Improved documentation.

mumax · Jul 9, 2020 · 53a0217 · 53a0217
1 parent 2d7bf8a
commit 53a0217
Show file tree

Hide file tree

Showing 6 changed files with 9 additions and 10 deletions.
diff --git a/cuda/madd.go b/cuda/madd.go
@@ -47,7 +47,7 @@ func Madd2(dst, src1, src2 *data.Slice, factor1, factor2 float32) {
 	}
 }
 
-// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3 * factor3
+// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3[i] * factor3
 func Madd3(dst, src1, src2, src3 *data.Slice, factor1, factor2, factor3 float32) {
 	N := dst.Len()
 	nComp := dst.NComp()
@@ -60,7 +60,7 @@ func Madd3(dst, src1, src2, src3 *data.Slice, factor1, factor2, factor3 float32)
 	}
 }
 
-// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3 * factor3
+// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3[i] * factor3 + src4[i] * factor4
 func Madd4(dst, src1, src2, src3, src4 *data.Slice, factor1, factor2, factor3, factor4 float32) {
 	N := dst.Len()
 	nComp := dst.NComp()
@@ -76,7 +76,7 @@ func Madd4(dst, src1, src2, src3, src4 *data.Slice, factor1, factor2, factor3, f
 	}
 }
 
-// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3 * factor3
+// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3[i] * factor3 + src4[i] * factor4 + src5[i] * factor5
 func Madd5(dst, src1, src2, src3, src4, src5 *data.Slice, factor1, factor2, factor3, factor4, factor5 float32) {
 	N := dst.Len()
 	nComp := dst.NComp()
@@ -93,7 +93,7 @@ func Madd5(dst, src1, src2, src3, src4, src5 *data.Slice, factor1, factor2, fact
 	}
 }
 
-// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3 * factor3
+// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3[i] * factor3 + src4[i] * factor4 + src5[i] * factor5 + src6[i] * factor6
 func Madd6(dst, src1, src2, src3, src4, src5, src6 *data.Slice, factor1, factor2, factor3, factor4, factor5, factor6 float32) {
 	N := dst.Len()
 	nComp := dst.NComp()
@@ -111,7 +111,7 @@ func Madd6(dst, src1, src2, src3, src4, src5, src6 *data.Slice, factor1, factor2
 	}
 }
 
-// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3 * factor3
+// multiply-add: dst[i] = src1[i] * factor1 + src2[i] * factor2 + src3[i] * factor3 + src4[i] * factor4 + src5[i] * factor5 + src6[i] * factor6 + src7[i] * factor7
 func Madd7(dst, src1, src2, src3, src4, src5, src6, src7 *data.Slice, factor1, factor2, factor3, factor4, factor5, factor6, factor7 float32) {
 	N := dst.Len()
 	nComp := dst.NComp()

diff --git a/cuda/madd4.cu b/cuda/madd4.cu
@@ -1,5 +1,5 @@
 
-// dst[i] = fac1 * src1[i] + fac2 * src2[i] + fac3 * src3[i]
+// dst[i] = src1[i] * fac1 + src2[i] * fac2 + src3[i] * fac3 + src4[i] * fac4
 extern "C" __global__ void
 madd4(float* __restrict__ dst,
       float* __restrict__ src1, float fac1,

diff --git a/cuda/madd5.cu b/cuda/madd5.cu
@@ -1,5 +1,5 @@
 
-// dst[i] = fac1 * src1[i] + fac2 * src2[i] + fac3 * src3[i]
+// dst[i] = src1[i] * fac1 + src2[i] * fac2 + src3[i] * fac3 + src4[i] * fac4 + src5[i] * fac5
 extern "C" __global__ void
 madd5(float* __restrict__ dst,
       float* __restrict__ src1, float fac1,

diff --git a/cuda/madd6.cu b/cuda/madd6.cu
@@ -1,5 +1,5 @@
 
-// dst[i] = fac1 * src1[i] + fac2 * src2[i] + fac3 * src3[i]
+// dst[i] = src1[i] * fac1 + src2[i] * fac2 + src3[i] * fac3 + src4[i] * fac4 + src5[i] * fac5 + src6[i] * fac6
 extern "C" __global__ void
 madd6(float* __restrict__ dst,
       float* __restrict__ src1, float fac1,

diff --git a/cuda/madd7.cu b/cuda/madd7.cu
@@ -1,5 +1,5 @@
 
-// dst[i] = fac1 * src1[i] + fac2 * src2[i] + fac3 * src3[i]
+// dst[i] = src1[i] * fac1 + src2[i] * fac2 + src3[i] * fac3 + src4[i] * fac4 + src5[i] * fac5 + src6[i] * fac6 + src7[i] * fac7
 extern "C" __global__ void
 madd7(float* __restrict__ dst,
       float* __restrict__ src1, float fac1,

diff --git a/test/sp4_madd_bench.mx3 b/test/sp4_madd_bench.mx3
@@ -47,4 +47,3 @@ run(10e-9)
 t7 := now()
 
 print(sprintf("Run time: RK23BS %6.6E, RK4 %6.6E, RK45DP %6.6E, RK56 %6.6E", t1.sub(t0).Seconds(), t3.sub(t2).Seconds(), t5.sub(t4).Seconds(), t7.sub(t6).Seconds()))
-