( cullBits4, _mm_cmpgt_ps( z1, minZ1 ) ); // NOTE: using minZ1
cullBits5 = _mm_or_ps( cullBits5, _mm_cmpgt_ps( maxW1, z1 ) );
}
cullBits0 = _mm_and_ps( cullBits0, vector_float_mask0 );
cullBits1 = _mm_and_ps( cullBits1, vector_float_mask1 );
cullBits2 = _mm_and_ps( cullBits2, vector_float_mask2 );
cullBits3 = _mm_and_ps( cullBits3, vector_float_mask3 );
cullBits4 = _mm_and_ps( cullBits4, vector_float_mask4 );
cullBits5 = _mm_and_ps( cullBits5, vector_float_mask5 );
cullBits0 = _mm_or_ps( cullBits0, cullBits1 );
cullBits2 = _mm_or_ps( cullBits2, cullBits3 );
cullBits4 = _mm_or_ps( cullBits4, cullBits5 );
cullBits0 = _mm_or_ps( cullBits0, cullBits2 );
cullBits0 = _mm_or_ps( cullBits0, cullBits4 );
cullBits0 = _mm_or_ps( cullBits0, _mm_perm_ps( cullBits0, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
cullBits0 = _mm_or_ps( cullBits0, _mm_perm_ps( cullBits0, _MM_SHUFFLE( 0, 1, 0, 1 ) ) );
int bits = _mm_cvtsi128_si32( ( const __m128i& )cullBits0 );
*outBits = ( byte )( bits ^ 63 );
return ( bits != 63 );
}
/*
========================
idRenderMatrix::ProjectedBounds
Calculates the bounds of the given bounding box projected with the given Model View Projection (MVP) matrix.
If 'windowSpace' is true then the calculated bounds along each axis are moved and clamped to the [0, 1] range.
The given bounding box is not clipped to the MVP so the projected bounds may not be as tight as possible.
If the given bounding box is W=0 clipped then the projected bounds will cover the full X-Y range.
Note that while projected[0][1] will be set to the minimum when the given bounding box is W=0 clipped,
projected[1][1] will still be valid and will NOT be set to the maximum when the given bounding box
is W=0 clipped.
========================
*/
void idRenderMatrix::ProjectedBounds( idBounds& projected, const idRenderMatrix& mvp, const idBounds& bounds, bool windowSpace )
{
__m128 mvp0 = _mm_loadu_ps( mvp[0] );
__m128 mvp1 = _mm_loadu_ps( mvp[1] );
__m128 mvp2 = _mm_loadu_ps( mvp[2] );
__m128 mvp3 = _mm_loadu_ps( mvp[3] );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
// compute four partial X,Y,Z,W values
__m128 parx = _mm_splat_ps( mvp0, 3 );
__m128 pary = _mm_splat_ps( mvp1, 3 );
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parx = _mm_madd_ps( vx, _mm_splat_ps( mvp0, 0 ), parx );
pary = _mm_madd_ps( vx, _mm_splat_ps( mvp1, 0 ), pary );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parx = _mm_madd_ps( vy, _mm_splat_ps( mvp0, 1 ), parx );
pary = _mm_madd_ps( vy, _mm_splat_ps( mvp1, 1 ), pary );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
// compute full X,Y,Z,W values
__m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
__m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
__m128 x0 = _mm_madd_ps( vz0, mvp0Z, parx );
__m128 y0 = _mm_madd_ps( vz0, mvp1Z, pary );
__m128 z0 = _mm_madd_ps( vz0, mvp2Z, parz );
__m128 w0 = _mm_madd_ps( vz0, mvp3Z, parw );
__m128 x1 = _mm_madd_ps( vz1, mvp0Z, parx );
__m128 y1 = _mm_madd_ps( vz1, mvp1Z, pary );
__m128 z1 = _mm_madd_ps( vz1, mvp2Z, parz );
__m128 w1 = _mm_madd_ps( vz1, mvp3Z, parw );
__m128 s0 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w0 );
__m128 s1 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w1 );
w0 = _mm_sel_ps( w0, vector_float_one, s0 );
w1 = _mm_sel_ps( w1, vector_float_one, s1 );
__m128 rw0 = _mm_rcp32_ps( w0 );
__m128 rw1 = _mm_rcp32_ps( w1 );
x0 = _mm_mul_ps( x0, rw0 );
y0 = _mm_mul_ps( y0, rw0 );
z0 = _mm_mul_ps( z0, rw0 );
x1 = _mm_mul_ps( x1, rw1 );
y1 = _mm_mul_ps( y1, rw1 );
z1 = _mm_mul_ps( z1, rw1 );
__m128 minX = _mm_min_ps( x0, x1 );
__m128 minY = _mm_min_ps( y0, y1 );
__m128 minZ = _mm_min_ps( z0, z1 );
__m128 maxX = _mm_max_ps( x0, x1 );
__m128 maxY = _mm_max_ps( y0, y1 );
__m128 maxZ = _mm_max_ps( z0, z1 );
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
s0 = _mm_or_ps( s0, s1 );
s0 = _mm_or_ps( s0, _mm_perm_ps( s0, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
s0 = _mm_or_ps( s0, _mm_perm_ps( s0, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minX = _mm_sel_ps( minX, vector_float_neg_infinity, s0 );
minY = _mm_sel_ps( minY, vector_float_neg_infinity, s0 );
minZ = _mm_sel_ps( minZ, vector_float_neg_infinity, s0 );
maxX = _mm_sel_ps( maxX, vector_float_pos_infinity, s0 );
maxY = _mm_sel_ps( maxY, vector_float_pos_infinity, s0 );
// NOTE: maxZ is valid either way
if( windowSpace )
{
minX = _mm_madd_ps( minX, vector_float_half, vector_float_half );
maxX = _mm_madd_ps( maxX, vector_float_half, vector_float_half );
minY = _mm_madd_ps( minY, vector_float_half, vector_float_half );
maxY = _mm_madd_ps( maxY, vector_float_half, vector_float_half );
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minZ = _mm_madd_ps( minZ, vector_float_half, vector_float_half );
maxZ = _mm_madd_ps( maxZ, vector_float_half, vector_float_half );
#endif
minX = _mm_max_ps( _mm_min_ps( minX, vector_float_one ), vector_float_zero );
maxX = _mm_max_ps( _mm_min_ps( maxX, vector_float_one ), vector_float_zero );
minY = _mm_max_ps( _mm_min_ps( minY, vector_float_one ), vector_float_zero );
maxY = _mm_max_ps( _mm_min_ps( maxY, vector_float_one ), vector_float_zero );
minZ = _mm_max_ps( _mm_min_ps( minZ, vector_float_one ), vector_float_zero );
maxZ = _mm_max_ps( _mm_min_ps( maxZ, vector_float_one ), vector_float_zero );
}
_mm_store_ss( & projected[0].x, minX );
_mm_store_ss( & projected[0].y, minY );
_mm_store_ss( & projected[0].z, minZ );
_mm_store_ss( & projected[1].x, maxX );
_mm_store_ss( & projected[1].y, maxY );
_mm_store_ss( & projected[1].z, maxZ );
}
/*
========================
idRenderMatrix::ProjectedNearClippedBounds
Calculates the bounds of the given bounding box projected with the given Model View Projection (MVP) matrix.
If 'windowSpace' is true then the calculated bounds along each axis are moved and clamped to the [0, 1] range.
The given bounding box is first near clipped so the projected bounds do not cover the full X-Y range when
the given bounding box crosses the W=0 plane. However, the given bounding box is not clipped against the
other planes so the projected bounds are still not as tight as they could be if the given bounding box
crosses a corner. Fortunately, clipping to the near clipping planes typically provides more than 50% of
the gain between not clipping at all and fully clipping the bounding box to all planes. Only clipping to
the near clipping plane is much cheaper than clipping to all planes and can be easily implemented with
completely branchless SIMD.
========================
*/
void idRenderMatrix::ProjectedNearClippedBounds( idBounds& projected, const idRenderMatrix& mvp, const idBounds& bounds, bool windowSpace )
{
/*
4----{E}---5
+ /| /|
Z {H} {I} {F} |
- / | / {J}
7--{G}-----6 |
| | | |
{L} 0----|-{A}-1
| / {K} / -
| {D} | {B} Y
|/ |/ +
3---{C}----2
- X +
*/
const __m128 mvp0 = _mm_loadu_ps( mvp[0] );
const __m128 mvp1 = _mm_loadu_ps( mvp[1] );
const __m128 mvp2 = _mm_loadu_ps( mvp[2] );
const __m128 mvp3 = _mm_loadu_ps( mvp[3] );
const __m128 b0 = _mm_loadu_bounds_0( bounds );
const __m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
const __m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
const __m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
const __m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
const __m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
const __m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
// compute four partial X,Y,Z,W values
__m128 parx = _mm_splat_ps( mvp0, 3 );
__m128 pary = _mm_splat_ps( mvp1, 3 );
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parx = _mm_madd_ps( vx, _mm_splat_ps( mvp0, 0 ), parx );
pary = _mm_madd_ps( vx, _mm_splat_ps( mvp1, 0 ), pary );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parx = _mm_madd_ps( vy, _mm_splat_ps( mvp0, 1 ), parx );
pary = _mm_madd_ps( vy, _mm_splat_ps( mvp1, 1 ), pary );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
// compute full X,Y,Z,W values
const __m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
const __m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
const __m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
const __m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
const __m128 x_0123 = _mm_madd_ps( vz0, mvp0Z, parx );
const __m128 y_0123 = _mm_madd_ps( vz0, mvp1Z, pary );
const __m128 z_0123 = _mm_madd_ps( vz0, mvp2Z, parz );
const __m128 w_0123 = _mm_madd_ps( vz0, mvp3Z, parw );
const __m128 x_4567 = _mm_madd_ps( vz1, mvp0Z, parx );
const __m128 y_4567 = _mm_madd_ps( vz1, mvp1Z, pary );
const __m128 z_4567 = _mm_madd_ps( vz1, mvp2Z, parz );
const __m128 w_4567 = _mm_madd_ps( vz1, mvp3Z, parw );
// rotate the X,Y,Z,W values up by one
const __m128 x_1230 = _mm_perm_ps( x_0123, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 y_1230 = _mm_perm_ps( y_0123, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 z_1230 = _mm_perm_ps( z_0123, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 w_1230 = _mm_perm_ps( w_0123, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 x_5674 = _mm_perm_ps( x_4567, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 y_5674 = _mm_perm_ps( y_4567, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 z_5674 = _mm_perm_ps( z_4567, _MM_SHUFFLE( 0, 3, 2, 1 ) );
const __m128 w_5674 = _mm_perm_ps( w_4567, _MM_SHUFFLE( 0, 3, 2, 1 ) );
#if defined( CLIP_SPACE_D3D ) // the D3D near plane is at Z=0 instead of Z=-1
const __m128 d_0123 = z_0123;
const __m128 d_4567 = z_4567;
const __m128 d_1230 = z_1230;
const __m128 d_5674 = z_5674;
#else
const __m128 d_0123 = _mm_add_ps( z_0123, w_0123 );
const __m128 d_4567 = _mm_add_ps( z_4567, w_4567 );
const __m128 d_1230 = _mm_add_ps( z_1230, w_1230 );
const __m128 d_5674 = _mm_add_ps( z_5674, w_5674 );
#endif
const __m128 deltaABCD = _mm_sub_ps( d_0123, d_1230 );
const __m128 deltaEFGH = _mm_sub_ps( d_4567, d_5674 );
const __m128 deltaIJKL = _mm_sub_ps( d_0123, d_4567 );
const __m128 maskABCD = _mm_cmpgt_ps( _mm_and_ps( deltaABCD, vector_float_abs_mask ), vector_float_smallest_non_denorm );
const __m128 maskEFGH = _mm_cmpgt_ps( _mm_and_ps( deltaEFGH, vector_float_abs_mask ), vector_float_smallest_non_denorm );
const __m128 maskIJKL = _mm_cmpgt_ps( _mm_and_ps( deltaIJKL, vector_float_abs_mask ), vector_float_smallest_non_denorm );
const __m128 fractionABCD = _mm_and_ps( _mm_div32_ps( d_0123, _mm_sel_ps( vector_float_one, deltaABCD, maskABCD ) ), maskABCD );
const __m128 fractionEFGH = _mm_and_ps( _mm_div32_ps( d_4567, _mm_sel_ps( vector_float_one, deltaEFGH, maskEFGH ) ), maskEFGH );
const __m128 fractionIJKL = _mm_and_ps( _mm_div32_ps( d_0123, _mm_sel_ps( vector_float_one, deltaIJKL, maskIJKL ) ), maskIJKL );
const __m128 clipABCD = _mm_and_ps( _mm_cmpgt_ps( fractionABCD, vector_float_zero ), _mm_cmpgt_ps( vector_float_one, fractionABCD ) );
const __m128 clipEFGH = _mm_and_ps( _mm_cmpgt_ps( fractionEFGH, vector_float_zero ), _mm_cmpgt_ps( vector_float_one, fractionEFGH ) );
const __m128 clipIJKL = _mm_and_ps( _mm_cmpgt_ps( fractionIJKL, vector_float_zero ), _mm_cmpgt_ps( vector_float_one, fractionIJKL ) );
const __m128 intersectionABCD_x = _mm_madd_ps( fractionABCD, _mm_sub_ps( x_1230, x_0123 ), x_0123 );
const __m128 intersectionABCD_y = _mm_madd_ps( fractionABCD, _mm_sub_ps( y_1230, y_0123 ), y_0123 );
const __m128 intersectionABCD_z = _mm_madd_ps( fractionABCD, _mm_sub_ps( z_1230, z_0123 ), z_0123 );
const __m128 intersectionABCD_w = _mm_madd_ps( fractionABCD, _mm_sub_ps( w_1230, w_0123 ), w_0123 );
const __m128 intersectionEFGH_x = _mm_madd_ps( fractionEFGH, _mm_sub_ps( x_5674, x_4567 ), x_4567 );
const __m128 intersectionEFGH_y = _mm_madd_ps( fractionEFGH, _mm_sub_ps( y_5674, y_4567 ), y_4567 );
const __m128 intersectionEFGH_z = _mm_madd_ps( fractionEFGH, _mm_sub_ps( z_5674, z_4567 ), z_4567 );
const __m128 intersectionEFGH_w = _mm_madd_ps( fractionEFGH, _mm_sub_ps( w_5674, w_4567 ), w_4567 );
const __m128 intersectionIJKL_x = _mm_madd_ps( fractionIJKL, _mm_sub_ps( x_4567, x_0123 ), x_0123 );
const __m128 intersectionIJKL_y = _mm_madd_ps( fractionIJKL, _mm_sub_ps( y_4567, y_0123 ), y_0123 );
const __m128 intersectionIJKL_z = _mm_madd_ps( fractionIJKL, _mm_sub_ps( z_4567, z_0123 ), z_0123 );
const __m128 intersectionIJKL_w = _mm_madd_ps( fractionIJKL, _mm_sub_ps( w_4567, w_0123 ), w_0123 );
const __m128 mask_0123 = _mm_cmpgt_ps( vector_float_zero, d_0123 );
const __m128 mask_1230 = _mm_cmpgt_ps( vector_float_zero, d_1230 );
const __m128 mask_4567 = _mm_cmpgt_ps( vector_float_zero, d_4567 );
const __m128 mask_5674 = _mm_cmpgt_ps( vector_float_zero, d_5674 );
const __m128 maskABCD_0123 = _mm_and_ps( clipABCD, mask_0123 );
const __m128 maskABCD_1230 = _mm_and_ps( clipABCD, mask_1230 );
const __m128 maskEFGH_4567 = _mm_and_ps( clipEFGH, mask_4567 );
const __m128 maskEFGH_5674 = _mm_and_ps( clipEFGH, mask_5674 );
const __m128 maskIJKL_0123 = _mm_and_ps( clipIJKL, mask_0123 );
const __m128 maskIJKL_4567 = _mm_and_ps( clipIJKL, mask_4567 );
__m128 edgeVertsABCD_x0 = _mm_sel_ps( x_0123, intersectionABCD_x, maskABCD_0123 );
__m128 edgeVertsABCD_y0 = _mm_sel_ps( y_0123, intersectionABCD_y, maskABCD_0123 );
__m128 edgeVertsABCD_z0 = _mm_sel_ps( z_0123, intersectionABCD_z, maskABCD_0123 );
__m128 edgeVertsABCD_w0 = _mm_sel_ps( w_0123, intersectionABCD_w, maskABCD_0123 );
__m128 edgeVertsABCD_x1 = _mm_sel_ps( x_1230, intersectionABCD_x, maskABCD_1230 );
__m128 edgeVertsABCD_y1 = _mm_sel_ps( y_1230, intersectionABCD_y, maskABCD_1230 );
__m128 edgeVertsABCD_z1 = _mm_sel_ps( z_1230, intersectionABCD_z, maskABCD_1230 );
__m128 edgeVertsABCD_w1 = _mm_sel_ps( w_1230, intersectionABCD_w, maskABCD_1230 );
__m128 edgeVertsEFGH_x0 = _mm_sel_ps( x_4567, intersectionEFGH_x, maskEFGH_4567 );
__m128 edgeVertsEFGH_y0 = _mm_sel_ps( y_4567, intersectionEFGH_y, maskEFGH_4567 );
__m128 edgeVertsEFGH_z0 = _mm_sel_ps( z_4567, intersectionEFGH_z, maskEFGH_4567 );
__m128 edgeVertsEFGH_w0 = _mm_sel_ps( w_4567, intersectionEFGH_w, maskEFGH_4567 );
__m128 edgeVertsEFGH_x1 = _mm_sel_ps( x_5674, intersectionEFGH_x, maskEFGH_5674 );
__m128 edgeVertsEFGH_y1 = _mm_sel_ps( y_5674, intersectionEFGH_y, maskEFGH_5674 );
__m128 edgeVertsEFGH_z1 = _mm_sel_ps( z_5674, intersectionEFGH_z, maskEFGH_5674 );
__m128 edgeVertsEFGH_w1 = _mm_sel_ps( w_5674, intersectionEFGH_w, maskEFGH_5674 );
__m128 edgeVertsIJKL_x0 = _mm_sel_ps( x_0123, intersectionIJKL_x, maskIJKL_0123 );
__m128 edgeVertsIJKL_y0 = _mm_sel_ps( y_0123, intersectionIJKL_y, maskIJKL_0123 );
__m128 edgeVertsIJKL_z0 = _mm_sel_ps( z_0123, intersectionIJKL_z, maskIJKL_0123 );
__m128 edgeVertsIJKL_w0 = _mm_sel_ps( w_0123, intersectionIJKL_w, maskIJKL_0123 );
__m128 edgeVertsIJKL_x1 = _mm_sel_ps( x_4567, intersectionIJKL_x, maskIJKL_4567 );
__m128 edgeVertsIJKL_y1 = _mm_sel_ps( y_4567, intersectionIJKL_y, maskIJKL_4567 );
__m128 edgeVertsIJKL_z1 = _mm_sel_ps( z_4567, intersectionIJKL_z, maskIJKL_4567 );
__m128 edgeVertsIJKL_w1 = _mm_sel_ps( w_4567, intersectionIJKL_w, maskIJKL_4567 );
const __m128 maskABCD_w0 = _mm_cmpgt_ps( edgeVertsABCD_w0, vector_float_smallest_non_denorm );
