Benchmark: Linear Blend Skinning 2

Linear Blend Skinning 2 (version: 1) Benchmark.js

Comparing performance of: Warmup vs Original vs UpdatedEmscripten vs LoicRewrite vs SubarrayOnTrans vs SubarrayOnPos vs SubarrayOnTransAndPos vs SubarrayOnTransAndPosWithFRound vs SubarrayOnTransAndPosWithoutPlus vs PosInOutOfTheLoopF32Array vs PosInOutOfTheLoopF32ArrayFromSubarray vs PosInOutOfTheLoopVars vs CacheTransWithSubarray vs CacheTransWithSlice vs CacheTransWithF32ArraySetFromSubarray vs CacheTransWithVarsFromSubarray vs CacheTransWithArrayFromSubarray vs SubarrayOnPosAndCacheTransWithSubarray vs SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData vs SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans vs SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans vs F64PosAndF64Trans vs F64PosAndF64TransAndF64Res vs I32Joint vs VarInsteadOfLetAndConst

Created: 2 years ago by: Registered User

Go to the latest result

Script Preparation code:

debugger;
var position_count = 15000;
var joint_count = 30;
var max_number_of_consecutive_joints = 15;
var vertex_dim = 3;

var MEM_SIZE = 100*1024*1024;
var MEM_STRIDE = 2*1024*1024;

var MEM = new ArrayBuffer(MEM_SIZE);
var HEAP32 = new Int32Array(MEM);
var HEAPU32 = new Uint32Array(MEM);
var HEAPF32 = new Float32Array(MEM);
var HEAPF64 = new Float64Array(MEM);
var MEM_HEAD = 0;

function MEM_ALLOC(len) {
    let cur = MEM_HEAD;
    len += MEM_STRIDE;
    MEM_HEAD += len + (16 - (len % 16));
    return cur;
}

function splitmix32(a) {
    return function() {
      a |= 0; a = a + 0x9e3779b9 | 0;
      var t = a ^ a >>> 16; t = Math.imul(t, 0x21f0aaad);
          t = t ^ t >>> 15; t = Math.imul(t, 0x735a2d97);
      return ((t = t ^ t >>> 15) >>> 0) / 4294967296;
    }
}

Math_random = splitmix32(0);
Math_imul = Math.imul;
Math_fround = Math.fround;

if (typeof window !== 'undefined' && window.navigator && window.navigator.wakeLock)
    window.navigator.wakeLock.request('screen');

var positions = MEM_ALLOC(position_count * 3 * 4);

for (let i = 0; i < position_count; ++i) {
    const p = (positions >> 2) + i * 3;
    HEAPF32[p] = Math_random() * 100;
    HEAPF32[p + 1] = Math_random() * 100;
    HEAPF32[p + 2] = Math_random() * 100;
}

var repeat_last_joint = -1;
var last_joint;
function get_joint() {
    if (repeat_last_joint < 0) {
        repeat_last_joint = (Math_random() * max_number_of_consecutive_joints) |0;
        last_joint = Math_random() * (joint_count - 1) | 0;
    }
    repeat_last_joint -= 1;
    return last_joint;
}

var joints = MEM_ALLOC(position_count * 4);

for (let i = 0; i < position_count; ++i) {
    const p = (joints >> 2) + i;
    HEAPU32[p] = get_joint();
}

var transformations = MEM_ALLOC(joint_count * 16 * 4);

for (let i = 0; i < joint_count; ++i) {
    const p = (transformations >> 2) + i * 16;
    for (let j = 0; j < 16; ++j)
        HEAPF32[p + j] = Math_random() * 100;
}

var res = MEM_ALLOC(position_count * 3 * 4);

var positions_and_joints = MEM_ALLOC(position_count * 4 * 4);

for (let i = 0; i < position_count; ++i) {
    const p = (positions_and_joints >> 2) + i * 4;
    HEAPU32[p] = get_joint();
    HEAPF32[p + 1] = Math_random() * 100;
    HEAPF32[p + 2] = Math_random() * 100;
    HEAPF32[p + 3] = Math_random() * 100;
}

var small_transformations = MEM_ALLOC(joint_count * 12 * 4);

for (let i = 0; i < joint_count; ++i) {
    const p = (small_transformations >> 2) + i * 12;
    for (let j = 0; j < 12; ++j)
        HEAPF32[p + j] = Math_random() * 100;
}

var positions_f64 = MEM_ALLOC(position_count * 3 * 8);

for (let i = 0; i < position_count; ++i) {
    const p = (positions_f64 >> 3) + i * 3;
    HEAPF64[p] = Math_random() * 100;
    HEAPF64[p + 1] = Math_random() * 100;
    HEAPF64[p + 2] = Math_random() * 100;
}

var transformations_f64 = MEM_ALLOC(joint_count * 16 * 8);

for (let i = 0; i < joint_count; ++i) {
    const p = (transformations_f64 >> 3) + i * 16;
    for (let j = 0; j < 16; ++j)
        HEAPF64[p + j] = Math_random() * 100;
}

var res_f64 = MEM_ALLOC(position_count * 3 * 8);

var vertexCount = position_count;

