Attempt to fix Apple M-series and NLPAR gpu

Signed-off by: David Rowenhorst <[email protected]>
drowenhorst-nrl · Jun 10, 2024 · d742733 · d742733
1 parent a7162ac
commit d742733
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/pyebsdindex/_ebsd_index_parallel.py b/pyebsdindex/_ebsd_index_parallel.py
@@ -312,7 +312,7 @@ def index_pats_distributed(
         ngpupro = min(max(6, ngpu*6), 12)  # number of processes that will serve data to the gpu
         #ngpupro = 8
         if n_cpu_nodes < 8:
-            ngpupro = min(ngpupro,8)
+            ngpupro = min(ngpupro, n_cpu_nodes)
         if n_cpu_nodes < 2:
             ngpupro = 2
         #if OSPLATFORM == 'Linux':

diff --git a/pyebsdindex/opencl/nlpar_cl.py b/pyebsdindex/opencl/nlpar_cl.py
@@ -163,7 +163,7 @@ def calcsigma_cl(self,nn=1,saturation_protect=True,automask=True, normalize_d=Fa
     #print(gpu_id)
     clparams.get_context(gpu_id=gpu_id, kfile = 'clnlpar.cl')
     clparams.get_queue()
-    target_mem = clparams.queue.device.max_mem_alloc_size//2
+    target_mem = min(clparams.queue.device.max_mem_alloc_size//2, int(4e9))
     ctx = clparams.ctx
     prg = clparams.prg
     queue = clparams.queue
@@ -400,7 +400,7 @@ def calcnlpar_cl(self, searchradius=None, lam = None, dthresh = None, saturation
     #print(gpu_id)
     clparams.get_context(gpu_id=gpu_id, kfile ='clnlpar.cl')
     clparams.get_queue()
-    target_mem = clparams.queue.device.max_mem_alloc_size//2
+    target_mem = min(clparams.queue.device.max_mem_alloc_size//4, int(2e9))
     ctx = clparams.ctx
     prg = clparams.prg
     queue = clparams.queue