Merge branch 'master' of https://bitbucket.org/GutenkunstLab/dadi

dadi/cuda/Integration.py

@@ -788,10 +788,10 @@ def _five_pops_temporal_params(phi, xx, T, initial_t, nu1_f, nu2_f, nu3_f, nu4_f
     pBuffer = pycuda.driver.mem_alloc(bsize_int)
 
     while current_t < T:
-        dt = min(dadi.Integration._compute_dt(dx, nu1, [m12, m13, m14], gamma1, h1),
-                 dadi.Integration._compute_dt(dy, nu2, [m21, m23, m24], gamma2, h2),
-                 dadi.Integration._compute_dt(dz, nu3, [m31, m32, m34], gamma3, h3),
-                 dadi.Integration._compute_dt(da, nu4, [m41, m42, m43], gamma4, h4),
+        dt = min(dadi.Integration._compute_dt(dx, nu1, [m12,m13,m14,m15], gamma1, h1),
+                 dadi.Integration._compute_dt(dy, nu2, [m21,m23,m24,m25], gamma2, h2),
+                 dadi.Integration._compute_dt(dz, nu3, [m31,m32,m34,m35], gamma3, h3),
+                 dadi.Integration._compute_dt(da, nu4, [m41,m42,m43,m45], gamma4, h4),
                  dadi.Integration._compute_dt(db, nu5, [m51,m52,m53,m54],gamma5,h5))
         this_dt = np.float64(min(dt, T - current_t))
 

dadi/cuda/kernels.cu

@@ -144,21 +144,3 @@ __global__ void compute_bc_nobc(double *dx, double *dfactor,
         c[ii*M + jj] = -dfactor[ii]*ctemp;
     }
 }
-
-__global__ void diff_iny_2D(double *in, double *out, int L, int M){
-    int ii = (blockIdx.x*blockDim.x + threadIdx.x) / (M-1);
-    int jj = (blockIdx.x*blockDim.x + threadIdx.x) % (M-1);
-
-    if(ii < L){
-        out[ii*(M-1) + jj] = in[ii*M+jj+1] - in[ii*M+jj];
-    }
-}
-
-__global__ void drop_last_col_2D(double *in, double *out, int L, int M){
-    int ii = (blockIdx.x*blockDim.x + threadIdx.x) / (M-1);
-    int jj = (blockIdx.x*blockDim.x + threadIdx.x) % (M-1);
-
-    if(ii < L){
-        out[ii*(M-1) + jj] = in[ii*M+jj];
-    }
-}
\ No newline at end of file

dadi/cuda/kernels.py

@@ -17,6 +17,4 @@ _Mfunc5D = mod.get_function("Mfunc5D")
 _cx0 = mod.get_function("cx0")
 _compute_ab_nobc = mod.get_function("compute_ab_nobc")
 _compute_bc_nobc = mod.get_function("compute_bc_nobc")
-_include_bc = mod.get_function("include_bc")
-diff_iny_2D = mod.get_function("diff_iny_2D")
-drop_last_col_2D = mod.get_function("drop_last_col_2D")
\ No newline at end of file
+_include_bc = mod.get_function("include_bc")
\ No newline at end of file

tests/test_CUDA.py

@@ -213,6 +213,34 @@ class CUDATestCase(unittest.TestCase):
                                                              gamma1=gamma1, gamma2=gamma2, gamma3=gamma3, gamma4=gamma4,
                                                              h1=h1, h2=h2, h3=h3, h4=h4, theta0=theta0,
                                                              frozen1=f1, frozen2=f2, frozen3=f3, frozen4=f4)
+                        self.assertTrue(np.allclose(phi_cpu, phi_gpu))
+
+    def test_5d_integration(self):
+        kwargs = {'T': 0.1,
+                  'nu1': 0.2, 'nu2': 1.3, 'nu3': 7.1, 'nu4': 27.1, 'nu5':lambda t: 0.9-8*t,
+                  'm12': 3, 'm13': 2.9, 'm14': 0.9, 'm15': 10,
+                  'm21': 3, 'm23': lambda t: 0.9+10*t, 'm24': 0.9, 'm25': lambda t: 2-15*t,
+                  'm31': 3.5, 'm32': 2.9, 'm34': 0.9, 'm35': 10,
+                  'm41': 3.3, 'm42': 2.2, 'm43': 0.1, 'm45': 9,
+                  'm51': 3.3, 'm52': 2.2, 'm53': 0.8, 'm54': 9.2,
+                  'gamma1': -1, 'gamma2': 2, 'gamma3': -1.9, 'gamma4': lambda t: 10*t, 'gamma5': -1, 
+                  'h1': 0.1, 'h2': 0.9, 'h3': 0.2, 'h4': 0.9, 'h5': 0.1, 
+                  'frozen1':False, 'frozen2':False, 'frozen3':False, 'frozen4':False, 'frozen5':False,
+                  'theta0': lambda t: 10*t}
+
+        pts = 5
+        xx = dadi.Numerics.default_grid(pts)
+        phi = dadi.PhiManip.phi_1D(xx)
+        phi = dadi.PhiManip.phi_1D_to_2D(xx, phi)
+        phi = dadi.PhiManip.phi_2D_to_3D(phi, 0, xx,xx,xx)
+        phi = dadi.PhiManip.phi_3D_to_4D(phi, 0, 0, xx,xx,xx,xx)
+        phi = dadi.PhiManip.phi_4D_to_5D(phi, 0,0,0, xx,xx,xx,xx,xx)
+        dadi.cuda_enabled(True)
+        phi_gpu = dadi.Integration.five_pops(phi.copy(), xx, **kwargs)
+        dadi.cuda_enabled(False)
+        phi_cpu = dadi.Integration.five_pops(phi.copy(), xx, **kwargs)
+
+        self.assertTrue(np.allclose(phi_cpu, phi_gpu))
 
 if dadi.cuda_enabled(True):
     suite = unittest.TestLoader().loadTestsFromTestCase(CUDATestCase)