var positions_view = HEAPF32.subarray(positions >> 2, (positions >> 2) + position_count * 3);
var joints_view = HEAPU32.subarray(joints >> 2, (joints >> 2) + position_count);
var transformations_view = HEAPF32.subarray(transformations >> 2, (transformations >> 2) + joint_count * 16);
var res_view = HEAPF32.subarray(res >> 2, (res >> 2) + position_count * 3);
var positions_and_joints_uview = HEAPU32.subarray(positions_and_joints >> 2, (positions_and_joints >> 2) + position_count * 4);
var positions_and_joints_fview = HEAPF32.subarray(positions_and_joints >> 2, (positions_and_joints >> 2) + position_count * 4);
var small_transformations_view = HEAPF32.subarray(small_transformations >> 2, (small_transformations >> 2) + joint_count * 12);
var positions_f64_view = HEAPF64.subarray(positions_f64 >> 3, (positions_f64 >> 3) + position_count * 3);
var transformations_f64_view = HEAPF64.subarray(transformations_f64 >> 3, (transformations_f64 >> 3) + joint_count * 16);
debugger;

function Warmup() {
    let a = 0;
    for (let i = 0; i < 8000000; i++) {
        a += Math.exp(Math_random()) * Math.exp(Math_random()) * Math.exp(Math_random()) * Math.exp(Math_random());
    }
}

function void_20mat_mult_4_2c_204_2c_204_2c_204_2c_20float__28int_2c_20float_20_28__29_20_5b4_5d_2c_20float_20const_20_28__29_20_5b4_5d_2c_20float_20const_20_28__29_20_5b4_5d_5b4_5d_29($0, $1, $2, $3) {
 var $4 = 0, $5 = 0, $6 = 0;
 if (($0 | 0) >= 1) {
  while (1) {
   $4 = $6 << 4;
   $5 = $4 + $1 | 0;
   $4 = $2 + $4 | 0;
   HEAPF32[$5 >> 2] = Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$4 >> 2] * HEAPF32[$3 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$4 + 4 >> 2] * HEAPF32[$3 + 16 >> 2])) + Math_fround(HEAPF32[$4 + 8 >> 2] * HEAPF32[$3 + 32 >> 2])) + Math_fround(HEAPF32[$4 + 12 >> 2] * HEAPF32[$3 + 48 >> 2]);
   HEAPF32[$5 + 4 >> 2] = Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$4 >> 2] * HEAPF32[$3 + 4 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$4 + 4 >> 2] * HEAPF32[$3 + 20 >> 2])) + Math_fround(HEAPF32[$4 + 8 >> 2] * HEAPF32[$3 + 36 >> 2])) + Math_fround(HEAPF32[$4 + 12 >> 2] * HEAPF32[$3 + 52 >> 2]);
   HEAPF32[$5 + 8 >> 2] = Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$4 >> 2] * HEAPF32[$3 + 8 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$4 + 4 >> 2] * HEAPF32[$3 + 24 >> 2])) + Math_fround(HEAPF32[$4 + 8 >> 2] * HEAPF32[$3 + 40 >> 2])) + Math_fround(HEAPF32[$4 + 12 >> 2] * HEAPF32[$3 + 56 >> 2]);
   HEAPF32[$5 + 12 >> 2] = Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$4 >> 2] * HEAPF32[$3 + 12 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$4 + 4 >> 2] * HEAPF32[$3 + 28 >> 2])) + Math_fround(HEAPF32[$4 + 8 >> 2] * HEAPF32[$3 + 44 >> 2])) + Math_fround(HEAPF32[$4 + 12 >> 2] * HEAPF32[$3 + 60 >> 2]);
   $6 = $6 + 1 | 0;
   if (($6 | 0) != ($0 | 0)) {
    continue;
   }
   break;
  }
 }
}
   
function do_affine_28float__2c_20unsigned_20long_2c_20int_2c_20float__2c_20unsigned_20int__2c_20float_20_28__29_20_5b4_5d_5b4_5d_29($0, $1, $2, $3, $4, $5) {
 var $6 = 0, $7 = 0, $8 = 0, $9 = 0;
 if (($1 | 0) > 0) {
  $9 = $2 + -2 | 0;
  while (1) {
   label$3 : {
    if ($9 >>> 0 > 2) {
     break label$3;
    }
    $2 = HEAP32[($7 << 2) + $4 >> 2];
    label$4 : {
     switch ($9 - 1 | 0) {
     default:
      $6 = $7 << 3;
      $8 = $6 + $0 | 0;
      $2 = ($2 << 6) + $5 | 0;
      $6 = $3 + $6 | 0;
      HEAPF32[$8 >> 2] = HEAPF32[$2 + 48 >> 2] + Math_fround(HEAPF32[$2 + 32 >> 2] + Math_fround(Math_fround(Math_fround(HEAPF32[$6 >> 2] * HEAPF32[$2 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$6 + 4 >> 2] * HEAPF32[$2 + 16 >> 2])));
      HEAPF32[$8 + 4 >> 2] = HEAPF32[$2 + 52 >> 2] + Math_fround(HEAPF32[$2 + 36 >> 2] + Math_fround(Math_fround(Math_fround(HEAPF32[$6 >> 2] * HEAPF32[$2 + 4 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$6 + 4 >> 2] * HEAPF32[$2 + 20 >> 2])));
      break label$3;
     case 0:
      $6 = Math_imul($7, 12);
      $8 = $6 + $0 | 0;
      $2 = ($2 << 6) + $5 | 0;
      $6 = $3 + $6 | 0;
      HEAPF32[$8 >> 2] = HEAPF32[$2 + 48 >> 2] + Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$6 >> 2] * HEAPF32[$2 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$6 + 4 >> 2] * HEAPF32[$2 + 16 >> 2])) + Math_fround(HEAPF32[$6 + 8 >> 2] * HEAPF32[$2 + 32 >> 2]));
      HEAPF32[$8 + 4 >> 2] = HEAPF32[$2 + 52 >> 2] + Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$6 >> 2] * HEAPF32[$2 + 4 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$6 + 4 >> 2] * HEAPF32[$2 + 20 >> 2])) + Math_fround(HEAPF32[$6 + 8 >> 2] * HEAPF32[$2 + 36 >> 2]));
      HEAPF32[$8 + 8 >> 2] = HEAPF32[$2 + 56 >> 2] + Math_fround(Math_fround(Math_fround(Math_fround(HEAPF32[$6 >> 2] * HEAPF32[$2 + 8 >> 2]) + Math_fround(0)) + Math_fround(HEAPF32[$6 + 4 >> 2] * HEAPF32[$2 + 24 >> 2])) + Math_fround(HEAPF32[$6 + 8 >> 2] * HEAPF32[$2 + 40 >> 2]));
      break label$3;
     case 1:
      break label$4;
     }
    }
    $6 = $7 << 4;
    void_20mat_mult_4_2c_204_2c_204_2c_204_2c_20float__28int_2c_20float_20_28__29_20_5b4_5d_2c_20float_20const_20_28__29_20_5b4_5d_2c_20float_20const_20_28__29_20_5b4_5d_5b4_5d_29(1, $6 + $0 | 0, $3 + $6 | 0, ($2 << 6) + $5 | 0);
   }
   $7 = $7 + 1 | 0;
   if (($7 | 0) != ($1 | 0)) {
    continue;
   }
   break;
  }
 }
}

