feat(dnn): add elemwise modes

GitOrigin-RevId: edea1a48b0

dnn/include/megdnn/oprs/general.h

@@ -313,7 +313,6 @@ protected:
             size_t workspace_in_bytes);
 };
 using Cumsum = CumsumForward;
-
 // mxx can be max or min
 class ArgmxxBase : public OperatorBase {
     DEF_OPR_IMPL_CTOR(ArgmxxBase, OperatorBase);

dnn/scripts/gen_elemwise_multi_type_utils.py

@@ -48,6 +48,19 @@ MODES = {
         "H_SWISH",
         "SILU",
         "GELU",
+        "SINH",
+        "COSH",
+        "ASINH",
+        "ACOSH",
+        "ATANH",
+        "TAN",
+        "SOFTPLUS",
+        "RELU6",
+        "HSIGMOID",
+        "LOGSIGMOID",
+        "SQRT",
+        "SQUARE",
+        "SIGN",
     ],
     2: [
         "ABS_GRAD",
@@ -76,8 +89,15 @@ MODES = {
         "FUSE_ADD_H_SWISH",
         "SILU_GRAD",
         "GELU_GRAD",
+        "PRELU",
+        "ASINH_GRAD",
+        "ACOSH_GRAD",
+        "ATANH_GRAD",
+        "SOFTPLUS_GRAD",
+        "RELU6_GRAD",
+        "HSIGMOID_GRAD",
     ],
-    3: ["COND_LEQ_MOV", "COND_LT_MOV", "FUSE_MUL_ADD3"],
+    3: ["COND_LEQ_MOV", "COND_LT_MOV", "FUSE_MUL_ADD3", "CLIP", "PRELU_GRAD"],
 }
 
 QINT4_MODES = {
@@ -107,8 +127,9 @@ QINT4_MODES = {
         "FUSE_ADD_TANH",
         "FUSE_ADD_SIGMOID",
         "FUSE_ADD_H_SWISH",
+        "PRELU",
     ],
-    3: ["COND_LEQ_MOV", "COND_LT_MOV", "FUSE_MUL_ADD3"],
+    3: ["COND_LEQ_MOV", "COND_LT_MOV", "FUSE_MUL_ADD3", "CLIP"],
 }
 
 QINT32_MODES = {

dnn/scripts/gen_elemwise_utils.py

@@ -12,7 +12,7 @@ DTYPES = {
 }
 
 MODES = {
-    (1, "INT"): ["RELU", "ABS", "NEGATE"],
+    (1, "INT"): ["RELU", "ABS", "NEGATE", "RELU6", "SQUARE", "SIGN"],
     (2, "INT"): [
         "ABS_GRAD",
         "ADD",
@@ -32,8 +32,9 @@ MODES = {
         "SHL",
         "SHR",
         "RMULH",
+        "PRELU",
     ],
-    (3, "INT"): ["COND_LEQ_MOV", "COND_LT_MOV"],
+    (3, "INT"): ["COND_LEQ_MOV", "COND_LT_MOV", "CLIP"],
     (1, "FLOAT"): [
         "RELU",
         "ABS",
@@ -59,6 +60,19 @@ MODES = {
         "H_SWISH",
         "SILU",
         "GELU",
+        "SINH",
+        "COSH",
+        "ASINH",
+        "ACOSH",
+        "ATANH",
+        "TAN",
+        "SOFTPLUS",
+        "RELU6",
+        "HSIGMOID",
+        "LOGSIGMOID",
+        "SQRT",
+        "SQUARE",
+        "SIGN",
     ],
     (2, "FLOAT"): [
         "ABS_GRAD",
@@ -87,8 +101,21 @@ MODES = {
         "FUSE_ADD_H_SWISH",
         "SILU_GRAD",
         "GELU_GRAD",
+        "PRELU",
+        "ASINH_GRAD",
+        "ACOSH_GRAD",
+        "ATANH_GRAD",
+        "SOFTPLUS_GRAD",
+        "RELU6_GRAD",
+        "HSIGMOID_GRAD",
+    ],
+    (3, "FLOAT"): [
+        "COND_LEQ_MOV",
+        "COND_LT_MOV",
+        "FUSE_MUL_ADD3",
+        "CLIP",
+        "PRELU_GRAD",
     ],
-    (3, "FLOAT"): ["COND_LEQ_MOV", "COND_LT_MOV", "FUSE_MUL_ADD3"],
     (1, "BOOL"): ["NOT"],
     (2, "BOOL"): ["AND", "OR", "XOR", "LT", "LEQ", "EQ"],
     (3, "BOOL"): [],

