Why didn't mali864 perform better than mali764?
hello, I encountered a problem recently:
In EGL_BUFFER_PRESERVED model, under exactly the same hardware condition(CPU, memory, GPU frequency..), it appeared that mali864 only showed half the performance compared with mali764.
for mali864 I used android 6.0.1 in 64bit and the GPU driver version is r12p0-04rel0;
for mali764 I used android 4.4.2 in 32bit and the GPU driver version is r4p1-02dev0;
so, here are the questions:
1.what may be the reason why this happen?
2.would a newer GPU driver help?
3.what can I do to fix this?
Hi Yueliu,
I'm not aware of any specific known issues in this regard.
Mali-T860 may make a different memory access pattern to Mali-T760 when accessing data resources, which may place increased pressure on system components such as a System MMU in situations where the System MMU TLB is close to thrashing.
If you are a Mali licensee please raise a support request via support@arm.com and our support team will follow up (it's likely we'll need some data off your platform to be able to help more).
Cheers, Pete
hi Peter Harrispeterharris,
for mali864,I used RK3399 hardware platform.
for mali764,I used RK3288 hardware platform.
Yueliu
hi
I'm using libhybris to take advantage of android graphic libarary.But the implemention of libhybris will cause a segment fault while using thread mali-hist-dump,I just quit the thread to avoid this fault.I don't know if this will have an impact on performance is as we expect.
What's the actual failure you are getting? Just quitting the thread isn't actually fixing anything, so it's possible there is another error lurking. In particular that thread doesn't do much, so if it's faulting I'd be more generally worried about some form of memory corruption ...
Any chance of a backtrace?
Cheers,
Pete
Hello, I found it is on the function "malioc_release_outputs" that the segment fault occured by tracing PC address.How can I fix that?
Error infomation is as follow:
[ 87.949121] mali-hist-dump[251]: unhandled level 2 translation fault (11) at 0x00000000, esr 0x92000006
[ 87.949137] pgd = ffffffc0e9fb3000
[ 87.949142] [00000000] *pgd=00000000e9b21003, *pud=00000000e9b21003, *pmd=0000000000000000
[ 87.949153]
[ 87.949160] CPU: 5 PID: 251 Comm: mali-hist-dump Not tainted 4.4.16 #2
[ 87.949165] Hardware name: rockchip,rk3399-excavator-edp (DT)
[ 87.949171] task: ffffffc0e9bc9880 ti: ffffffc0e7814000 task.ti: ffffffc0e7814000
[ 87.949177] PC is at 0x7f843c5154
[ 87.949181] LR is at 0x7f843c5574
[ 87.949186] pc : [<0000007f843c5154>] lr : [<0000007f843c5574>] pstate: a0000000
[ 87.949190] sp : 0000007f7c5da450
[ 87.949195] x29: 0000007f7c5da7c0 x28: 0000000000000000
[ 87.949203] x27: 0000000000000000 x26: 0000007f852825ba
[ 87.949211] x25: 0000000000000000 x24: 0000000000000000
[ 87.949218] x23: 0000007f7c5da570 x22: 0000007f857f0000
[ 87.949226] x21: 00000000ffffffd8 x20: 0000007f852ace10
[ 87.949234] x19: 0000007f852825b9 x18: 0000000000000000
[ 87.949244] x17: 0000000000000073 x16: 0000000000000000
[ 87.949253] x15: 0000007f852acdb8 x14: 0000007f852ac000
[ 87.949263] x13: 000000000000000b x12: 00000000ffffffd0
[ 87.949272] x11: 0000000000000043 x10: 0000007f852825b9
[ 87.949283] x9 : 0000007f7c5da680 x8 : 0000007f7c5da6b0
[ 87.949293] x7 : 00000000ffffffd8 x6 : 0000007f852acdcc
[ 87.949300] x5 : 0000000000000000 x4 : 0000000000000000
[ 87.949308] x3 : 00000000ffffffff x2 : 0000007f843c4824
[ 87.949316] x1 : 0000007f843c4988 x0 : 0000000000000073
[ 87.949324]
prio:6 tag:mali text:==>[INIT] (eglimage) CDBG: In file: hardware/arm/maliT760/driver/product/cdbg/src/mali_cdbg_env.c line: 764 cdbgp_populate_from_system_environment
Initialization of a handle to the system environment failed (3)
This exception would influence the error above?
That message shouldn't impact anything.
However I'm pretty sure that if you're seeing those messages and the memory histogram at all, you're running a debug build of the driver, which may explain the performance issues you are seeing. Can you try a release build?
If you are a licensee (which would make sense if you are using debug builds of the driver), you make get a faster response to your questions via support@arm.com.