[Bug 101325] UE4Editor crash after pressing "play" with radeon southern island card (7850 HD)

[bugzilla-daemon@freedesktop.org]

[-- Attachment #1.1: Type: text/plain, Size: 2768 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=101325

--- Comment #10 from Julien Isorce <julien.isorce@gmail.com> ---
Hi Luke,

I can reproduce the ring stalled issue using your apitrace 100% of the time.

In apitrace output I could notice:
warning: extension `GL_EXT_gpu_shader4' unsupported in fragment shader
message: api issue 4: FBO incomplete: no attachments and default width or
height is 0 [-1]
message: api performance issue 5: using glBufferSubData(buffer 1599, offset 0,
size 288) to update a GL_STATIC_DRAW buffer
message: shader compiler issue 2: LLVM diagnostic (remark): <unknown>:0:0: 895
instructions in function
message: shader compiler issue 3: too many identical messages; ignoring

Using GALLIUM_DDEBUG and check_vm I could get some msg like:


[1;32mPS - Constant buffer slot 4 (GPU list):ESC[0m
        ESC[1;33mSQ_BUF_RSRC_WORD0ESC[0m <- 0x000de0e0
        ESC[1;33mSQ_BUF_RSRC_WORD1ESC[0m <- BASE_ADDRESS_HI = 0
                             STRIDE = 7200 (0x1c20)
                             CACHE_SWIZZLE = 0
                             SWIZZLE_ENABLE = 0
        ESC[1;33mSQ_BUF_RSRC_WORD2ESC[0m <- 0x4000c007
        ESC[1;33mSQ_BUF_RSRC_WORD3ESC[0m <- DST_SEL_X = SQ_SEL_X
                             DST_SEL_Y = SQ_SEL_0
                             DST_SEL_Z = SQ_SEL_0
                             DST_SEL_W = SQ_SEL_1
                             NUM_FORMAT = BUF_NUM_FORMAT_UNORM
                             DATA_FORMAT = BUF_DATA_FORMAT_INVALID
                             ELEMENT_SIZE = 2
                             INDEX_STRIDE = 2
                             ADD_TID_ENABLE = 1
                             ATC = 0
                             HASH_ENABLE = 0
                             HEAP = 0
                             MTYPE = 2
                             TYPE = SQ_RSRC_BUF_RSVD_2
                             USER_VM_ENABLE = 0
                             USER_VM_MODE = 1
                             NV = 0
!!!!! This slot was corrupted in GPU memory !!!!!

I am using lastest git mesa, xserver, ddx, llvm and amd-staging-4.9 kernel.

Similar result with amdgpu though it does not detect the gpu hang, all I can
see is with radeontop, the graphic pipe is stuck at 100% for ever, but no error
in dmesg.

And indeed R600_DEBUG=nowc does not help in that case.

I have the feeling that this is a llvm issue due to exotic shaders but only
maintainers will be able to help you at this point. 

Providing this is 100% reproductible on SI chip class (at least) on both radeon
and amdgpu drivers, it would be intersting to know if it happens on a more
recent graphic card too.

-- 
You are receiving this mail because:
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 3657 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel