Call __fmm_release with gpu id and not index

__fmm_release expects gpu_id as parameter. fmm_release passed it instead the
index of the GPU in the gpu_mem array.

Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>


[ROCm/ROCR-Runtime commit: aba396bc52]

projects/rocr-runtime/src/fmm.c

@@ -627,7 +627,7 @@ static void __fmm_release(uint32_t gpu_id, void *address,
 	pthread_mutex_lock(&aperture->fmm_mutex);
 
 	/* Find the object to retrieve the handle */
-	object = vm_find_object_by_address(aperture, address, 0);
+	object = vm_find_object_by_address(aperture, address, MemorySizeInBytes);
 	if (!object) {
 		pthread_mutex_unlock(&aperture->fmm_mutex);
 		return;
@@ -654,7 +654,7 @@ void fmm_release(void *address, uint64_t MemorySizeInBytes)
 		if (address >= gpu_mem[i].gpuvm_aperture.base &&
 			address <= gpu_mem[i].gpuvm_aperture.limit) {
 			found = true;
-			__fmm_release(i, address, MemorySizeInBytes);
+			__fmm_release(gpu_mem[i].gpu_id, address, MemorySizeInBytes);
 			fmm_print(gpu_mem[i].gpu_id);
 		}
 	}