const __m128 maskABCD_w1 = _mm_cmpgt_ps( edgeVertsABCD_w1, vector_float_smallest_non_denorm );
const __m128 maskEFGH_w0 = _mm_cmpgt_ps( edgeVertsEFGH_w0, vector_float_smallest_non_denorm );
const __m128 maskEFGH_w1 = _mm_cmpgt_ps( edgeVertsEFGH_w1, vector_float_smallest_non_denorm );
const __m128 maskIJKL_w0 = _mm_cmpgt_ps( edgeVertsIJKL_w0, vector_float_smallest_non_denorm );
const __m128 maskIJKL_w1 = _mm_cmpgt_ps( edgeVertsIJKL_w1, vector_float_smallest_non_denorm );
edgeVertsABCD_w0 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsABCD_w0, maskABCD_w0 ) );
edgeVertsABCD_w1 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsABCD_w1, maskABCD_w1 ) );
edgeVertsEFGH_w0 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsEFGH_w0, maskEFGH_w0 ) );
edgeVertsEFGH_w1 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsEFGH_w1, maskEFGH_w1 ) );
edgeVertsIJKL_w0 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsIJKL_w0, maskIJKL_w0 ) );
edgeVertsIJKL_w1 = _mm_rcp32_ps( _mm_sel_ps( vector_float_one, edgeVertsIJKL_w1, maskIJKL_w1 ) );
edgeVertsABCD_x0 = _mm_mul_ps( edgeVertsABCD_x0, edgeVertsABCD_w0 );
edgeVertsABCD_x1 = _mm_mul_ps( edgeVertsABCD_x1, edgeVertsABCD_w1 );
edgeVertsEFGH_x0 = _mm_mul_ps( edgeVertsEFGH_x0, edgeVertsEFGH_w0 );
edgeVertsEFGH_x1 = _mm_mul_ps( edgeVertsEFGH_x1, edgeVertsEFGH_w1 );
edgeVertsIJKL_x0 = _mm_mul_ps( edgeVertsIJKL_x0, edgeVertsIJKL_w0 );
edgeVertsIJKL_x1 = _mm_mul_ps( edgeVertsIJKL_x1, edgeVertsIJKL_w1 );
edgeVertsABCD_y0 = _mm_mul_ps( edgeVertsABCD_y0, edgeVertsABCD_w0 );
edgeVertsABCD_y1 = _mm_mul_ps( edgeVertsABCD_y1, edgeVertsABCD_w1 );
edgeVertsEFGH_y0 = _mm_mul_ps( edgeVertsEFGH_y0, edgeVertsEFGH_w0 );
edgeVertsEFGH_y1 = _mm_mul_ps( edgeVertsEFGH_y1, edgeVertsEFGH_w1 );
edgeVertsIJKL_y0 = _mm_mul_ps( edgeVertsIJKL_y0, edgeVertsIJKL_w0 );
edgeVertsIJKL_y1 = _mm_mul_ps( edgeVertsIJKL_y1, edgeVertsIJKL_w1 );
edgeVertsABCD_z0 = _mm_mul_ps( edgeVertsABCD_z0, edgeVertsABCD_w0 );
edgeVertsABCD_z1 = _mm_mul_ps( edgeVertsABCD_z1, edgeVertsABCD_w1 );
edgeVertsEFGH_z0 = _mm_mul_ps( edgeVertsEFGH_z0, edgeVertsEFGH_w0 );
edgeVertsEFGH_z1 = _mm_mul_ps( edgeVertsEFGH_z1, edgeVertsEFGH_w1 );
edgeVertsIJKL_z0 = _mm_mul_ps( edgeVertsIJKL_z0, edgeVertsIJKL_w0 );
edgeVertsIJKL_z1 = _mm_mul_ps( edgeVertsIJKL_z1, edgeVertsIJKL_w1 );
const __m128 posInf = vector_float_pos_infinity;
const __m128 negInf = vector_float_neg_infinity;
const __m128 minX0 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsABCD_x0, maskABCD_w0 ), _mm_sel_ps( posInf, edgeVertsABCD_x1, maskABCD_w1 ) );
const __m128 minX1 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsEFGH_x0, maskEFGH_w0 ), _mm_sel_ps( posInf, edgeVertsEFGH_x1, maskEFGH_w1 ) );
const __m128 minX2 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsIJKL_x0, maskIJKL_w0 ), _mm_sel_ps( posInf, edgeVertsIJKL_x1, maskIJKL_w1 ) );
const __m128 minY0 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsABCD_y0, maskABCD_w0 ), _mm_sel_ps( posInf, edgeVertsABCD_y1, maskABCD_w1 ) );
const __m128 minY1 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsEFGH_y0, maskEFGH_w0 ), _mm_sel_ps( posInf, edgeVertsEFGH_y1, maskEFGH_w1 ) );
const __m128 minY2 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsIJKL_y0, maskIJKL_w0 ), _mm_sel_ps( posInf, edgeVertsIJKL_y1, maskIJKL_w1 ) );
const __m128 minZ0 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsABCD_z0, maskABCD_w0 ), _mm_sel_ps( posInf, edgeVertsABCD_z1, maskABCD_w1 ) );
const __m128 minZ1 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsEFGH_z0, maskEFGH_w0 ), _mm_sel_ps( posInf, edgeVertsEFGH_z1, maskEFGH_w1 ) );
const __m128 minZ2 = _mm_min_ps( _mm_sel_ps( posInf, edgeVertsIJKL_z0, maskIJKL_w0 ), _mm_sel_ps( posInf, edgeVertsIJKL_z1, maskIJKL_w1 ) );
const __m128 maxX0 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsABCD_x0, maskABCD_w0 ), _mm_sel_ps( negInf, edgeVertsABCD_x1, maskABCD_w1 ) );
const __m128 maxX1 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsEFGH_x0, maskEFGH_w0 ), _mm_sel_ps( negInf, edgeVertsEFGH_x1, maskEFGH_w1 ) );
const __m128 maxX2 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsIJKL_x0, maskIJKL_w0 ), _mm_sel_ps( negInf, edgeVertsIJKL_x1, maskIJKL_w1 ) );
const __m128 maxY0 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsABCD_y0, maskABCD_w0 ), _mm_sel_ps( negInf, edgeVertsABCD_y1, maskABCD_w1 ) );
const __m128 maxY1 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsEFGH_y0, maskEFGH_w0 ), _mm_sel_ps( negInf, edgeVertsEFGH_y1, maskEFGH_w1 ) );
const __m128 maxY2 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsIJKL_y0, maskIJKL_w0 ), _mm_sel_ps( negInf, edgeVertsIJKL_y1, maskIJKL_w1 ) );
const __m128 maxZ0 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsABCD_z0, maskABCD_w0 ), _mm_sel_ps( negInf, edgeVertsABCD_z1, maskABCD_w1 ) );
const __m128 maxZ1 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsEFGH_z0, maskEFGH_w0 ), _mm_sel_ps( negInf, edgeVertsEFGH_z1, maskEFGH_w1 ) );
const __m128 maxZ2 = _mm_max_ps( _mm_sel_ps( negInf, edgeVertsIJKL_z0, maskIJKL_w0 ), _mm_sel_ps( negInf, edgeVertsIJKL_z1, maskIJKL_w1 ) );
__m128 minX = _mm_min_ps( minX0, _mm_min_ps( minX1, minX2 ) );
__m128 minY = _mm_min_ps( minY0, _mm_min_ps( minY1, minY2 ) );
__m128 minZ = _mm_min_ps( minZ0, _mm_min_ps( minZ1, minZ2 ) );
__m128 maxX = _mm_max_ps( maxX0, _mm_max_ps( maxX1, maxX2 ) );
__m128 maxY = _mm_max_ps( maxY0, _mm_max_ps( maxY1, maxY2 ) );
__m128 maxZ = _mm_max_ps( maxZ0, _mm_max_ps( maxZ1, maxZ2 ) );
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
if( windowSpace )
{
minX = _mm_madd_ps( minX, vector_float_half, vector_float_half );
maxX = _mm_madd_ps( maxX, vector_float_half, vector_float_half );
minY = _mm_madd_ps( minY, vector_float_half, vector_float_half );
maxY = _mm_madd_ps( maxY, vector_float_half, vector_float_half );
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minZ = _mm_madd_ps( minZ, vector_float_half, vector_float_half );
maxZ = _mm_madd_ps( maxZ, vector_float_half, vector_float_half );
#endif
minX = _mm_max_ps( _mm_min_ps( minX, vector_float_one ), vector_float_zero );
maxX = _mm_max_ps( _mm_min_ps( maxX, vector_float_one ), vector_float_zero );
minY = _mm_max_ps( _mm_min_ps( minY, vector_float_one ), vector_float_zero );
maxY = _mm_max_ps( _mm_min_ps( maxY, vector_float_one ), vector_float_zero );
minZ = _mm_max_ps( _mm_min_ps( minZ, vector_float_one ), vector_float_zero );
maxZ = _mm_max_ps( _mm_min_ps( maxZ, vector_float_one ), vector_float_zero );
}
_mm_store_ss( & projected[0].x, minX );
_mm_store_ss( & projected[0].y, minY );
_mm_store_ss( & projected[0].z, minZ );
_mm_store_ss( & projected[1].x, maxX );
_mm_store_ss( & projected[1].y, maxY );
_mm_store_ss( & projected[1].z, maxZ );
}
#if 0
/*
========================
LocalViewOriginFromMVP
========================
*/
static idVec3 LocalViewOriginFromMVP( const idRenderMatrix& mvp )
{
const float nearX = mvp[3][0] + mvp[2][0];
const float nearY = mvp[3][1] + mvp[2][1];
const float nearZ = mvp[3][2] + mvp[2][2];
const float s = idMath::InvSqrt( nearX * nearX + nearY * nearY + nearZ * nearZ );
idRenderMatrix inverseMVP;
idRenderMatrix::Inverse( mvp, inverseMVP );
const float invW = 1.0f / inverseMVP[3][3];
const float x = ( inverseMVP[0][3] - nearX * s ) * invW;
const float y = ( inverseMVP[1][3] - nearY * s ) * invW;
const float z = ( inverseMVP[2][3] - nearZ * s ) * invW;
return idVec3( x, y, z );
}
#endif
/*
========================
LocalNearClipCenterFromMVP
Based on whether the depth range is [0,1] or [-1,1], either transform (0,0,0) or (0,0,-1) with the inverse MVP.
========================
*/
static idVec3 LocalNearClipCenterFromMVP( const idRenderMatrix& mvp )
{
idRenderMatrix inverseMVP;
idRenderMatrix::Inverse( mvp, inverseMVP );
#if defined( CLIP_SPACE_D3D ) // the D3D near plane is at Z=0 instead of Z=-1
const float x = inverseMVP[0][3];
const float y = inverseMVP[1][3];
const float z = inverseMVP[2][3];
const float w = inverseMVP[3][3];
#else
const float x = inverseMVP[0][3] - inverseMVP[0][2];
const float y = inverseMVP[1][3] - inverseMVP[1][2];
const float z = inverseMVP[2][3] - inverseMVP[2][2];
const float w = inverseMVP[3][3] - inverseMVP[3][2];
#endif
const float invW = 1.0f / w;
return idVec3( x * invW, y * invW, z * invW );
}
/*
========================
ClipHomogeneousPolygonToSide
Clips a polygon with homogeneous coordinates to the axis aligned plane[axis] = sign * offset.