function Original() {
    do_affine_28float__2c_20unsigned_20long_2c_20int_2c_20float__2c_20unsigned_20int__2c_20float_20_28__29_20_5b4_5d_5b4_5d_29(res, vertexCount, vertex_dim, positions, joints, transformations);
}

function UpdatedEmscripten() {
 var $3 = 0, $9 = Math_fround(0), $10 = 0, $11 = 0, $14 = 0, $15 = 0, $17 = 0, $24 = 0, $25 = 0, $27 = 0, $28 = 0;
 var $5 = position_count, $7 = transformations, $4 = res, $2 = vertex_dim, $12 = positions, $16 = joints, $8 = 0;
 if (($5 | 0) > 0) {
  $17 = $2 - 2 | 0;
  while (1) {
   $14 = HEAP32[$16 + ($8 << 2) >> 2];
   label$594: {
    label$595: {
     switch ($17 | 0) {
     case 0:
      $2 = $8 << 3;
      $24 = $2 + $12 | 0;
      $25 = $2 + $4 | 0;
      $11 = 0;
      $2 = 1;
      while (1) {
       $9 = Math_fround(0);
       $10 = 1;
       $3 = 0;
       while (1) {
        $15 = $11 << 2;
        $27 = $7 + ($14 << 6) | 0;
        $9 = Math_fround(Math_fround(HEAPF32[$24 + ($3 << 2) >> 2] * HEAPF32[$15 + ($27 + ($3 << 4) | 0) >> 2]) + $9);
        $3 = 1;
        $28 = $10 & 1;
        $10 = 0;
        if ($28) {
         continue;
        }
        break;
       }
       $3 = $15 + $27 | 0;
       HEAPF32[$15 + $25 >> 2] = Math_fround($9 + HEAPF32[$3 + 32 >> 2]) + HEAPF32[$3 + 48 >> 2];
       $11 = 1;
       $3 = $2;
       $2 = 0;
       if ($3) {
        continue;
       }
       break;
      }
      ;
      break label$594;
     case 1:
      $2 = Math_imul($8, 12);
      $11 = $2 + $12 | 0;
      $15 = $2 + $4 | 0;
      $10 = 0;
      while (1) {
       $9 = Math_fround(0);
       $3 = 0;
       while (1) {
        $2 = $10 << 2;
        $24 = $7 + ($14 << 6) | 0;
        $9 = Math_fround(Math_fround(HEAPF32[$11 + ($3 << 2) >> 2] * HEAPF32[$2 + ($24 + ($3 << 4) | 0) >> 2]) + $9);
        $3 = $3 + 1 | 0;
        if (($3 | 0) != 3) {
         continue;
        }
        break;
       }
       HEAPF32[$2 + $15 >> 2] = $9 + HEAPF32[($2 + $24 | 0) + 48 >> 2];
       $10 = $10 + 1 | 0;
       if (($10 | 0) != 3) {
        continue;
       }
       break;
      }
      ;
      break label$594;
     case 2:
      break label$595;
     default:
      break label$594;
     }
    }
    $2 = $8 << 4;
    $11 = $2 + $12 | 0;
    $2 = $2 + $4 | 0;
    $10 = 0;
    while (1) {
     $9 = Math_fround(0);
     $3 = 0;
     while (1) {
      $15 = $10 << 2;
      $9 = Math_fround(Math_fround(HEAPF32[$11 + ($3 << 2) >> 2] * HEAPF32[$15 + (($7 + ($14 << 6) | 0) + ($3 << 4) | 0) >> 2]) + $9);
      $3 = $3 + 1 | 0;
      if (($3 | 0) != 4) {
       continue;
      }
      break;
     }
     HEAPF32[$2 + $15 >> 2] = $9;
     $10 = $10 + 1 | 0;
     if (($10 | 0) != 4) {
      continue;
     }
     break;
    }
   }
   $8 = $8 + 1 | 0;
   if (($5 | 0) != ($8 | 0)) {
    continue;
   }
   break;
  }
 }
 $10 = 15;
}

function LoicRewrite() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +HEAPF32[M        ] + P2 * +HEAPF32[M + 4 | 0] + P3 * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +HEAPF32[M + 1 | 0] + P2 * +HEAPF32[M + 5 | 0] + P3 * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +HEAPF32[M + 2 | 0] + P2 * +HEAPF32[M + 6 | 0] + P3 * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnTrans() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const MAT = HEAPF32.subarray(M, M + 16 | 0);
    
