--attr=irmode=down;oscale=per_oc:2.25;post_ops='sum:1.5;relu'
--cfg=u8s8u8s32 --batch=conv_vgg_19
--cfg=u8s8f32s32 --batch=conv_googlenet_v2
+--cfg=u8s8s32s32 --batch=conv_tails
--cfg=s8s8u8s32 --batch=conv_vgg_19
--cfg=s8s8f32s32 --batch=conv_googlenet_v2
+--cfg=s8s8s32s32 --batch=conv_tails
--dir=FWD_D
--attr=irmode=nearest;oscale=common:2.25;post_ops='sum:1.5'
--cfg=u8s8s8s32 --batch=conv_googlenet_v3
--cfg=u8s8s32s32 --batch=conv_alexnet
+--cfg=u8s8s32s32 --batch=conv_tails
--cfg=s8s8s8s32 --batch=conv_googlenet_v3
--cfg=s8s8s32s32 --batch=conv_alexnet
+--cfg=s8s8s32s32 --batch=conv_tails
# f32
--reset --cfg=f32
--mb=2
--skip-impl="ref:gemm" # ! test jit version only
--allow-unimpl=true
---dir=FWD_B --attr=post_ops='sum;relu' --batch=conv_resnet_50
---dir=FWD_B --attr=post_ops='sum;relu' --batch=conv_3d
---dir=FWD_B --attr=post_ops='sum;relu' --batch=conv_1d
+
+--dir=FWD_B
+--attr=post_ops='sum;relu' --batch=conv_resnet_50
+--attr=post_ops='sum;relu:0.5' --batch=conv_tails
+--attr=post_ops='sum;tanh' --batch=conv_tails
+--attr=post_ops='sum;elu:0.5' --batch=conv_tails
+--attr=post_ops='sum;abs' --batch=conv_tails
+--attr=post_ops='sum;sqrt' --batch=conv_tails
+--attr=post_ops='sum;linear:0.5:1.5' --batch=conv_tails
+--attr=post_ops='sum;brelu:0.5' --batch=conv_tails
+--attr=post_ops='sum;logistic' --batch=conv_tails
+--cfg=f32_no_limits # square and srelu might overrun int_max_exact
+--attr=post_ops='sum;square' --batch=conv_tails
+--attr=post_ops='sum;srelu' --batch=conv_tails
# f32_wino
--reset --alg=wino --cfg=f32_wino
--attr=irmode=nearest;oscale=common:2.25;post_ops='sum:1.5'
--cfg=u8s8s8s32_wino --batch=conv_googlenet_v3
--cfg=u8s8s32s32_wino --batch=conv_resnet_50
+
+# i8 conv + f32 leaky relu
+--reset --dir=FWD_B --mb=2
+--skip-impl="ref:gemm" # ! test jit version only
+--allow-unimpl=true
+--attr=post_ops='relu:0.5'
+--cfg=s8s8f32s32 --batch=conv_yolov2
+--cfg=u8s8f32s32 --batch=conv_yolov2
+--attr=post_ops='relu:0.5;sum'
+--cfg=s8s8f32s32 --batch=conv_yolov2
+--cfg=u8s8f32s32 --batch=conv_yolov2
+--attr=post_ops='sum;relu:0.5'
+--cfg=s8s8f32s32 --batch=conv_yolov2
+--cfg=u8s8f32s32 --batch=conv_yolov2