========================
*/
static void ClipHomogeneousPolygonToSide_SSE2( idVec4* __restrict newPoints, idVec4* __restrict points, int& numPoints,
const int axis, const __m128& sign, const __m128& offset )
{
assert( newPoints != points );
const __m128 side = _mm_mul_ps( sign, offset );
__m128i mask = _mm_sub_epi32( vector_int_0123, _mm_shuffle_epi32( _mm_cvtsi32_si128( numPoints ), 0 ) );
__m128i index = _mm_setzero_si128();
ALIGNTYPE16 unsigned short indices[16 * 2];
ALIGNTYPE16 float clipFractions[16];
int localNumPoint = numPoints;
for( int i = 0; i < localNumPoint; i += 4 )
{
const int i0 = ( i + 0 ) & ( ( i + 0 - localNumPoint ) >> 31 );
const int i1 = ( i + 1 ) & ( ( i + 1 - localNumPoint ) >> 31 );
const int i2 = ( i + 2 ) & ( ( i + 2 - localNumPoint ) >> 31 );
const int i3 = ( i + 3 ) & ( ( i + 3 - localNumPoint ) >> 31 );
const int i4 = ( i + 4 ) & ( ( i + 4 - localNumPoint ) >> 31 );
const __m128 p0A = _mm_load_ss( &points[i0][axis] );
const __m128 p1A = _mm_load_ss( &points[i1][axis] );
const __m128 p2A = _mm_load_ss( &points[i2][axis] );
const __m128 p3A = _mm_load_ss( &points[i3][axis] );
const __m128 p4A = _mm_load_ss( &points[i4][axis] );
const __m128 p0W = _mm_load_ss( &points[i0][3] );
const __m128 p1W = _mm_load_ss( &points[i1][3] );
const __m128 p2W = _mm_load_ss( &points[i2][3] );
const __m128 p3W = _mm_load_ss( &points[i3][3] );
const __m128 p4W = _mm_load_ss( &points[i4][3] );
const __m128 t0 = _mm_unpacklo_ps( p0A, p2A );
const __m128 t1 = _mm_unpacklo_ps( p1A, p3A );
const __m128 pa0 = _mm_unpacklo_ps( t0, t1 );
const __m128 pa1 = _mm_sld_ps( pa0, p4A, 4 );
const __m128 r0 = _mm_unpacklo_ps( p0W, p2W );
const __m128 r1 = _mm_unpacklo_ps( p1W, p3W );
const __m128 pw0 = _mm_unpacklo_ps( r0, r1 );
const __m128 pw1 = _mm_sld_ps( pw0, p4W, 4 );
{
const __m128 bside0 = _mm_cmpgt_ps( _mm_mul_ps( offset, pw0 ), _mm_mul_ps( sign, pa0 ) );
const __m128 bside1 = _mm_cmpgt_ps( _mm_mul_ps( offset, pw1 ), _mm_mul_ps( sign, pa1 ) );
const __m128i side0 = _mm_and_si128( __m128c( bside0 ), vector_int_1 );
const __m128i side1 = _mm_and_si128( __m128c( bside1 ), vector_int_1 );
const __m128i xorSide = _mm_xor_si128( side0, side1 );
const __m128i interleavedSide0 = _mm_unpacklo_epi32( side0, xorSide );
const __m128i interleavedSide1 = _mm_unpackhi_epi32( side0, xorSide );
const __m128i packedSide = _mm_packs_epi32( interleavedSide0, interleavedSide1 );
const __m128i packedMaskedSide = _mm_and_si128( packedSide, _mm_srai_epi32( mask, 31 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 2 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 4 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 6 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 8 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 10 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 12 ) );
index = _mm_add_epi16( index, _mm_slli_si128( packedMaskedSide, 14 ) );
_mm_store_si128( ( __m128i* )&indices[i * 2], index );
mask = _mm_add_epi32( mask, vector_int_4 );
index = _mm_add_epi16( index, packedMaskedSide );
index = _mm_shufflehi_epi16( index, _MM_SHUFFLE( 3, 3, 3, 3 ) );
index = _mm_shuffle_epi32( index, _MM_SHUFFLE( 3, 3, 3, 3 ) );
}
{
const __m128 d0 = _mm_nmsub_ps( pw0, side, pa0 );
const __m128 d1 = _mm_nmsub_ps( pw1, side, pa1 );
const __m128 delta = _mm_sub_ps( d0, d1 );
const __m128 deltaAbs = _mm_and_ps( delta, vector_float_abs_mask );
const __m128 clamp = _mm_cmpgt_ps( vector_float_smallest_non_denorm, deltaAbs );
const __m128 deltaClamped = _mm_sel_ps( delta, vector_float_one, clamp );
const __m128 fraction = _mm_mul_ps( d0, _mm_rcp32_ps( deltaClamped ) );
const __m128 fractionClamped0 = _mm_sel_ps( fraction, vector_float_one, clamp );
const __m128 fractionClamped1 = _mm_max_ps( fractionClamped0, vector_float_zero );
const __m128 fractionClamped2 = _mm_min_ps( fractionClamped1, vector_float_one );
_mm_store_ps( &clipFractions[i], fractionClamped2 );
}
}
numPoints = _mm_cvtsi128_si32( index ) & 0xFFFF;
for( int i = 0; i < localNumPoint; i += 4 )
{
const int i0 = ( i + 0 ) & ( ( i + 0 - localNumPoint ) >> 31 );
const int i1 = ( i + 1 ) & ( ( i + 1 - localNumPoint ) >> 31 );
const int i2 = ( i + 2 ) & ( ( i + 2 - localNumPoint ) >> 31 );
const int i3 = ( i + 3 ) & ( ( i + 3 - localNumPoint ) >> 31 );
const int i4 = ( i + 4 ) & ( ( i + 4 - localNumPoint ) >> 31 );
const __m128 p0 = _mm_load_ps( points[i0].ToFloatPtr() );
const __m128 p1 = _mm_load_ps( points[i1].ToFloatPtr() );
const __m128 p2 = _mm_load_ps( points[i2].ToFloatPtr() );
const __m128 p3 = _mm_load_ps( points[i3].ToFloatPtr() );
const __m128 p4 = _mm_load_ps( points[i4].ToFloatPtr() );
const __m128 fraction = _mm_load_ps( &clipFractions[i] );
const __m128 c0 = _mm_madd_ps( _mm_splat_ps( fraction, 0 ), _mm_sub_ps( p1, p0 ), p0 );
const __m128 c1 = _mm_madd_ps( _mm_splat_ps( fraction, 1 ), _mm_sub_ps( p2, p1 ), p1 );
const __m128 c2 = _mm_madd_ps( _mm_splat_ps( fraction, 2 ), _mm_sub_ps( p3, p2 ), p2 );
const __m128 c3 = _mm_madd_ps( _mm_splat_ps( fraction, 3 ), _mm_sub_ps( p4, p3 ), p3 );
_mm_store_ps( newPoints[indices[i * 2 + 0]].ToFloatPtr(), p0 );
_mm_store_ps( newPoints[indices[i * 2 + 1]].ToFloatPtr(), c0 );
_mm_store_ps( newPoints[indices[i * 2 + 2]].ToFloatPtr(), p1 );
_mm_store_ps( newPoints[indices[i * 2 + 3]].ToFloatPtr(), c1 );
_mm_store_ps( newPoints[indices[i * 2 + 4]].ToFloatPtr(), p2 );
_mm_store_ps( newPoints[indices[i * 2 + 5]].ToFloatPtr(), c2 );
_mm_store_ps( newPoints[indices[i * 2 + 6]].ToFloatPtr(), p3 );
_mm_store_ps( newPoints[indices[i * 2 + 7]].ToFloatPtr(), c3 );
}
}
/*
========================
ClipHomogeneousPolygonToUnitCube
Clips a polygon with homogeneous coordinates to all six axis aligned unit cube planes.
========================
*/
static int ClipHomogeneousPolygonToUnitCube_SSE2( idVec4* points, int numPoints )
{
assert( numPoints < 16 - 6 );
ALIGNTYPE16 idVec4 newPoints[16 * 2];
#if defined( CLIP_SPACE_D3D ) // the D3D near plane is at Z=0 instead of Z=-1
ClipHomogeneousPolygonToSide_SSE2( newPoints, points, numPoints, 2, vector_float_neg_one, vector_float_zero ); // near
#else
ClipHomogeneousPolygonToSide_SSE2( newPoints, points, numPoints, 2, vector_float_neg_one, vector_float_one ); // near
#endif
ClipHomogeneousPolygonToSide_SSE2( points, newPoints, numPoints, 2, vector_float_pos_one, vector_float_one ); // far
ClipHomogeneousPolygonToSide_SSE2( newPoints, points, numPoints, 1, vector_float_neg_one, vector_float_one ); // bottom
ClipHomogeneousPolygonToSide_SSE2( points, newPoints, numPoints, 1, vector_float_pos_one, vector_float_one ); // top
ClipHomogeneousPolygonToSide_SSE2( newPoints, points, numPoints, 0, vector_float_neg_one, vector_float_one ); // left
ClipHomogeneousPolygonToSide_SSE2( points, newPoints, numPoints, 0, vector_float_pos_one, vector_float_one ); // right
return numPoints;
}
/*
========================
idRenderMatrix::ProjectedFullyClippedBounds
Calculates the bounds of the given bounding box projected with the given Model View Projection (MVP) matrix.
If 'windowSpace' is true then the calculated bounds along each axis are moved and clamped to the [0, 1] range.
The given bounding box is first fully clipped to the MVP to get the smallest projected bounds.
Note that this code assumes the MVP matrix has an infinite far clipping plane. When the far plane is at
infinity the bounds are never far clipped and it is sufficient to test whether or not the center of the
near clip plane is inside the bounds to calculate the correct minimum Z. If the far plane is not at
infinity then this code would also have to test for the view frustum being completely contained inside
the given bounds in which case the projected bounds should be set to fully cover the view frustum.
========================
*/
void idRenderMatrix::ProjectedFullyClippedBounds( idBounds& projected, const idRenderMatrix& mvp, const idBounds& bounds, bool windowSpace )
{
const __m128 mvp0 = _mm_loadu_ps( mvp[0] );
const __m128 mvp1 = _mm_loadu_ps( mvp[1] );
const __m128 mvp2 = _mm_loadu_ps( mvp[2] );
const __m128 mvp3 = _mm_loadu_ps( mvp[3] );
const __m128 t0 = _mm_unpacklo_ps( mvp0, mvp2 ); // mvp[0][0], mvp[2][0], mvp[0][1], mvp[2][1]
const __m128 t1 = _mm_unpackhi_ps( mvp0, mvp2 ); // mvp[0][2], mvp[2][2], mvp[0][3], mvp[2][3]
const __m128 t2 = _mm_unpacklo_ps( mvp1, mvp3 ); // mvp[1][0], mvp[3][0], mvp[1][1], mvp[3][1]
const __m128 t3 = _mm_unpackhi_ps( mvp1, mvp3 ); // mvp[1][2], mvp[3][2], mvp[1][3], mvp[3][3]
const __m128 mvpX = _mm_unpacklo_ps( t0, t2 ); // mvp[0][0], mvp[1][0], mvp[2][0], mvp[3][0]
const __m128 mvpY = _mm_unpackhi_ps( t0, t2 ); // mvp[0][1], mvp[1][1], mvp[2][1], mvp[3][1]
const __m128 mvpZ = _mm_unpacklo_ps( t1, t3 ); // mvp[0][2], mvp[1][2], mvp[2][2], mvp[3][2]
const __m128 mvpW = _mm_unpackhi_ps( t1, t3 ); // mvp[0][3], mvp[1][3], mvp[2][3], mvp[3][3]
const __m128 b0 = _mm_loadu_bounds_0( bounds );
const __m128 b1 = _mm_loadu_bounds_1( bounds );
const __m128 b0X = _mm_splat_ps( b0, 0 );
const __m128 b0Y = _mm_splat_ps( b0, 1 );
const __m128 b0Z = _mm_splat_ps( b0, 2 );
const __m128 b1X = _mm_splat_ps( b1, 0 );
const __m128 b1Y = _mm_splat_ps( b1, 1 );
const __m128 b1Z = _mm_splat_ps( b1, 2 );
const __m128 p0 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 p1 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 p2 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 p3 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 p4 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 p5 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 p6 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 p7 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
ALIGNTYPE16 idVec4 projectedPoints[8];
_mm_store_ps( projectedPoints[0].ToFloatPtr(), p0 );
_mm_store_ps( projectedPoints[1].ToFloatPtr(), p1 );
_mm_store_ps( projectedPoints[2].ToFloatPtr(), p2 );
_mm_store_ps( projectedPoints[3].ToFloatPtr(), p3 );
_mm_store_ps( projectedPoints[4].ToFloatPtr(), p4 );
_mm_store_ps( projectedPoints[5].ToFloatPtr(), p5 );
_mm_store_ps( projectedPoints[6].ToFloatPtr(), p6 );
_mm_store_ps( projectedPoints[7].ToFloatPtr(), p7 );
ALIGNTYPE16 idVec4 clippedPoints[6 * 16];
int numClippedPoints = 0;
for( int i = 0; i < 6; i++ )
{
_mm_store_ps( clippedPoints[numClippedPoints + 0].ToFloatPtr(), _mm_load_ps( projectedPoints[boxPolygonVertices[i][0]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 1].ToFloatPtr(), _mm_load_ps( projectedPoints[boxPolygonVertices[i][1]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 2].ToFloatPtr(), _mm_load_ps( projectedPoints[boxPolygonVertices[i][2]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 3].ToFloatPtr(), _mm_load_ps( projectedPoints[boxPolygonVertices[i][3]].ToFloatPtr() ) );
numClippedPoints += ClipHomogeneousPolygonToUnitCube_SSE2( &clippedPoints[numClippedPoints], 4 );
}
// repeat the first clipped point at the end to get a multiple of 4 clipped points
const __m128 point0 = _mm_load_ps( clippedPoints[0].ToFloatPtr() );
for( int i = numClippedPoints; ( i & 3 ) != 0; i++ )
{
_mm_store_ps( clippedPoints[i].ToFloatPtr(), point0 );
}
// test if the center of the near clip plane is inside the given bounding box
const idVec3 localNearClipCenter = LocalNearClipCenterFromMVP( mvp );
const bool inside = bounds.Expand( RENDER_MATRIX_PROJECTION_EPSILON ).ContainsPoint( localNearClipCenter );
__m128 minX = vector_float_pos_infinity;
__m128 minY = vector_float_pos_infinity;
__m128 minZ = inside ? vector_float_neg_one : vector_float_pos_infinity;
__m128 maxX = vector_float_neg_infinity;
__m128 maxY = vector_float_neg_infinity;
__m128 maxZ = vector_float_neg_infinity;
for( int i = 0; i < numClippedPoints; i += 4 )
{
const __m128 cp0 = _mm_load_ps( clippedPoints[i + 0].ToFloatPtr() );
const __m128 cp1 = _mm_load_ps( clippedPoints[i + 1].ToFloatPtr() );
const __m128 cp2 = _mm_load_ps( clippedPoints[i + 2].ToFloatPtr() );
const __m128 cp3 = _mm_load_ps( clippedPoints[i + 3].ToFloatPtr() );
const __m128 s0 = _mm_unpacklo_ps( cp0, cp2 ); // cp0[0], cp2[0], cp0[1], cp2[1]
const __m128 s1 = _mm_unpackhi_ps( cp0, cp2 ); // cp0[2], cp2[2], cp0[3], cp2[3]
const __m128 s2 = _mm_unpacklo_ps( cp1, cp3 ); // cp1[0], cp3[0], cp1[1], cp3[1]
const __m128 s3 = _mm_unpackhi_ps( cp1, cp3 ); // cp1[2], cp3[2], cp1[3], cp3[3]
const __m128 cpX = _mm_unpacklo_ps( s0, s2 ); // cp0[0], cp1[0], cp2[0], cp3[0]
const __m128 cpY = _mm_unpackhi_ps( s0, s2 ); // cp0[1], cp1[1], cp2[1], cp3[1]
const __m128 cpZ = _mm_unpacklo_ps( s1, s3 ); // cp0[2], cp1[2], cp2[2], cp3[2]
const __m128 cpW = _mm_unpackhi_ps( s1, s3 ); // cp0[3], cp1[3], cp2[3], cp3[3]
const __m128 rW = _mm_rcp32_ps( cpW );
const __m128 rX = _mm_mul_ps( cpX, rW );
const __m128 rY = _mm_mul_ps( cpY, rW );
const __m128 rZ = _mm_mul_ps( cpZ, rW );
minX = _mm_min_ps( minX, rX );
minY = _mm_min_ps( minY, rY );
minZ = _mm_min_ps( minZ, rZ );
maxX = _mm_max_ps( maxX, rX );
maxY = _mm_max_ps( maxY, rY );
maxZ = _mm_max_ps( maxZ, rZ );
}
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minX = _mm_min_ps( minX, _mm_perm_ps( minX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minY = _mm_min_ps( minY, _mm_perm_ps( minY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxX = _mm_max_ps( maxX, _mm_perm_ps( maxX, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxY = _mm_max_ps( maxY, _mm_perm_ps( maxY, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
if( windowSpace )
{
minX = _mm_madd_ps( minX, vector_float_half, vector_float_half );
maxX = _mm_madd_ps( maxX, vector_float_half, vector_float_half );
minY = _mm_madd_ps( minY, vector_float_half, vector_float_half );
maxY = _mm_madd_ps( maxY, vector_float_half, vector_float_half );
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minZ = _mm_madd_ps( minZ, vector_float_half, vector_float_half );
maxZ = _mm_madd_ps( maxZ, vector_float_half, vector_float_half );
#endif
minX = _mm_max_ps( _mm_min_ps( minX, vector_float_one ), vector_float_zero );
maxX = _mm_max_ps( _mm_min_ps( maxX, vector_float_one ), vector_float_zero );
minY = _mm_max_ps( _mm_min_ps( minY, vector_float_one ), vector_float_zero );
maxY = _mm_max_ps( _mm_min_ps( maxY, vector_float_one ), vector_float_zero );
minZ = _mm_max_ps( _mm_min_ps( minZ, vector_float_one ), vector_float_zero );
maxZ = _mm_max_ps( _mm_min_ps( maxZ, vector_float_one ), vector_float_zero );
}
_mm_store_ss( & projected[0].x, minX );
_mm_store_ss( & projected[0].y, minY );
_mm_store_ss( & projected[0].z, minZ );
_mm_store_ss( & projected[1].x, maxX );
_mm_store_ss( & projected[1].y, maxY );
_mm_store_ss( & projected[1].z, maxZ );
}
/*
========================
idRenderMatrix::DepthBoundsForBounds
Calculates the depth bounds of the given bounding box projected with the given Model View Projection (MVP) matrix.
If 'windowSpace' is true then the calculated depth bounds are moved and clamped to the [0, 1] range.
The given bounding box is not clipped to the MVP so the depth bounds may not be as tight as possible.
========================
*/
void idRenderMatrix::DepthBoundsForBounds( float& min, float& max, const idRenderMatrix& mvp, const idBounds& bounds, bool windowSpace )
{
__m128 mvp2 = _mm_loadu_ps( mvp[2] );
__m128 mvp3 = _mm_loadu_ps( mvp[3] );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
// compute four partial Z,W values
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
__m128 z0 = _mm_madd_ps( vz0, _mm_splat_ps( mvp2, 2 ), parz );
__m128 w0 = _mm_madd_ps( vz0, _mm_splat_ps( mvp3, 2 ), parw );
__m128 z1 = _mm_madd_ps( vz1, _mm_splat_ps( mvp2, 2 ), parz );
__m128 w1 = _mm_madd_ps( vz1, _mm_splat_ps( mvp3, 2 ), parw );
__m128 s0 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w0 );
w0 = _mm_or_ps( w0, _mm_and_ps( vector_float_smallest_non_denorm, s0 ) );
__m128 rw0 = _mm_rcp32_ps( w0 );
z0 = _mm_mul_ps( z0, rw0 );
z0 = _mm_sel_ps( z0, vector_float_neg_infinity, s0 );
__m128 s1 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w1 );
w1 = _mm_or_ps( w1, _mm_and_ps( vector_float_smallest_non_denorm, s1 ) );
__m128 rw1 = _mm_rcp32_ps( w1 );
z1 = _mm_mul_ps( z1, rw1 );
z1 = _mm_sel_ps( z1, vector_float_neg_infinity, s1 );
__m128 minv = _mm_min_ps( z0, z1 );
__m128 maxv = _mm_max_ps( z0, z1 );
minv = _mm_min_ps( minv, _mm_perm_ps( minv, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minv = _mm_min_ps( minv, _mm_perm_ps( minv, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxv = _mm_max_ps( maxv, _mm_perm_ps( maxv, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxv = _mm_max_ps( maxv, _mm_perm_ps( maxv, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
if( windowSpace )
{
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minv = _mm_madd_ps( minv, vector_float_half, vector_float_half );
maxv = _mm_madd_ps( maxv, vector_float_half, vector_float_half );
#endif
minv = _mm_max_ps( minv, vector_float_zero );
maxv = _mm_min_ps( maxv, vector_float_one );
}
_mm_store_ss( & min, minv );
_mm_store_ss( & max, maxv );
}
/*
========================
idRenderMatrix::DepthBoundsForExtrudedBounds
Calculates the depth bounds of the given extruded bounding box projected with the given Model View Projection (MVP) matrix.
The given bounding box is extruded in the 'extrudeDirection' up to the 'clipPlane'.
If 'windowSpace' is true then the calculated depth bounds are moved and clamped to the [0, 1] range.