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +MAT[0] + P2 * +MAT[4] + P3 * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +MAT[1] + P2 * +MAT[5] + P3 * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +MAT[2] + P2 * +MAT[6] + P3 * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnPos() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +HEAPF32[M        ] + +POS[1] * +HEAPF32[M + 4 | 0] + +POS[2] * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +HEAPF32[M + 1 | 0] + +POS[1] * +HEAPF32[M + 5 | 0] + +POS[2] * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +HEAPF32[M + 2 | 0] + +POS[1] * +HEAPF32[M + 6 | 0] + +POS[2] * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnTransAndPos() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const MAT = HEAPF32.subarray(M, M + 16 | 0);
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +MAT[0] + +POS[1] * +MAT[4] + +POS[2] * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +MAT[1] + +POS[1] * +MAT[5] + +POS[2] * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +MAT[2] + +POS[1] * +MAT[6] + +POS[2] * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnTransAndPosWithFRound() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const MAT = HEAPF32.subarray(M, M + 16 | 0);
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = Math.fround(Math.fround(Math.fround(Math.fround(Math.fround(POS[0]) * Math.fround(MAT[0])) + Math.fround(Math.fround(POS[1]) * Math.fround(MAT[4]))) + Math.fround(Math.fround(POS[2]) * Math.fround(MAT[ 8]))) + Math.fround(MAT[12]));
        HEAPF32[resPosition_F32 + 1 | 0] = Math.fround(Math.fround(Math.fround(Math.fround(Math.fround(POS[0]) * Math.fround(MAT[1])) + Math.fround(Math.fround(POS[1]) * Math.fround(MAT[5]))) + Math.fround(Math.fround(POS[2]) * Math.fround(MAT[ 9]))) + Math.fround(MAT[13]));
        HEAPF32[resPosition_F32 + 2 | 0] = Math.fround(Math.fround(Math.fround(Math.fround(Math.fround(POS[0]) * Math.fround(MAT[2])) + Math.fround(Math.fround(POS[1]) * Math.fround(MAT[6]))) + Math.fround(Math.fround(POS[2]) * Math.fround(MAT[10]))) + Math.fround(MAT[14]));
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnTransAndPosWithoutPlus() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        const MAT = HEAPF32.subarray(M, M + 16 | 0);
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = POS[0] * MAT[0] + POS[1] * MAT[4] + POS[2] * MAT[ 8] + MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = POS[0] * MAT[1] + POS[1] * MAT[5] + POS[2] * MAT[ 9] + MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = POS[0] * MAT[2] + POS[1] * MAT[6] + POS[2] * MAT[10] + MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function PosInOutOfTheLoopF32Array() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let POS = new Float32Array(3);

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;

        POS[0] = +HEAPF32[position_F32];
        POS[1] = +HEAPF32[position_F32 + 1 | 0];
        POS[2] = +HEAPF32[position_F32 + 2 | 0];

        HEAPF32[resPosition_F32        ] = +POS[0] * +HEAPF32[M        ] + +POS[1] * +HEAPF32[M + 4 | 0] + +POS[2] * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +HEAPF32[M + 1 | 0] + +POS[1] * +HEAPF32[M + 5 | 0] + +POS[2] * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +HEAPF32[M + 2 | 0] + +POS[1] * +HEAPF32[M + 6 | 0] + +POS[2] * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function PosInOutOfTheLoopF32ArrayFromSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let POS = new Float32Array(3);

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        POS.set(HEAPF32.subarray(position_F32, position_F32 + 3 | 0));
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +HEAPF32[M        ] + +POS[1] * +HEAPF32[M + 4 | 0] + +POS[2] * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +HEAPF32[M + 1 | 0] + +POS[1] * +HEAPF32[M + 5 | 0] + +POS[2] * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +HEAPF32[M + 2 | 0] + +POS[1] * +HEAPF32[M + 6 | 0] + +POS[2] * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function PosInOutOfTheLoopVars() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let P1, P2, P3;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F32 + (J << 4) | 0;
        P1 = +HEAPF32[position_F32];
        P2 = +HEAPF32[position_F32 + 1 | 0];
        P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +HEAPF32[M        ] + P2 * +HEAPF32[M + 4 | 0] + P3 * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +HEAPF32[M + 1 | 0] + P2 * +HEAPF32[M + 5 | 0] + P3 * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +HEAPF32[M + 2 | 0] + P2 * +HEAPF32[M + 6 | 0] + P3 * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function CacheTransWithSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            MAT = HEAPF32.subarray(M, M + 16 | 0);    
        }
    
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +MAT[0] + P2 * +MAT[4] + P3 * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +MAT[1] + P2 * +MAT[5] + P3 * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +MAT[2] + P2 * +MAT[6] + P3 * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function CacheTransWithSlice() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            MAT = HEAPF32.slice(M, M + 16 | 0);    
        }
    
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +MAT[0] + P2 * +MAT[4] + P3 * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +MAT[1] + P2 * +MAT[5] + P3 * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +MAT[2] + P2 * +MAT[6] + P3 * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function CacheTransWithF32ArraySetFromSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0, MAT = new Float32Array(16);

