[test] further testing of buit-in immc benchmarks

fangq · Mar 10, 2024 · 237713b · 237713b
1 parent 8e44f67
commit 237713b
Show file tree

Hide file tree

Showing 4 changed files with 27 additions and 4 deletions.
diff --git a/examples/validation/createmesh.m b/examples/validation/createmesh.m
@@ -6,5 +6,5 @@
 elem(:, 1:4) = meshreorient(node, elem(:, 1:4));
 
 srcpos = [30.1, 30.2, 0];
-savemmcmesh(sessionid, node, elem, []);
+savemmcmesh(sessionid, node, elem);
 eid = tsearchn(node, elem(:, 1:4), srcpos);
diff --git a/src/mmc_bench.c b/src/mmc_bench.c
@@ -280,7 +280,7 @@ MSTR(
          "DoMismatch":1,
          "DebugFlag":"TP",
          "RayTracer":"g",
-        "OutputFormat": "bin",
+         "OutputFormat": "bin",
          "Photons":1e6
      },
      "Domain":{
@@ -365,7 +365,7 @@ MSTR(
          "DoMismatch":1,
          "DebugFlag":"TP",
          "RayTracer":"g",
-        "OutputFormat": "bin",
+         "OutputFormat": "bin",
          "Photons":1e6
      },
      "Domain":{

diff --git a/src/mmc_cu_host.cu b/src/mmc_cu_host.cu
@@ -51,6 +51,18 @@ and the ray tracer precomputed data (tracer).
 ******************************************************************************/
 #define CUDA_ASSERT(a)                                                         \
     mcx_cu_assess((a), __FILE__, __LINE__) ///< macro to report CUDA error
+
+/**
+ * @brief Utility function to calculate the GPU stream processors (cores) per SM
+ *
+ * Obtain GPU core number per MP, this replaces
+ * ConvertSMVer2Cores() in libcudautils to avoid
+ * extra dependency.
+ *
+ * @param[in] v1: the major version of an NVIDIA GPU
+ * @param[in] v2: the minor version of an NVIDIA GPU
+ */
+
 int mcx_corecount(int v1, int v2) {
     int v = v1 * 10 + v2;
 
@@ -62,8 +74,10 @@ int mcx_corecount(int v1, int v2) {
         return 48;
     } else if (v < 50) {
         return 192;
-    } else {
+    } else if (v < 60 || v == 61 || v >= 89) {
         return 128;
+    } else {
+        return 64;
     }
 }
 

diff --git a/src/mmc_utils.c b/src/mmc_utils.c
@@ -1381,6 +1381,7 @@ int mcx_loadjson(cJSON* root, mcconfig* cfg) {
 
                     cfg->roitype = (collen == 6) ? rtEdge : ((collen == 1 || rowlen == 1) ? rtNode : (collen == 4 ? rtFace : rtNone));
                     cfg->roidata = (float*)malloc(sizeof(float) * rowlen * collen);
+                    cfg->implicit = (cfg->roitype != rtNone) + (cfg->roitype == rtFace);
 
                     for (int i = 0; i < rowlen; i++) {
                         if (cJSON_GetArraySize(subitem) != collen) {
@@ -2830,6 +2831,14 @@ void mcx_validatecfg(mcconfig* cfg) {
         cfg->basisorder = 0;
     }
 
+    if (cfg->implicit && (int)(cfg->gpuid) >= 0) {
+        MMC_ERROR(-2, "Implicit MMC is currently only supported in the CPU, please set -G -1 or cfg.gpuid=-1");
+    }
+
+    if (cfg->srcnum > 1 && (int)(cfg->gpuid) >= 0) {
+        MMC_ERROR(-2, "Photon-sharing MMC is currently only supported in the CPU, please set -G -1 or cfg.gpuid=-1");
+    }
+
     for (i = 0; i < MAX_DEVICE; i++)
         if (cfg->deviceid[i] == '0') {
             cfg->deviceid[i] = '\0';