HDR: speed up reading by around 4x (AcademySoftwareFoundation#3590)

Most of the cost in HDR file reading was converting from RGBE to floats,
inside ldexpf(), which at least on Windows/VS2022 spends most of the
time inside _ctrlfp(). But we only have 256 possible exponents in total,
so just use a precomputed table.

On my PC (Windows 10, VS2022, Ryzen 5950X) this change gets file read
time for an 8x resolution .HDR image from 1.10s down to 0.27s

Author: aras-p

src/hdr.imageio/hdrinput.cpp

@@ -17,7 +17,7 @@ OIIO_PLUGIN_NAMESPACE_BEGIN
 // .hdr / .rgbe files - HDR files from Radiance
 //
 // General info on the hdr/rgbe format can be found at:
-//     http://local.wasp.uwa.edu.au/~pbourke/dataformats/pic/
+//     http://paulbourke.net/dataformats/pic/
 //
 // Also see Greg Ward's "Real Pixels" chapter in Graphics Gems II for an
 // explanation of the encoding that's used in Radiance rgba files.
@@ -115,15 +115,83 @@ OIIO_EXPORT const char* hdr_input_extensions[] = { "hdr", "rgbe", nullptr };
 OIIO_PLUGIN_EXPORTS_END
 
 
+// Table of ldexpf(1.0f, i - (int)(128 + 8)) values.
+// Note: Ward uses ldexp(col+0.5,exp-(128+8)).  However we wanted pixels
+//       in the range [0,1] to map back into the range [0,1].
+static const float exponent_table[256] = {
+    1.1479437e-41f, 2.2958874e-41f, 4.5917748e-41f, 9.1835496e-41f,
+    1.8367099e-40f, 3.6734198e-40f, 7.3468397e-40f, 1.4693679e-39f,
+    2.9387359e-39f, 5.8774718e-39f, 1.1754944e-38f, 2.3509887e-38f,
+    4.7019774e-38f, 9.4039548e-38f, 1.880791e-37f,  3.7615819e-37f,
+    7.5231638e-37f, 1.5046328e-36f, 3.0092655e-36f, 6.0185311e-36f,
+    1.2037062e-35f, 2.4074124e-35f, 4.8148249e-35f, 9.6296497e-35f,
+    1.9259299e-34f, 3.8518599e-34f, 7.7037198e-34f, 1.540744e-33f,
+    3.0814879e-33f, 6.1629758e-33f, 1.2325952e-32f, 2.4651903e-32f,
+    4.9303807e-32f, 9.8607613e-32f, 1.9721523e-31f, 3.9443045e-31f,
+    7.8886091e-31f, 1.5777218e-30f, 3.1554436e-30f, 6.3108872e-30f,
+    1.2621774e-29f, 2.5243549e-29f, 5.0487098e-29f, 1.009742e-28f,
+    2.0194839e-28f, 4.0389678e-28f, 8.0779357e-28f, 1.6155871e-27f,
+    3.2311743e-27f, 6.4623485e-27f, 1.2924697e-26f, 2.5849394e-26f,
+    5.1698788e-26f, 1.0339758e-25f, 2.0679515e-25f, 4.1359031e-25f,
+    8.2718061e-25f, 1.6543612e-24f, 3.3087225e-24f, 6.6174449e-24f,
+    1.323489e-23f,  2.646978e-23f,  5.2939559e-23f, 1.0587912e-22f,
+    2.1175824e-22f, 4.2351647e-22f, 8.4703295e-22f, 1.6940659e-21f,
+    3.3881318e-21f, 6.7762636e-21f, 1.3552527e-20f, 2.7105054e-20f,
+    5.4210109e-20f, 1.0842022e-19f, 2.1684043e-19f, 4.3368087e-19f,
+    8.6736174e-19f, 1.7347235e-18f, 3.469447e-18f,  6.9388939e-18f,
+    1.3877788e-17f, 2.7755576e-17f, 5.5511151e-17f, 1.110223e-16f,
+    2.220446e-16f,  4.4408921e-16f, 8.8817842e-16f, 1.7763568e-15f,
+    3.5527137e-15f, 7.1054274e-15f, 1.4210855e-14f, 2.8421709e-14f,
+    5.6843419e-14f, 1.1368684e-13f, 2.2737368e-13f, 4.5474735e-13f,
+    9.094947e-13f,  1.8189894e-12f, 3.6379788e-12f, 7.2759576e-12f,
+    1.4551915e-11f, 2.910383e-11f,  5.8207661e-11f, 1.1641532e-10f,