The extruded bounding box is not clipped to the MVP so the depth bounds may not be as tight as possible.
========================
*/
void idRenderMatrix::DepthBoundsForExtrudedBounds( float& min, float& max, const idRenderMatrix& mvp, const idBounds& bounds, const idVec3& extrudeDirection, const idPlane& clipPlane, bool windowSpace )
{
assert( idMath::Fabs( extrudeDirection * clipPlane.Normal() ) >= idMath::FLT_SMALLEST_NON_DENORMAL );
__m128 mvp2 = _mm_loadu_ps( mvp[2] );
__m128 mvp3 = _mm_loadu_ps( mvp[3] );
__m128 b0 = _mm_loadu_bounds_0( bounds );
__m128 b1 = _mm_loadu_bounds_1( bounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
__m128 minv;
__m128 maxv;
// calculate the min/max depth values for the bounding box corners
{
// compute four partial Z,W values
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
__m128 z0 = _mm_madd_ps( vz0, _mm_splat_ps( mvp2, 2 ), parz );
__m128 w0 = _mm_madd_ps( vz0, _mm_splat_ps( mvp3, 2 ), parw );
__m128 z1 = _mm_madd_ps( vz1, _mm_splat_ps( mvp2, 2 ), parz );
__m128 w1 = _mm_madd_ps( vz1, _mm_splat_ps( mvp3, 2 ), parw );
__m128 s0 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w0 );
w0 = _mm_or_ps( w0, _mm_and_ps( vector_float_smallest_non_denorm, s0 ) );
__m128 rw0 = _mm_rcp32_ps( w0 );
z0 = _mm_mul_ps( z0, rw0 );
z0 = _mm_sel_ps( z0, vector_float_neg_infinity, s0 );
__m128 s1 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w1 );
w1 = _mm_or_ps( w1, _mm_and_ps( vector_float_smallest_non_denorm, s1 ) );
__m128 rw1 = _mm_rcp32_ps( w1 );
z1 = _mm_mul_ps( z1, rw1 );
z1 = _mm_sel_ps( z1, vector_float_neg_infinity, s1 );
minv = _mm_min_ps( z0, z1 );
maxv = _mm_max_ps( z0, z1 );
}
// calculate and include the min/max depth value for the extruded bounding box corners
{
__m128 clipX = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 0 ), 0 );
__m128 clipY = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 1 ), 0 );
__m128 clipZ = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 2 ), 0 );
__m128 clipW = _mm_splat_ps( _mm_load_ss( clipPlane.ToFloatPtr() + 3 ), 0 );
__m128 extrudeX = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 0 ), 0 );
__m128 extrudeY = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 1 ), 0 );
__m128 extrudeZ = _mm_splat_ps( _mm_load_ss( extrudeDirection.ToFloatPtr() + 2 ), 0 );
__m128 closing = _mm_madd_ps( clipX, extrudeX, _mm_madd_ps( clipY, extrudeY, _mm_mul_ps( clipZ, extrudeZ ) ) );
__m128 invClosing = _mm_rcp32_ps( closing );
invClosing = _mm_xor_ps( invClosing, vector_float_sign_bit );
__m128 dt = _mm_madd_ps( clipX, vx, _mm_madd_ps( clipY, vy, clipW ) );
__m128 d0 = _mm_madd_ps( clipZ, vz0, dt );
__m128 d1 = _mm_madd_ps( clipZ, vz1, dt );
d0 = _mm_mul_ps( d0, invClosing );
d1 = _mm_mul_ps( d1, invClosing );
__m128 vx0 = _mm_madd_ps( extrudeX, d0, vx );
__m128 vx1 = _mm_madd_ps( extrudeX, d1, vx );
__m128 vy0 = _mm_madd_ps( extrudeY, d0, vy );
__m128 vy1 = _mm_madd_ps( extrudeY, d1, vy );
vz0 = _mm_madd_ps( extrudeZ, d0, vz0 );
vz1 = _mm_madd_ps( extrudeZ, d1, vz1 );
__m128 mvp2X = _mm_splat_ps( mvp2, 0 );
__m128 mvp3X = _mm_splat_ps( mvp3, 0 );
__m128 mvp2W = _mm_splat_ps( mvp2, 3 );
__m128 mvp3W = _mm_splat_ps( mvp3, 3 );
__m128 z0 = _mm_madd_ps( vx0, mvp2X, mvp2W );
__m128 w0 = _mm_madd_ps( vx0, mvp3X, mvp3W );
__m128 z1 = _mm_madd_ps( vx1, mvp2X, mvp2W );
__m128 w1 = _mm_madd_ps( vx1, mvp3X, mvp3W );
__m128 mvp2Y = _mm_splat_ps( mvp2, 1 );
__m128 mvp3Y = _mm_splat_ps( mvp3, 1 );
z0 = _mm_madd_ps( vy0, mvp2Y, z0 );
w0 = _mm_madd_ps( vy0, mvp3Y, w0 );
z1 = _mm_madd_ps( vy1, mvp2Y, z1 );
w1 = _mm_madd_ps( vy1, mvp3Y, w1 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
z0 = _mm_madd_ps( vz0, mvp2Z, z0 );
w0 = _mm_madd_ps( vz0, mvp3Z, w0 );
z1 = _mm_madd_ps( vz1, mvp2Z, z1 );
w1 = _mm_madd_ps( vz1, mvp3Z, w1 );
__m128 s0 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w0 );
w0 = _mm_or_ps( w0, _mm_and_ps( vector_float_smallest_non_denorm, s0 ) );
__m128 rw0 = _mm_rcp32_ps( w0 );
z0 = _mm_mul_ps( z0, rw0 );
z0 = _mm_sel_ps( z0, vector_float_neg_infinity, s0 );
__m128 s1 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w1 );
w1 = _mm_or_ps( w1, _mm_and_ps( vector_float_smallest_non_denorm, s1 ) );
__m128 rw1 = _mm_rcp32_ps( w1 );
z1 = _mm_mul_ps( z1, rw1 );
z1 = _mm_sel_ps( z1, vector_float_neg_infinity, s1 );
minv = _mm_min_ps( minv, z0 );
maxv = _mm_max_ps( maxv, z0 );
minv = _mm_min_ps( minv, z1 );
maxv = _mm_max_ps( maxv, z1 );
}
minv = _mm_min_ps( minv, _mm_perm_ps( minv, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minv = _mm_min_ps( minv, _mm_perm_ps( minv, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxv = _mm_max_ps( maxv, _mm_perm_ps( maxv, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxv = _mm_max_ps( maxv, _mm_perm_ps( maxv, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
if( windowSpace )
{
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minv = _mm_madd_ps( minv, vector_float_half, vector_float_half );
maxv = _mm_madd_ps( maxv, vector_float_half, vector_float_half );
#endif
minv = _mm_max_ps( minv, vector_float_zero );
maxv = _mm_min_ps( maxv, vector_float_one );
}
_mm_store_ss( & min, minv );
_mm_store_ss( & max, maxv );
}
/*
========================
PointInsideInfiniteShadow
Returns true if the 'localPoint' is inside the infinite shadow volume cast
from the given occluder bounding box and the given light position.
========================
*/
static bool PointInsideInfiniteShadow( const idBounds& occluderBounds, const idVec3& localLightOrigin, const idVec3& localPoint, const float epsilon )
{
// Expand the bounds with an epsilon.
const idBounds expandedBounds = occluderBounds.Expand( epsilon );
// If the point is inside the bounding box then the point
// is also inside the shadow projection.
if( expandedBounds.ContainsPoint( localPoint ) )
{
return true;
}
// If the light is inside the bounding box then the shadow is projected
// in all directions and any point is inside the infinte shadow projection.
if( expandedBounds.ContainsPoint( localPoint ) )
{
return true;
}
// If the line from localLightOrigin to localPoint intersects the
// bounding box then the point is inside the infinite shadow projection.
if( expandedBounds.LineIntersection( localLightOrigin, localPoint ) )
{
return true;
}
// The point is definitely not inside the projected shadow.
return false;
}
/*
========================
idRenderMatrix::DepthBoundsForShadowBounds
Calculates the depth bounds of the infinite shadow volume projected with the given Model View Projection (MVP) matrix.
The infinite shadow volume is cast from the given occluder bounding box and the given light position.
If 'windowSpace' is true then the calculated depth bounds are moved and clamped to the [0, 1] range.
The infinite shadow volume is fully clipped to the MVP to get the tightest possible bounds.
Note that this code assumes the MVP matrix has an infinite far clipping plane. When the far plane is at
infinity the shadow volume is never far clipped and it is sufficient to test whether or not the center
of the near clip plane is inside the shadow volume to calculate the correct minimum Z. If the far plane
is not at infinity then this code would also have to test for the view frustum being completely contained
inside the shadow volume to also calculate the correct maximum Z. This could be done, for instance, by
testing if the center of the far clipping plane is contained inside the shadow volume.
