Merge pull request #3964 from asmorkalov:as/threshold_otsu_fix

asmorkalov · web-flow · commit d943e1d61c8b · 2025-06-27T18:40:05.000+03:00
Fixed out-of-bound access in CUDA Otsu threshold implementation.
diff --git a/modules/cudaarithm/src/cuda/threshold.cu b/modules/cudaarithm/src/cuda/threshold.cu
@@ -178,7 +178,7 @@ otsu_score(uint *otsu_threshold, uint *threshold_sums, float2 *variance)
 {
     const uint32_t n_thresholds = 256;
 
-    __shared__ float shared_memory[n_thresholds / WARP_SIZE];
+    __shared__ float shared_memory[n_thresholds];
 
     int threshold = threadIdx.x;
 

Original file line number	Diff line number	Diff line change
`@@ -178,7 +178,7 @@ otsu_score(uint otsu_threshold, uint threshold_sums, float2 *variance)`
`178`	`178`	`{`
`179`	`179`	`const uint32_t n_thresholds = 256;`
`180`	`180`
`181`		`- __shared__ float shared_memory[n_thresholds / WARP_SIZE];`
	`181`	`+ __shared__ float shared_memory[n_thresholds];`
`182`	`182`
`183`	`183`	`int threshold = threadIdx.x;`
`184`	`184`