HIP: force max threads per block to be 1024

Some old compilers still use 256. Explicitly set it to 1024 to get correct
result from ops like ARGMAX and GROUP_NORM.

Related: #10610, #11619
Signed-off-by: fxzjshm <fxzjshm@163.com>

ggml/src/ggml-hip/CMakeLists.txt

@@ -40,6 +40,9 @@ find_package(hip     REQUIRED)
 find_package(hipblas REQUIRED)
 find_package(rocblas REQUIRED)
 
+# Workaround old compilers
+set(CMAKE_HIP_FLAGS "${CMAKE_HIP_FLAGS} --gpu-max-threads-per-block=1024")
+
 if (${hip_VERSION} VERSION_LESS 5.5)
     message(FATAL_ERROR "At least ROCM/HIP V5.5 is required")
 endif()