========================
*/
void idRenderMatrix::DepthBoundsForShadowBounds( float& min, float& max, const idRenderMatrix& mvp, const idBounds& bounds, const idVec3& localLightOrigin, bool windowSpace )
{
const __m128 mvp0 = _mm_loadu_ps( mvp[0] );
const __m128 mvp1 = _mm_loadu_ps( mvp[1] );
const __m128 mvp2 = _mm_loadu_ps( mvp[2] );
const __m128 mvp3 = _mm_loadu_ps( mvp[3] );
const __m128 t0 = _mm_unpacklo_ps( mvp0, mvp2 ); // mvp[0][0], mvp[2][0], mvp[0][1], mvp[2][1]
const __m128 t1 = _mm_unpackhi_ps( mvp0, mvp2 ); // mvp[0][2], mvp[2][2], mvp[0][3], mvp[2][3]
const __m128 t2 = _mm_unpacklo_ps( mvp1, mvp3 ); // mvp[1][0], mvp[3][0], mvp[1][1], mvp[3][1]
const __m128 t3 = _mm_unpackhi_ps( mvp1, mvp3 ); // mvp[1][2], mvp[3][2], mvp[1][3], mvp[3][3]
const __m128 mvpX = _mm_unpacklo_ps( t0, t2 ); // mvp[0][0], mvp[1][0], mvp[2][0], mvp[3][0]
const __m128 mvpY = _mm_unpackhi_ps( t0, t2 ); // mvp[0][1], mvp[1][1], mvp[2][1], mvp[3][1]
const __m128 mvpZ = _mm_unpacklo_ps( t1, t3 ); // mvp[0][2], mvp[1][2], mvp[2][2], mvp[3][2]
const __m128 mvpW = _mm_unpackhi_ps( t1, t3 ); // mvp[0][3], mvp[1][3], mvp[2][3], mvp[3][3]
const __m128 b0 = _mm_loadu_bounds_0( bounds );
const __m128 b1 = _mm_loadu_bounds_1( bounds );
const __m128 lightOriginX = _mm_load_ss( localLightOrigin.ToFloatPtr() + 0 );
const __m128 lightOriginY = _mm_load_ss( localLightOrigin.ToFloatPtr() + 1 );
const __m128 lightOriginZ = _mm_load_ss( localLightOrigin.ToFloatPtr() + 2 );
const __m128 lightOrigin = _mm_unpacklo_ps( _mm_unpacklo_ps( lightOriginX, lightOriginZ ), lightOriginY );
// calculate the front facing polygon bits
int frontBits = GetBoxFrontBits_SSE2( b0, b1, lightOrigin );
const __m128 b0X = _mm_splat_ps( b0, 0 );
const __m128 b0Y = _mm_splat_ps( b0, 1 );
const __m128 b0Z = _mm_splat_ps( b0, 2 );
const __m128 b1X = _mm_splat_ps( b1, 0 );
const __m128 b1Y = _mm_splat_ps( b1, 1 );
const __m128 b1Z = _mm_splat_ps( b1, 2 );
// bounding box corners
const __m128 np0 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 np1 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 np2 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 np3 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b0Z, mvpZ, mvpW ) ) );
const __m128 np4 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 np5 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b0Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 np6 = _mm_madd_ps( b1X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
const __m128 np7 = _mm_madd_ps( b0X, mvpX, _mm_madd_ps( b1Y, mvpY, _mm_madd_ps( b1Z, mvpZ, mvpW ) ) );
ALIGNTYPE16 idVec4 projectedNearPoints[8];
_mm_store_ps( projectedNearPoints[0].ToFloatPtr(), np0 );
_mm_store_ps( projectedNearPoints[1].ToFloatPtr(), np1 );
_mm_store_ps( projectedNearPoints[2].ToFloatPtr(), np2 );
_mm_store_ps( projectedNearPoints[3].ToFloatPtr(), np3 );
_mm_store_ps( projectedNearPoints[4].ToFloatPtr(), np4 );
_mm_store_ps( projectedNearPoints[5].ToFloatPtr(), np5 );
_mm_store_ps( projectedNearPoints[6].ToFloatPtr(), np6 );
_mm_store_ps( projectedNearPoints[7].ToFloatPtr(), np7 );
// subtract the light position from the bounding box
const __m128 lightX = _mm_splat_ps( lightOriginX, 0 );
const __m128 lightY = _mm_splat_ps( lightOriginY, 0 );
const __m128 lightZ = _mm_splat_ps( lightOriginZ, 0 );
const __m128 d0X = _mm_sub_ps( b0X, lightX );
const __m128 d0Y = _mm_sub_ps( b0Y, lightY );
const __m128 d0Z = _mm_sub_ps( b0Z, lightZ );
const __m128 d1X = _mm_sub_ps( b1X, lightX );
const __m128 d1Y = _mm_sub_ps( b1Y, lightY );
const __m128 d1Z = _mm_sub_ps( b1Z, lightZ );
// bounding box corners projected to infinity from the light position
const __m128 fp0 = _mm_madd_ps( d0X, mvpX, _mm_madd_ps( d0Y, mvpY, _mm_mul_ps( d0Z, mvpZ ) ) );
const __m128 fp1 = _mm_madd_ps( d1X, mvpX, _mm_madd_ps( d0Y, mvpY, _mm_mul_ps( d0Z, mvpZ ) ) );
const __m128 fp2 = _mm_madd_ps( d1X, mvpX, _mm_madd_ps( d1Y, mvpY, _mm_mul_ps( d0Z, mvpZ ) ) );
const __m128 fp3 = _mm_madd_ps( d0X, mvpX, _mm_madd_ps( d1Y, mvpY, _mm_mul_ps( d0Z, mvpZ ) ) );
const __m128 fp4 = _mm_madd_ps( d0X, mvpX, _mm_madd_ps( d0Y, mvpY, _mm_mul_ps( d1Z, mvpZ ) ) );
const __m128 fp5 = _mm_madd_ps( d1X, mvpX, _mm_madd_ps( d0Y, mvpY, _mm_mul_ps( d1Z, mvpZ ) ) );
const __m128 fp6 = _mm_madd_ps( d1X, mvpX, _mm_madd_ps( d1Y, mvpY, _mm_mul_ps( d1Z, mvpZ ) ) );
const __m128 fp7 = _mm_madd_ps( d0X, mvpX, _mm_madd_ps( d1Y, mvpY, _mm_mul_ps( d1Z, mvpZ ) ) );
ALIGNTYPE16 idVec4 projectedFarPoints[8];
_mm_store_ps( projectedFarPoints[0].ToFloatPtr(), fp0 );
_mm_store_ps( projectedFarPoints[1].ToFloatPtr(), fp1 );
_mm_store_ps( projectedFarPoints[2].ToFloatPtr(), fp2 );
_mm_store_ps( projectedFarPoints[3].ToFloatPtr(), fp3 );
_mm_store_ps( projectedFarPoints[4].ToFloatPtr(), fp4 );
_mm_store_ps( projectedFarPoints[5].ToFloatPtr(), fp5 );
_mm_store_ps( projectedFarPoints[6].ToFloatPtr(), fp6 );
_mm_store_ps( projectedFarPoints[7].ToFloatPtr(), fp7 );
ALIGNTYPE16 idVec4 clippedPoints[( 6 + 12 ) * 16];
int numClippedPoints = 0;
// clip the front facing bounding box polygons at the near cap
const frontPolygons_t& frontPolygons = boxFrontPolygonsForFrontBits[frontBits];
for( int i = 0; i < frontPolygons.count; i++ )
{
const int polygon = frontPolygons.indices[i];
_mm_store_ps( clippedPoints[numClippedPoints + 0].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxPolygonVertices[polygon][0]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 1].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxPolygonVertices[polygon][1]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 2].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxPolygonVertices[polygon][2]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 3].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxPolygonVertices[polygon][3]].ToFloatPtr() ) );
numClippedPoints += ClipHomogeneousPolygonToUnitCube_SSE2( &clippedPoints[numClippedPoints], 4 );
}
// clip the front facing bounding box polygons projected to the far cap
for( int i = 0; i < frontPolygons.count; i++ )
{
const int polygon = frontPolygons.indices[i];
_mm_store_ps( clippedPoints[numClippedPoints + 0].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxPolygonVertices[polygon][0]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 1].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxPolygonVertices[polygon][1]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 2].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxPolygonVertices[polygon][2]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 3].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxPolygonVertices[polygon][3]].ToFloatPtr() ) );
numClippedPoints += ClipHomogeneousPolygonToUnitCube_SSE2( &clippedPoints[numClippedPoints], 4 );
}
// clip the silhouette edge polygons that stretch to infinity
const silhouetteEdges_t& silhouetteEdges = boxSilhouetteEdgesForFrontBits[frontBits];
for( int i = 0; i < silhouetteEdges.count; i++ )
{
const int edge = silhouetteEdges.indices[i];
_mm_store_ps( clippedPoints[numClippedPoints + 0].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxEdgeVertices[edge][0]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 1].ToFloatPtr(), _mm_load_ps( projectedNearPoints[boxEdgeVertices[edge][1]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 2].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxEdgeVertices[edge][1]].ToFloatPtr() ) );
_mm_store_ps( clippedPoints[numClippedPoints + 3].ToFloatPtr(), _mm_load_ps( projectedFarPoints[boxEdgeVertices[edge][0]].ToFloatPtr() ) );
numClippedPoints += ClipHomogeneousPolygonToUnitCube_SSE2( &clippedPoints[numClippedPoints], 4 );
}
// repeat the first clipped point at the end to get a multiple of 4 clipped points
const __m128 point0 = _mm_load_ps( clippedPoints[0].ToFloatPtr() );
for( int i = numClippedPoints; ( i & 3 ) != 0; i++ )
{
_mm_store_ps( clippedPoints[i].ToFloatPtr(), point0 );
}
// test if the center of the near clip plane is inside the infinite shadow volume
const idVec3 localNearClipCenter = LocalNearClipCenterFromMVP( mvp );
const bool inside = PointInsideInfiniteShadow( bounds, localLightOrigin, localNearClipCenter, RENDER_MATRIX_PROJECTION_EPSILON );
__m128 minZ = inside ? vector_float_neg_one : vector_float_pos_infinity;
__m128 maxZ = vector_float_neg_infinity;
for( int i = 0; i < numClippedPoints; i += 4 )
{
const __m128 cp0 = _mm_load_ps( clippedPoints[i + 0].ToFloatPtr() );
const __m128 cp1 = _mm_load_ps( clippedPoints[i + 1].ToFloatPtr() );
const __m128 cp2 = _mm_load_ps( clippedPoints[i + 2].ToFloatPtr() );
const __m128 cp3 = _mm_load_ps( clippedPoints[i + 3].ToFloatPtr() );
const __m128 s1 = _mm_unpackhi_ps( cp0, cp2 ); // cp0[2], cp2[2], cp0[3], cp2[3]
const __m128 s3 = _mm_unpackhi_ps( cp1, cp3 ); // cp1[2], cp3[2], cp1[3], cp3[3]
const __m128 cpZ = _mm_unpacklo_ps( s1, s3 ); // cp0[2], cp1[2], cp2[2], cp3[2]
const __m128 cpW = _mm_unpackhi_ps( s1, s3 ); // cp0[3], cp1[3], cp2[3], cp3[3]
const __m128 rW = _mm_rcp32_ps( cpW );
const __m128 rZ = _mm_mul_ps( cpZ, rW );
minZ = _mm_min_ps( minZ, rZ );
maxZ = _mm_max_ps( maxZ, rZ );
}
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
minZ = _mm_min_ps( minZ, _mm_perm_ps( minZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 1, 0, 3, 2 ) ) );
maxZ = _mm_max_ps( maxZ, _mm_perm_ps( maxZ, _MM_SHUFFLE( 2, 3, 0, 1 ) ) );
if( windowSpace )
{
#if !defined( CLIP_SPACE_D3D ) // the D3D clip space Z is already in the range [0,1]
minZ = _mm_madd_ps( minZ, vector_float_half, vector_float_half );
maxZ = _mm_madd_ps( maxZ, vector_float_half, vector_float_half );
#endif
minZ = _mm_max_ps( _mm_min_ps( minZ, vector_float_one ), vector_float_zero );
maxZ = _mm_max_ps( _mm_min_ps( maxZ, vector_float_one ), vector_float_zero );
}
_mm_store_ss( & min, minZ );
_mm_store_ss( & max, maxZ );
}
/*
========================
idRenderMatrix::GetFrustumPlanes
Normally the clip space extends from -1.0 to 1.0 on each axis, but by setting 'zeroToOne'
to true, the clip space will extend from 0.0 to 1.0 on each axis for a light projection matrix.
