Skip to content

GitLab

Commit c799ea08 authored by Peter Eastman's avatar Peter Eastman
Browse files

Fixed bug introduced by previous checkin

parent bd7cce70
Show whitespace changes
Inline Side-by-side
Showing with 8 additions and 8 deletions
platforms/opencl/src/kernels/gbsaObc_default.cl
View file @ c799ea08
...@@ -79,7 +79,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -79,7 +79,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2); (0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2);
if (params1.x < params2.x-r) if (params1.x < params2.x-r)
bornSum += 2.0f*RECIP(params1.x-l_ij); bornSum += 2.0f*(RECIP(params1.x)-l_ij);
} }
} }
} }
...@@ -147,7 +147,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -147,7 +147,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2); (0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2);
if (params1.x < params2.x-r) if (params1.x < params2.x-r)
bornSum += 2.0f*RECIP(params1.x-l_ij); bornSum += 2.0f*(RECIP(params1.x)-l_ij);
} }
float rScaledRadiusI = r+params1.y; float rScaledRadiusI = r+params1.y;
if (params2.x < rScaledRadiusI) { if (params2.x < rScaledRadiusI) {
...@@ -159,7 +159,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -159,7 +159,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2); (0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2);
if (params2.x < params1.x-r) if (params2.x < params1.x-r)
term += 2.0f*RECIP(params2.x-l_ij); term += 2.0f*(RECIP(params2.x)-l_ij);
localData[baseLocalAtom+tj+forceBufferOffset].bornSum += term; localData[baseLocalAtom+tj+forceBufferOffset].bornSum += term;
} }
} }
......
platforms/opencl/src/kernels/gbsaObc_nvidia.cl
View file @ c799ea08
...@@ -79,7 +79,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -79,7 +79,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2); (0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2);
if (params1.x < params2.x-r) if (params1.x < params2.x-r)
bornSum += 2.0f*RECIP(params1.x-l_ij); bornSum += 2.0f*(RECIP(params1.x)-l_ij);
} }
} }
} }
...@@ -144,7 +144,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -144,7 +144,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2); (0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2);
if (params1.x < params2.x-r) if (params1.x < params2.x-r)
bornSum += 2.0f*RECIP(params1.x-l_ij); bornSum += 2.0f*(RECIP(params1.x)-l_ij);
} }
float rScaledRadiusI = r+params1.y; float rScaledRadiusI = r+params1.y;
if (params2.x < rScaledRadiusI) { if (params2.x < rScaledRadiusI) {
...@@ -156,7 +156,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -156,7 +156,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2); (0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2);
if (params2.x < params1.x-r) if (params2.x < params1.x-r)
term += 2.0f*RECIP(params2.x-l_ij); term += 2.0f*(RECIP(params2.x)-l_ij);
tempBuffer[get_local_id(0)] = term; tempBuffer[get_local_id(0)] = term;
} }
} }
...@@ -212,7 +212,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -212,7 +212,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + bornSum += l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2); (0.25f*params2.y*params2.y*invR)*(l_ij2-u_ij2);
if (params1.x < params2.x-r) if (params1.x < params2.x-r)
bornSum += 2.0f*RECIP(params1.x-l_ij); bornSum += 2.0f*(RECIP(params1.x)-l_ij);
} }
float rScaledRadiusI = r+params1.y; float rScaledRadiusI = r+params1.y;
if (params2.x < rScaledRadiusI) { if (params2.x < rScaledRadiusI) {
...@@ -224,7 +224,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -224,7 +224,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) + float term = l_ij - u_ij + 0.25f*r*(u_ij2-l_ij2) + (0.50f*invR*ratio) +
(0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2); (0.25f*params1.y*params1.y*invR)*(l_ij2-u_ij2);
if (params2.x < params1.x-r) if (params2.x < params1.x-r)
term += 2.0f*RECIP(params2.x-l_ij); term += 2.0f*(RECIP(params2.x)-l_ij);
localData[tbx+tj].bornSum += term; localData[tbx+tj].bornSum += term;
} }
} }
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment