On 6/24/2020 3:05 PM, Alex Deucher wrote:
[CAUTION: External Email] The failures with ROCm only happen with noretry=1, so enable runtime pm when noretry=0 (the current default). Signed-off-by: Alex Deucher <alexander.deuc...@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 0fec39eed164..341d072edd95 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -176,7 +176,6 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags) case CHIP_BONAIRE: case CHIP_HAWAII: #endif - case CHIP_VEGA10: case CHIP_VEGA20: case CHIP_ARCTURUS: case CHIP_SIENNA_CICHLID: @@ -184,6 +183,11 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags) if (amdgpu_runtime_pm > 0) adev->runpm = true; break; + case CHIP_VEGA10: + /* turn runpm on if noretry=0 */ + if (!amdgpu_noretry) + adev->runpm = true; + break;
Though it fixes the ROCm pytorch issue but aren't there any stability and performance optimization concerns as it will impact recoverable page faults?
I have no objection to this otherwise. + felix Acked-by: Rajneesh Bhardwaj <rajneesh.bhard...@amd.com>
default: /* enable runpm on VI+ */ adev->runpm = true; -- 2.25.4 _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Famd-gfx&data=02%7C01%7Crajneesh.bhardwaj%40amd.com%7Cc985ef0414bd41b48eb508d8187196ed%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637286223437553622&sdata=wRJbu3%2F3zu%2BHZ3KA%2FZmyh1yhgATM2zONRr%2FvI5KsxrM%3D&reserved=0
_______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx