Debug CUDA

unitaryfund · Nov 2, 2024 · 8a7fdea · 8a7fdea
1 parent 7490f96
commit 8a7fdea
Showing 1 changed file with 9 additions and 2 deletions.
diff --git a/src/common/qengine.cu b/src/common/qengine.cu
@@ -698,15 +698,15 @@ __global__ void decomposeprob(qCudaCmplx* stateVec, bitCapIntOcl* bitCapIntOclPt
             remainderStateAngle[k] += qCudaArg(amp) * (qCudaReal1)nrm;
         }
 
-        if (partProb > REAL1_EPSILON) {
+        if (partProb > REAL1_EPSILON_CUDA) {
             partStateAngle[lcv] /= partProb;
         }
         partStateProb[lcv] = partProb;
     }
 
     for (bitCapIntOcl lcv = ID; lcv < remainderPower; lcv += Nthreads) {
         const real1 prob = remainderStateProb[lcv];
-        if (prob > REAL1_EPSILON) {
+        if (prob > REAL1_EPSILON_CUDA) {
             remainderStateAngle[lcv] /= prob;
         }
     }
@@ -766,6 +766,13 @@ __global__ void disposeprob(qCudaCmplx* stateVec, bitCapIntOcl* bitCapIntOclPtr,
             remainderStateAngle[k] += qCudaArg(amp) * (qCudaReal1)nrm;
         }
     }
+
+    for (bitCapIntOcl lcv = ID; lcv < remainderPower; lcv += Nthreads) {
+        const real1 prob = remainderStateProb[lcv];
+        if (prob > REAL1_EPSILON_CUDA) {
+            remainderStateAngle[lcv] /= prob;
+        }
+    }
 }
 
 __global__ void dispose(qCudaCmplx* stateVec, bitCapIntOcl* bitCapIntOclPtr, qCudaCmplx* nStateVec)