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            MAT.set(HEAPF32.subarray(M, M + 16 | 0));    
        }
    
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +MAT[0] + P2 * +MAT[4] + P3 * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +MAT[1] + P2 * +MAT[5] + P3 * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +MAT[2] + P2 * +MAT[6] + P3 * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function CacheTransWithVarsFromSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0;
    // let M0 = Math.fround(0), M1 = Math.fround(0), M2 = Math.fround(0), M4 = Math.fround(0), M5 = Math.fround(0), M6 = Math.fround(0), M8 = Math.fround(0), M9 = Math.fround(0), M10 = Math.fround(0), M12 = Math.fround(0), M13 = Math.fround(0), M14 = Math.fround(0);
    let M0 , M1 , M2 , M4 , M5 , M6 , M8 , M9 , M10 , M12 , M13 , M14 ;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            const MAT = HEAPF32.subarray(M, M + 16 | 0);
            M0 = MAT[0];
            M1 = MAT[1];
            M2 = MAT[2];
            M4 = MAT[4];
            M5 = MAT[5];
            M6= MAT[6];
            M8 = MAT[8];
            M9 = MAT[9];
            M10 = MAT[10];
            M12 = MAT[12];
            M13 = MAT[13];
            M14 = MAT[14];
        }

        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];

        HEAPF32[resPosition_F32        ] = P1 * M0 + P2 * M4 + P3 *  M8 + M12;
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * M1 + P2 * M5 + P3 *  M9 + M13;
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * M2 + P2 * M6 + P3 * M10 + M14;

        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function CacheTransWithArrayFromSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0;
    // let M = [Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0), Math.fround(0)];
    let M = [];

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            const MAT = HEAPF32.subarray(M, M + 16 | 0);
            M[0] = MAT[0];
            M[1] = MAT[1];
            M[2] = MAT[2];
            M[3] = MAT[4];
            M[4] = MAT[5];
            M[5] = MAT[6];
            M[6] = MAT[8];
            M[7] = MAT[9];
            M[8] = MAT[10];
            M[9] = MAT[12];
            M[10] = MAT[13];
            M[11] = MAT[14];
        }
    
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +M[0] + P2 * +M[3] + P3 * +M[6] + +M[9];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +M[1] + P2 * +M[4] + P3 * +M[7] + +M[10];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +M[2] + P2 * +M[5] + P3 * +M[8] + +M[11];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnPosAndCacheTransWithSubarray() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            MAT = HEAPF32.subarray(M, M + 16 | 0);    
        }
    
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +MAT[0] + +POS[1] * +MAT[4] + +POS[2] * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +MAT[1] + +POS[1] * +MAT[5] + +POS[2] * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +MAT[2] + +POS[1] * +MAT[6] + +POS[2] * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let positions_and_joints_32 = positions_and_joints >> 2;
    const transformations_F32 = transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[positions_and_joints_32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + (J << 4) | 0;
            MAT = HEAPF32.subarray(M, M + 16 | 0);    
        }
    
        const POS = HEAPF32.subarray(positions_and_joints_32 + 1 | 0, positions_and_joints_32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +MAT[0] + +POS[1] * +MAT[4] + +POS[2] * +MAT[ 8] + +MAT[12];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +MAT[1] + +POS[1] * +MAT[5] + +POS[2] * +MAT[ 9] + +MAT[13];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +MAT[2] + +POS[1] * +MAT[6] + +POS[2] * +MAT[10] + +MAT[14];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        positions_and_joints_32 = positions_and_joints_32 + 4 | 0;
    }
}

function SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_U32 = joints >> 2;
    const transformations_F32 = small_transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[joint_U32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + Math_imul(J, 12) | 0;
            MAT = HEAPF32.subarray(M, M + 12 | 0);
        }
    
        const POS = HEAPF32.subarray(position_F32, position_F32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +MAT[0] + +POS[1] * +MAT[3] + +POS[2] * +MAT[6] + +MAT[ 9];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +MAT[1] + +POS[1] * +MAT[4] + +POS[2] * +MAT[7] + +MAT[10];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +MAT[2] + +POS[1] * +MAT[5] + +POS[2] * +MAT[8] + +MAT[11];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let positions_and_joints_32 = positions_and_joints >> 2;
    const transformations_F32 = small_transformations >> 2;

    let J = -1|0, MAT;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const NJ = HEAPU32[positions_and_joints_32];
        if (NJ != J) {
            J = NJ;
            const M = transformations_F32 + Math_imul(J, 12) | 0;
            MAT = HEAPF32.subarray(M, M + 12 | 0);
        }
    
        const POS = HEAPF32.subarray(positions_and_joints_32 + 1 | 0, positions_and_joints_32 + 3 | 0);
    
        HEAPF32[resPosition_F32        ] = +POS[0] * +MAT[0] + +POS[1] * +MAT[3] + +POS[2] * +MAT[6] + +MAT[ 9];
        HEAPF32[resPosition_F32 + 1 | 0] = +POS[0] * +MAT[1] + +POS[1] * +MAT[4] + +POS[2] * +MAT[7] + +MAT[10];
        HEAPF32[resPosition_F32 + 2 | 0] = +POS[0] * +MAT[2] + +POS[1] * +MAT[5] + +POS[2] * +MAT[8] + +MAT[11];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        positions_and_joints_32 = positions_and_joints_32 + 4 | 0;
    }
}

function F64PosAndF64Trans() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F64 = positions_f64 >> 3;
    let joint_U32 = joints >> 2;
    const transformations_F64 = transformations_f64 >> 3;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F64 + (J << 4) | 0;
        const P1 = +HEAPF64[position_F64];
        const P2 = +HEAPF64[position_F64 + 1 | 0];
        const P3 = +HEAPF64[position_F64 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +HEAPF64[M        ] + P2 * +HEAPF64[M + 4 | 0] + P3 * +HEAPF64[M +  8 | 0] + +HEAPF64[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +HEAPF64[M + 1 | 0] + P2 * +HEAPF64[M + 5 | 0] + P3 * +HEAPF64[M +  9 | 0] + +HEAPF64[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +HEAPF64[M + 2 | 0] + P2 * +HEAPF64[M + 6 | 0] + P3 * +HEAPF64[M + 10 | 0] + +HEAPF64[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F64 = position_F64 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function F64PosAndF64TransAndF64Res() {
    let resPosition_F64 = res_f64 >> 3;
    const endResPosition_F64 = (res_f64 >> 3) + Math_imul(vertexCount, 3) | 0;
    let position_F64 = positions_f64 >> 3;
    let joint_U32 = joints >> 2;
    const transformations_F64 = transformations_f64 >> 3;

    while (resPosition_F64 >>> 0 < endResPosition_F64 >>> 0) {
        const J = HEAPU32[joint_U32];
        const M = transformations_F64 + (J << 4) | 0;
        const P1 = +HEAPF64[position_F64];
        const P2 = +HEAPF64[position_F64 + 1 | 0];
        const P3 = +HEAPF64[position_F64 + 2 | 0];
    
        HEAPF32[resPosition_F64        ] = P1 * +HEAPF64[M        ] + P2 * +HEAPF64[M + 4 | 0] + P3 * +HEAPF64[M +  8 | 0] + +HEAPF64[M + 12 | 0];
        HEAPF32[resPosition_F64 + 1 | 0] = P1 * +HEAPF64[M + 1 | 0] + P2 * +HEAPF64[M + 5 | 0] + P3 * +HEAPF64[M +  9 | 0] + +HEAPF64[M + 13 | 0];
        HEAPF32[resPosition_F64 + 2 | 0] = P1 * +HEAPF64[M + 2 | 0] + P2 * +HEAPF64[M + 6 | 0] + P3 * +HEAPF64[M + 10 | 0] + +HEAPF64[M + 14 | 0];
    
        resPosition_F64 = resPosition_F64 + 3 | 0;
        position_F64 = position_F64 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

function I32Joint() {
    let resPosition_F32 = res >> 2;
    const endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    let position_F32 = positions >> 2;
    let joint_I32 = joints >> 2;
    const transformations_F32 = transformations >> 2;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        const J = HEAP32[joint_I32];
        const M = transformations_F32 + Math_imul(J, 16) | 0;
        const P1 = +HEAPF32[position_F32];
        const P2 = +HEAPF32[position_F32 + 1 | 0];
        const P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +HEAPF32[M        ] + P2 * +HEAPF32[M + 4 | 0] + P3 * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +HEAPF32[M + 1 | 0] + P2 * +HEAPF32[M + 5 | 0] + P3 * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +HEAPF32[M + 2 | 0] + P2 * +HEAPF32[M + 6 | 0] + P3 * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_I32 = joint_I32 + 1 | 0;
    }
}

function VarInsteadOfLetAndConst() {
    var resPosition_F32 = res >> 2;
    var endResPosition_F32 = (res >> 2) + Math_imul(vertexCount, 3) | 0;
    var position_F32 = positions >> 2;
    var joint_U32 = joints >> 2;
    var transformations_F32 = transformations >> 2;
    var J, M, P1, P2, P3;

    while (resPosition_F32 >>> 0 < endResPosition_F32 >>> 0) {
        J = HEAPU32[joint_U32];
        M = transformations_F32 + (J << 4) | 0;
        P1 = +HEAPF32[position_F32];
        P2 = +HEAPF32[position_F32 + 1 | 0];
        P3 = +HEAPF32[position_F32 + 2 | 0];
    
        HEAPF32[resPosition_F32        ] = P1 * +HEAPF32[M        ] + P2 * +HEAPF32[M + 4 | 0] + P3 * +HEAPF32[M +  8 | 0] + +HEAPF32[M + 12 | 0];
        HEAPF32[resPosition_F32 + 1 | 0] = P1 * +HEAPF32[M + 1 | 0] + P2 * +HEAPF32[M + 5 | 0] + P3 * +HEAPF32[M +  9 | 0] + +HEAPF32[M + 13 | 0];
        HEAPF32[resPosition_F32 + 2 | 0] = P1 * +HEAPF32[M + 2 | 0] + P2 * +HEAPF32[M + 6 | 0] + P3 * +HEAPF32[M + 10 | 0] + +HEAPF32[M + 14 | 0];
    
        resPosition_F32 = resPosition_F32 + 3 | 0;
        position_F32 = position_F32 + 3 | 0;
        joint_U32 = joint_U32 + 1 | 0;
    }
}

Tests:

Warmup
Warmup()
Original
Original()
UpdatedEmscripten
UpdatedEmscripten()
LoicRewrite
LoicRewrite()
SubarrayOnTrans
SubarrayOnTrans()
SubarrayOnPos
SubarrayOnPos()
SubarrayOnTransAndPos
SubarrayOnTransAndPos()
SubarrayOnTransAndPosWithFRound
SubarrayOnTransAndPosWithFRound()
SubarrayOnTransAndPosWithoutPlus
SubarrayOnTransAndPosWithoutPlus()
PosInOutOfTheLoopF32Array
PosInOutOfTheLoopF32Array()
PosInOutOfTheLoopF32ArrayFromSubarray
PosInOutOfTheLoopF32ArrayFromSubarray()
PosInOutOfTheLoopVars
PosInOutOfTheLoopVars()
CacheTransWithSubarray
CacheTransWithSubarray()
CacheTransWithSlice
CacheTransWithSlice()
CacheTransWithF32ArraySetFromSubarray
CacheTransWithF32ArraySetFromSubarray()
CacheTransWithVarsFromSubarray
CacheTransWithVarsFromSubarray()
CacheTransWithArrayFromSubarray
CacheTransWithArrayFromSubarray()
SubarrayOnPosAndCacheTransWithSubarray
SubarrayOnPosAndCacheTransWithSubarray()
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData()
SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans
SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans()
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans()
F64PosAndF64Trans
F64PosAndF64Trans()
F64PosAndF64TransAndF64Res
F64PosAndF64TransAndF64Res()
I32Joint
I32Joint()
VarInsteadOfLetAndConst
VarInsteadOfLetAndConst()

Historical Performance Data aggregated from 14 community runs

Desktop Results:

Chrome 6 runs

SubarrayOnPosAndCacheTransWithSubarray

257/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData

254/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans

240/s

SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans

239/s

SubarrayOnTransAndPos

208/s

CacheTransWithVarsFromSubarray

207/s

SubarrayOnTransAndPosWithoutPlus

206/s

CacheTransWithSlice

203/s

CacheTransWithSubarray

202/s

CacheTransWithF32ArraySetFromSubarray

201/s

SubarrayOnTrans

175/s

SubarrayOnPos

86/s

PosInOutOfTheLoopF32ArrayFromSubarray

85/s

LoicRewrite

80/s

PosInOutOfTheLoopVars

79/s

PosInOutOfTheLoopF32Array

79/s

CacheTransWithArrayFromSubarray

54/s

SubarrayOnTransAndPosWithFRound

19/s

Original

18/s

UpdatedEmscripten

4/s

Warmup

0/s

Fastest: SubarrayOnPosAndCacheTransWithSubarray (1574.1x faster)

View exact numbers

✓ SubarrayOnPosAndCacheTransWithSubarray: 257 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData: 254 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans: 240 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans: 239 ops/sec
• SubarrayOnTransAndPos: 208 ops/sec
• CacheTransWithVarsFromSubarray: 207 ops/sec
• SubarrayOnTransAndPosWithoutPlus: 206 ops/sec
• CacheTransWithSlice: 203 ops/sec
• CacheTransWithSubarray: 202 ops/sec
• CacheTransWithF32ArraySetFromSubarray: 201 ops/sec
• SubarrayOnTrans: 175 ops/sec
• SubarrayOnPos: 86 ops/sec
• PosInOutOfTheLoopF32ArrayFromSubarray: 85 ops/sec
• LoicRewrite: 80 ops/sec
• PosInOutOfTheLoopVars: 79 ops/sec
• PosInOutOfTheLoopF32Array: 79 ops/sec
• CacheTransWithArrayFromSubarray: 54 ops/sec
• SubarrayOnTransAndPosWithFRound: 19 ops/sec
• Original: 18 ops/sec
• UpdatedEmscripten: 4 ops/sec
• Warmup: 0 ops/sec

Mobile Results:

Chrome 8 runs

SubarrayOnPosAndCacheTransWithSubarray

118/s

SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans

117/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans

115/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData

114/s

CacheTransWithSubarray

103/s

CacheTransWithF32ArraySetFromSubarray

103/s

CacheTransWithSlice

101/s

CacheTransWithVarsFromSubarray

101/s

SubarrayOnTransAndPos

100/s

SubarrayOnTransAndPosWithoutPlus

99/s

SubarrayOnTrans

86/s

SubarrayOnPos

42/s

PosInOutOfTheLoopF32ArrayFromSubarray

42/s

PosInOutOfTheLoopVars

41/s

LoicRewrite

41/s

PosInOutOfTheLoopF32Array

40/s

CacheTransWithArrayFromSubarray

30/s

SubarrayOnTransAndPosWithFRound

9/s

Original

9/s

UpdatedEmscripten

2/s

Warmup

0/s

Fastest: SubarrayOnPosAndCacheTransWithSubarray (1391.2x faster)

View exact numbers

✓ SubarrayOnPosAndCacheTransWithSubarray: 118 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans: 117 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans: 115 ops/sec
• SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData: 114 ops/sec
• CacheTransWithSubarray: 103 ops/sec
• CacheTransWithF32ArraySetFromSubarray: 103 ops/sec
• CacheTransWithSlice: 101 ops/sec
• CacheTransWithVarsFromSubarray: 101 ops/sec
• SubarrayOnTransAndPos: 100 ops/sec
• SubarrayOnTransAndPosWithoutPlus: 99 ops/sec
• SubarrayOnTrans: 86 ops/sec
• SubarrayOnPos: 42 ops/sec
• PosInOutOfTheLoopF32ArrayFromSubarray: 42 ops/sec
• PosInOutOfTheLoopVars: 41 ops/sec
• LoicRewrite: 41 ops/sec
• PosInOutOfTheLoopF32Array: 40 ops/sec
• CacheTransWithArrayFromSubarray: 30 ops/sec
• SubarrayOnTransAndPosWithFRound: 9 ops/sec
• Original: 9 ops/sec
• UpdatedEmscripten: 2 ops/sec
• Warmup: 0 ops/sec

Results show median performance across all community submissions. Run this benchmark yourself to verify on your device.