+    2.3283064e-10f, 4.6566129e-10f, 9.3132257e-10f, 1.8626451e-09f,
+    3.7252903e-09f, 7.4505806e-09f, 1.4901161e-08f, 2.9802322e-08f,
+    5.9604645e-08f, 1.1920929e-07f, 2.3841858e-07f, 4.7683716e-07f,
+    9.5367432e-07f, 1.9073486e-06f, 3.8146973e-06f, 7.6293945e-06f,
+    1.5258789e-05f, 3.0517578e-05f, 6.1035156e-05f, 0.00012207031f,
+    0.00024414062f, 0.00048828125f, 0.0009765625f,  0.001953125f,
+    0.00390625f,    0.0078125f,     0.015625f,      0.03125f,
+    0.0625f,        0.125f,         0.25f,          0.5f,
+    1.0f,           2.0f,           4.0f,           8.0f,
+    16.0f,          32.0f,          64.0f,          128.0f,
+    256.0f,         512.0f,         1024.0f,        2048.0f,
+    4096.0f,        8192.0f,        16384.0f,       32768.0f,
+    65536.0f,       131072.0f,      262144.0f,      524288.0f,
+    1048576.0f,     2097152.0f,     4194304.0f,     8388608.0f,
+    16777216.0f,    33554432.0f,    67108864.0f,    1.3421773e+08f,
+    2.6843546e+08f, 5.3687091e+08f, 1.0737418e+09f, 2.1474836e+09f,
+    4.2949673e+09f, 8.5899346e+09f, 1.7179869e+10f, 3.4359738e+10f,
+    6.8719477e+10f, 1.3743895e+11f, 2.7487791e+11f, 5.4975581e+11f,
+    1.0995116e+12f, 2.1990233e+12f, 4.3980465e+12f, 8.796093e+12f,
+    1.7592186e+13f, 3.5184372e+13f, 7.0368744e+13f, 1.4073749e+14f,
+    2.8147498e+14f, 5.6294995e+14f, 1.1258999e+15f, 2.2517998e+15f,
+    4.5035996e+15f, 9.0071993e+15f, 1.8014399e+16f, 3.6028797e+16f,
+    7.2057594e+16f, 1.4411519e+17f, 2.8823038e+17f, 5.7646075e+17f,
+    1.1529215e+18f, 2.305843e+18f,  4.611686e+18f,  9.223372e+18f,
+    1.8446744e+19f, 3.6893488e+19f, 7.3786976e+19f, 1.4757395e+20f,
+    2.9514791e+20f, 5.9029581e+20f, 1.1805916e+21f, 2.3611832e+21f,
+    4.7223665e+21f, 9.444733e+21f,  1.8889466e+22f, 3.7778932e+22f,
+    7.5557864e+22f, 1.5111573e+23f, 3.0223145e+23f, 6.0446291e+23f,
+    1.2089258e+24f, 2.4178516e+24f, 4.8357033e+24f, 9.6714066e+24f,
+    1.9342813e+25f, 3.8685626e+25f, 7.7371252e+25f, 1.547425e+26f,
+    3.0948501e+26f, 6.1897002e+26f, 1.23794e+27f,   2.4758801e+27f,
+    4.9517602e+27f, 9.9035203e+27f, 1.9807041e+28f, 3.9614081e+28f,
+    7.9228163e+28f, 1.5845633e+29f, 3.1691265e+29f, 6.338253e+29f,
+    1.2676506e+30f, 2.5353012e+30f, 5.0706024e+30f, 1.0141205e+31f,
+    2.028241e+31f,  4.0564819e+31f, 8.1129638e+31f, 1.6225928e+32f,
+    3.2451855e+32f, 6.4903711e+32f, 1.2980742e+33f, 2.5961484e+33f,
+    5.1922969e+33f, 1.0384594e+34f, 2.0769187e+34f, 4.1538375e+34f,
+    8.307675e+34f,  1.661535e+35f,  3.32307e+35f,   6.64614e+35f,
+};
+
 
 /* standard conversion from rgbe to float pixels */
-/* note: Ward uses ldexp(col+0.5,exp-(128+8)).  However we wanted pixels */
-/*       in the range [0,1] to map back into the range [0,1].            */
 inline void
 rgbe2float(float& red, float& green, float& blue, unsigned char rgbe[4])
 {
     if (rgbe[3]) { /*nonzero pixel*/
-        float f = ldexpf(1.0f, rgbe[3] - (int)(128 + 8));
+        float f = exponent_table[rgbe[3]];
         red     = rgbe[0] * f;
         green   = rgbe[1] * f;
         blue    = rgbe[2] * f;