Avoid exceeding the configured thread count in x86_64 TOBF16 (#4748)

martin-frbg · web-flow · commit dd7efcf9efe8 · 2024-06-14T14:21:13.000+02:00
* avoid setting nthreads higher than available
diff --git a/kernel/x86_64/tobf16.c b/kernel/x86_64/tobf16.c
@@ -144,10 +144,11 @@ void CNAME(BLASLONG n, FLOAT_TYPE * in, BLASLONG inc_in, bfloat16 * out, BLASLON
     if (inc_in == 0 || inc_out == 0 || n <= 100000) {
         nthreads = 1;
     } else {
+        nthreads = num_cpu_avail(1);
         if (n/100000 < 100) {
-            nthreads = 4;
-        } else {
-            nthreads = 16;
+            nthreads = MAX(nthreads,4);
+//        } else {
+//            nthreads = MAX(nthreads,16);
         }
     }
 

Original file line number	Diff line number	Diff line change
`@@ -144,10 +144,11 @@ void CNAME(BLASLONG n, FLOAT_TYPE * in, BLASLONG inc_in, bfloat16 * out, BLASLON`
`144`	`144`	`if (inc_in == 0 \|\| inc_out == 0 \|\| n <= 100000) {`
`145`	`145`	`nthreads = 1;`
`146`	`146`	`} else {`
	`147`	`+ nthreads = num_cpu_avail(1);`
`147`	`148`	`if (n/100000 < 100) {`
`148`		`- nthreads = 4;`
`149`		`- } else {`
`150`		`- nthreads = 16;`
	`149`	`+ nthreads = MAX(nthreads,4);`
	`150`	`+// } else {`
	`151`	`+// nthreads = MAX(nthreads,16);`
`151`	`152`	`}`
`152`	`153`	`}`
`153`	`154`