Rendered benchmark preparation results:

Suite status: <idle, ready to run>

Previous results

Test case name	Result
Warmup
Original
UpdatedEmscripten
LoicRewrite
SubarrayOnTrans
SubarrayOnPos
SubarrayOnTransAndPos
SubarrayOnTransAndPosWithFRound
SubarrayOnTransAndPosWithoutPlus
PosInOutOfTheLoopF32Array
PosInOutOfTheLoopF32ArrayFromSubarray
PosInOutOfTheLoopVars
CacheTransWithSubarray
CacheTransWithSlice
CacheTransWithF32ArraySetFromSubarray
CacheTransWithVarsFromSubarray
CacheTransWithArrayFromSubarray
SubarrayOnPosAndCacheTransWithSubarray
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData
SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans
F64PosAndF64Trans
F64PosAndF64TransAndF64Res
I32Joint
VarInsteadOfLetAndConst

Fastest: N/A

Slowest: N/A

Latest run results:

Run details: (Test run date: 2 years ago)

User agent: Mozilla/5.0 (Android 13; Mobile; rv:122.0) Gecko/122.0 Firefox/122.0

Browser/OS: Firefox Mobile 122 on Android

View result in a separate tab

F64PosAndF64TransAndF64Res

4K/s

LoicRewrite

4K/s

F64PosAndF64Trans

3K/s

PosInOutOfTheLoopVars

3K/s

I32Joint

3K/s

VarInsteadOfLetAndConst

3K/s

PosInOutOfTheLoopF32Array

2K/s

Original

2K/s

CacheTransWithSlice

1K/s

CacheTransWithSubarray

1K/s

UpdatedEmscripten

1K/s

CacheTransWithVarsFromSubarray

982/s

CacheTransWithF32ArraySetFromSubarray

873/s

SubarrayOnTrans

326/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData

280/s

SubarrayOnPosAndCacheTransWithSubarray

274/s

SubarrayOnPos

269/s

SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans

240/s

PosInOutOfTheLoopF32ArrayFromSubarray

201/s

SubarrayOnTransAndPosWithoutPlus

193/s

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans

190/s

SubarrayOnTransAndPosWithFRound

175/s

SubarrayOnTransAndPos

168/s

CacheTransWithArrayFromSubarray

124/s

Warmup

1/s

View exact numbers

Test name	Executions per second
✓ F64PosAndF64TransAndF64Res	3,679 Ops/sec
LoicRewrite	3,642 Ops/sec
F64PosAndF64Trans	3,330 Ops/sec
PosInOutOfTheLoopVars	3,296 Ops/sec
I32Joint	3,220 Ops/sec
VarInsteadOfLetAndConst	2,740 Ops/sec
PosInOutOfTheLoopF32Array	2,442 Ops/sec
Original	1,674 Ops/sec
CacheTransWithSlice	1,280 Ops/sec
CacheTransWithSubarray	1,161 Ops/sec
UpdatedEmscripten	1,111 Ops/sec
CacheTransWithVarsFromSubarray	982 Ops/sec
CacheTransWithF32ArraySetFromSubarray	873 Ops/sec
SubarrayOnTrans	326 Ops/sec
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData	280 Ops/sec
SubarrayOnPosAndCacheTransWithSubarray	274 Ops/sec
SubarrayOnPos	269 Ops/sec
SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans	240 Ops/sec
PosInOutOfTheLoopF32ArrayFromSubarray	201 Ops/sec
SubarrayOnTransAndPosWithoutPlus	193 Ops/sec
SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans	190 Ops/sec
SubarrayOnTransAndPosWithFRound	175 Ops/sec
SubarrayOnTransAndPos	168 Ops/sec
CacheTransWithArrayFromSubarray	124 Ops/sec
Warmup	1 Ops/sec

Autogenerated LLM Summary (model llama3.2:3b, generated one year ago):

LLMs can make mistakes. Check important info.

Linear Blend Skinning 2 (version: 1) Benchmark.js

Created: 2 years ago by: Registered User

Go to the latest result

Warmup

Original

UpdatedEmscripten

LoicRewrite

SubarrayOnTrans

SubarrayOnPos

SubarrayOnTransAndPos

SubarrayOnTransAndPosWithFRound

SubarrayOnTransAndPosWithoutPlus

PosInOutOfTheLoopF32Array

PosInOutOfTheLoopF32ArrayFromSubarray

PosInOutOfTheLoopVars

CacheTransWithSubarray

CacheTransWithSlice

CacheTransWithF32ArraySetFromSubarray

CacheTransWithVarsFromSubarray

CacheTransWithArrayFromSubarray

SubarrayOnPosAndCacheTransWithSubarray

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingData

SubarrayOnPosAndCacheTransWithSubarrayAndSmallTrans

SubarrayOnPosAndCacheTransWithSubarrayAndInterleavingDataAndSmallTrans

F64PosAndF64Trans

F64PosAndF64TransAndF64Res

I32Joint

VarInsteadOfLetAndConst

Historical Performance Data aggregated from 14 community runs

Desktop Results:

Chrome 6 runs

Mobile Results:

Chrome 8 runs

Suite status: <idle, ready to run>

Fastest: N/A

Slowest: N/A

Autogenerated LLM Summary (model llama3.2:3b, generated one year ago):

Related benchmarks