aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add support for fast division of varying int values by small constants.

Browse Source 
For varying int8/16/32 types, divides by small constants can be
implemented efficiently through multiplies and shifts with integer
types of twice the bit-width; this commit adds this optimization.
    
(Implementation is based on Halide.)
This commit is contained in:
Matt Pharr
2013-07-23 16:49:56 -07:00
parent 0277ba1aaa
commit 83e1630fbc
3 changed files with 819 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
tests/idiv.ispc Normal file
View File

@@ -0,0 +1,75 @@
export uniform int width() { return programCount; }
export void f_f(uniform float RET[], uniform float aFOO[]) {
uniform int errorCount = 0;
for (unsigned int8 num = 0; num < 255; ++num) {
for (uniform unsigned int8 div = 2; div < 255; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
for (int8 num = 0; num < 127; ++num) {
for (uniform int8 div = 2; div < 127; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
for (int16 num = 0; num < 32767; ++num) {
for (uniform int16 div = 2; div < 256; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
for (unsigned int16 num = 0; num < 0xffff; ++num) {
for (uniform unsigned int16 div = 2; div < 256; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
// randomly sample int32s...
uniform RNGState state;
seed_rng(&state, 1234);
for (uniform int i = 0; i < 1M; ++i) {
unsigned int32 num = random(&state);
for (uniform unsigned int32 div = 2; div < 256; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("ui32 error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
for (uniform int64 i = 0; i < 1M; ++i) {
int32 num = random(&state);
if (num < 0)
continue;
for (uniform int32 div = 2; div < 256; ++div) {
if (__fast_idiv(num, div) != num/div) {
++errorCount;
print("si32 error %/% = %, got %\n", num, div, num/div, __fast_idiv(num,div));
}
}
}
RET[programIndex] = errorCount;
}
export void result(uniform float RET[]) {
RET[programIndex] = 0;
}