========================
*/
void idRenderMatrix::GetFrustumPlanes( idPlane planes[6], const idRenderMatrix& frustum, bool zeroToOne, bool normalize )
{
// FIXME: need to know whether or not this is a D3D MVP.
// We cannot just assume that it's an D3D MVP matrix when
// zeroToOne = false and CLIP_SPACE_D3D is defined because
// this code may be called for non-MVP matrices.
const bool isZeroOneZ = false;
if( zeroToOne )
{
// left: inside(p) = p * frustum[0] > 0
planes[0][0] = frustum[0][0];
planes[0][1] = frustum[0][1];
planes[0][2] = frustum[0][2];
planes[0][3] = frustum[0][3];
// bottom: inside(p) = p * frustum[1] > 0
planes[2][0] = frustum[1][0];
planes[2][1] = frustum[1][1];
planes[2][2] = frustum[1][2];
planes[2][3] = frustum[1][3];
// near: inside(p) = p * frustum[2] > 0
planes[4][0] = frustum[2][0];
planes[4][1] = frustum[2][1];
planes[4][2] = frustum[2][2];
planes[4][3] = frustum[2][3];
}
else
{
// left: inside(p) = p * frustum[0] > - ( p * frustum[3] )
planes[0][0] = frustum[3][0] + frustum[0][0];
planes[0][1] = frustum[3][1] + frustum[0][1];
planes[0][2] = frustum[3][2] + frustum[0][2];
planes[0][3] = frustum[3][3] + frustum[0][3];
// bottom: inside(p) = p * frustum[1] > -( p * frustum[3] )
planes[2][0] = frustum[3][0] + frustum[1][0];
planes[2][1] = frustum[3][1] + frustum[1][1];
planes[2][2] = frustum[3][2] + frustum[1][2];
planes[2][3] = frustum[3][3] + frustum[1][3];
// near: inside(p) = p * frustum[2] > -( p * frustum[3] )
planes[4][0] = isZeroOneZ ? ( frustum[2][0] ) : ( frustum[3][0] + frustum[2][0] );
planes[4][1] = isZeroOneZ ? ( frustum[2][1] ) : ( frustum[3][1] + frustum[2][1] );
planes[4][2] = isZeroOneZ ? ( frustum[2][2] ) : ( frustum[3][2] + frustum[2][2] );
planes[4][3] = isZeroOneZ ? ( frustum[2][3] ) : ( frustum[3][3] + frustum[2][3] );
}
// right: inside(p) = p * frustum[0] < p * frustum[3]
planes[1][0] = frustum[3][0] - frustum[0][0];
planes[1][1] = frustum[3][1] - frustum[0][1];
planes[1][2] = frustum[3][2] - frustum[0][2];
planes[1][3] = frustum[3][3] - frustum[0][3];
// top: inside(p) = p * frustum[1] < p * frustum[3]
planes[3][0] = frustum[3][0] - frustum[1][0];
planes[3][1] = frustum[3][1] - frustum[1][1];
planes[3][2] = frustum[3][2] - frustum[1][2];
planes[3][3] = frustum[3][3] - frustum[1][3];
// far: inside(p) = p * frustum[2] < p * frustum[3]
planes[5][0] = frustum[3][0] - frustum[2][0];
planes[5][1] = frustum[3][1] - frustum[2][1];
planes[5][2] = frustum[3][2] - frustum[2][2];
planes[5][3] = frustum[3][3] - frustum[2][3];
// optionally normalize the planes
if( normalize )
{
for( int i = 0; i < 6; i++ )
{
float s = idMath::InvSqrt( planes[i].Normal().LengthSqr() );
planes[i][0] *= s;
planes[i][1] *= s;
planes[i][2] *= s;
planes[i][3] *= s;
}
}
}
/*
========================
idRenderMatrix::GetFrustumCorners
========================
*/
void idRenderMatrix::GetFrustumCorners( frustumCorners_t& corners, const idRenderMatrix& frustumTransform, const idBounds& frustumBounds )
{
assert_16_byte_aligned( &corners );
__m128 mvp0 = _mm_loadu_ps( frustumTransform[0] );
__m128 mvp1 = _mm_loadu_ps( frustumTransform[1] );
__m128 mvp2 = _mm_loadu_ps( frustumTransform[2] );
__m128 mvp3 = _mm_loadu_ps( frustumTransform[3] );
__m128 b0 = _mm_loadu_bounds_0( frustumBounds );
__m128 b1 = _mm_loadu_bounds_1( frustumBounds );
// take the four points on the X-Y plane
__m128 vxy = _mm_unpacklo_ps( b0, b1 ); // min X, max X, min Y, max Y
__m128 vx = _mm_perm_ps( vxy, _MM_SHUFFLE( 1, 0, 1, 0 ) ); // min X, max X, min X, max X
__m128 vy = _mm_perm_ps( vxy, _MM_SHUFFLE( 3, 3, 2, 2 ) ); // min Y, min Y, max Y, max Y
__m128 vz0 = _mm_splat_ps( b0, 2 ); // min Z, min Z, min Z, min Z
__m128 vz1 = _mm_splat_ps( b1, 2 ); // max Z, max Z, max Z, max Z
// compute four partial X,Y,Z,W values
__m128 parx = _mm_splat_ps( mvp0, 3 );
__m128 pary = _mm_splat_ps( mvp1, 3 );
__m128 parz = _mm_splat_ps( mvp2, 3 );
__m128 parw = _mm_splat_ps( mvp3, 3 );
parx = _mm_madd_ps( vx, _mm_splat_ps( mvp0, 0 ), parx );
pary = _mm_madd_ps( vx, _mm_splat_ps( mvp1, 0 ), pary );
parz = _mm_madd_ps( vx, _mm_splat_ps( mvp2, 0 ), parz );
parw = _mm_madd_ps( vx, _mm_splat_ps( mvp3, 0 ), parw );
parx = _mm_madd_ps( vy, _mm_splat_ps( mvp0, 1 ), parx );
pary = _mm_madd_ps( vy, _mm_splat_ps( mvp1, 1 ), pary );
parz = _mm_madd_ps( vy, _mm_splat_ps( mvp2, 1 ), parz );
parw = _mm_madd_ps( vy, _mm_splat_ps( mvp3, 1 ), parw );
__m128 mvp0Z = _mm_splat_ps( mvp0, 2 );
__m128 mvp1Z = _mm_splat_ps( mvp1, 2 );
__m128 mvp2Z = _mm_splat_ps( mvp2, 2 );
__m128 mvp3Z = _mm_splat_ps( mvp3, 2 );
__m128 x0 = _mm_madd_ps( vz0, mvp0Z, parx );
__m128 y0 = _mm_madd_ps( vz0, mvp1Z, pary );
__m128 z0 = _mm_madd_ps( vz0, mvp2Z, parz );
__m128 w0 = _mm_madd_ps( vz0, mvp3Z, parw );
__m128 x1 = _mm_madd_ps( vz1, mvp0Z, parx );
__m128 y1 = _mm_madd_ps( vz1, mvp1Z, pary );
__m128 z1 = _mm_madd_ps( vz1, mvp2Z, parz );
__m128 w1 = _mm_madd_ps( vz1, mvp3Z, parw );
__m128 s0 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w0 );
__m128 s1 = _mm_cmpgt_ps( vector_float_smallest_non_denorm, w1 );
w0 = _mm_sel_ps( w0, vector_float_one, s0 );
w1 = _mm_sel_ps( w1, vector_float_one, s1 );
__m128 rw0 = _mm_rcp32_ps( w0 );
__m128 rw1 = _mm_rcp32_ps( w1 );
x0 = _mm_mul_ps( x0, rw0 );
y0 = _mm_mul_ps( y0, rw0 );
z0 = _mm_mul_ps( z0, rw0 );
x1 = _mm_mul_ps( x1, rw1 );
y1 = _mm_mul_ps( y1, rw1 );
z1 = _mm_mul_ps( z1, rw1 );
_mm_store_ps( corners.x + 0, x0 );
_mm_store_ps( corners.x + 4, x1 );
_mm_store_ps( corners.y + 0, y0 );
_mm_store_ps( corners.y + 4, y1 );
_mm_store_ps( corners.z + 0, z0 );
_mm_store_ps( corners.z + 4, z1 );
}
/*
========================
idRenderMatrix::CullFrustumCornersToPlane
========================
*/
frustumCull_t idRenderMatrix::CullFrustumCornersToPlane( const frustumCorners_t& corners, const idPlane& plane )
{
assert_16_byte_aligned( &corners );
__m128 vp = _mm_loadu_ps( plane.ToFloatPtr() );
__m128 x0 = _mm_load_ps( corners.x + 0 );
__m128 y0 = _mm_load_ps( corners.y + 0 );
__m128 z0 = _mm_load_ps( corners.z + 0 );
__m128 x1 = _mm_load_ps( corners.x + 4 );
__m128 y1 = _mm_load_ps( corners.y + 4 );
__m128 z1 = _mm_load_ps( corners.z + 4 );
__m128 p0 = _mm_splat_ps( vp, 0 );
__m128 p1 = _mm_splat_ps( vp, 1 );
__m128 p2 = _mm_splat_ps( vp, 2 );
__m128 p3 = _mm_splat_ps( vp, 3 );
__m128 d0 = _mm_madd_ps( x0, p0, _mm_madd_ps( y0, p1, _mm_madd_ps( z0, p2, p3 ) ) );
__m128 d1 = _mm_madd_ps( x1, p0, _mm_madd_ps( y1, p1, _mm_madd_ps( z1, p2, p3 ) ) );
int b0 = _mm_movemask_ps( d0 );
int b1 = _mm_movemask_ps( d1 );
unsigned int front = ( ( unsigned int ) - ( ( b0 & b1 ) ^ 15 ) ) >> 31;
unsigned int back = ( ( unsigned int ) - ( b0 | b1 ) ) >> 31;
compile_time_assert( FRUSTUM_CULL_FRONT == 1 );
compile_time_assert( FRUSTUM_CULL_BACK == 2 );
compile_time_assert( FRUSTUM_CULL_CROSS == 3 );
return ( frustumCull_t )( front | ( back << 1 ) );
}
Reference in a new issue View git blame Copy permalink