dnn/scripts/opr_param_defs.py

@@ -424,6 +424,28 @@ pdef('Elemwise').add_enum(
     Doc('NEQ = 61', 'binary: x != y'),
     Doc('ISNAN = 62', 'unary: isnan(x)'),
     Doc('ISINF = 63', 'unary: isinf(x)'),
+    Doc('SINH = 64', 'unary: sinh(x)'),
+    Doc('COSH = 65', 'unary: cosh(x)'),
+    Doc('ASINH = 66', 'unary: asinh(x)'),
+    Doc('ACOSH = 67', 'unary: acosh(x)'),
+    Doc('ATANH = 68', 'unary: atanh(x)'),
+    Doc('TAN = 69', 'unary: tan(x)'),
+    Doc('ASINH_GRAD = 70', 'binary: y / sqrt(x^2 + 1)'),
+    Doc('ACOSH_GRAD = 71', 'binary: y / sqrt(x^2 - 1) (x > 1)'),
+    Doc('ATANH_GRAD = 72', 'binary: y / (1 - x^2) (|x| < 1)'),
+    Doc('PRELU = 73', 'binary: x > 0 ? x : x * y'),
+    Doc('CLIP = 74', 'ternary: x <= y ? y : (x <= z ? x : z)'),
+    Doc('PRELU_GRAD = 75', 'ternary: x > 0 ? y : y * z'),
+    Doc('SOFTPLUS = 76', 'unary: log(1 + e^x)'),
+    Doc('SOFTPLUS_GRAD = 77', 'binary: y * e^x / (1 + e^x)'),
+    Doc('RELU6 = 78', 'unary: min(max(0, x), 6)'),
+    Doc('RELU6_GRAD = 79', 'binary: x < 0 ? 0 : (x > 6 ? 0 : y)'),
+    Doc('HSIGMOID = 80', 'unary: relu6(x + 3) / 6'),
+    Doc('HSIGMOID_GRAD = 81', 'binary: x < -3 ? 0 : (x > 3 ? 0 : y / 6)'),
+    Doc('LOGSIGMOID = 82', 'unary: -log(1 + e^(-x))'),
+    Doc('SQRT = 83', 'unary: x^(1/2)'),
+    Doc('SQUARE = 84', 'unary: x^2'),
+    Doc('SIGN = 85', 'unary: sgn(x)'),
 )
 
 pdef('ElemwiseMultiType').add_enum(

dnn/src/common/elemwise/each_mode.inl

@@ -25,12 +25,28 @@
     MEGDNN_ELEMWISE_MODE_ENABLE(ERFCINV, cb)         \
     MEGDNN_ELEMWISE_MODE_ENABLE(H_SWISH, cb)         \
     MEGDNN_ELEMWISE_MODE_ENABLE(SILU, cb)            \
-    MEGDNN_ELEMWISE_MODE_ENABLE(GELU, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(GELU, cb)            \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SINH, cb)            \
+    MEGDNN_ELEMWISE_MODE_ENABLE(COSH, cb)            \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ASINH, cb)           \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH, cb)           \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ATANH, cb)           \
+    MEGDNN_ELEMWISE_MODE_ENABLE(TAN, cb)             \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS, cb)        \
+    MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)           \
+    MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID, cb)        \
+    MEGDNN_ELEMWISE_MODE_ENABLE(LOGSIGMOID, cb)      \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SQRT, cb)            \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)          \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_UNARY_INT(cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(RELU, cb)          \
     MEGDNN_ELEMWISE_MODE_ENABLE(ABS, cb)           \
-    MEGDNN_ELEMWISE_MODE_ENABLE(NEGATE, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(NEGATE, cb)        \
+    MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)         \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)        \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_BOOL(cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(AND, cb)             \
@@ -66,7 +82,14 @@
     MEGDNN_ELEMWISE_MODE_ENABLE(H_SWISH_GRAD, cb)     \
     MEGDNN_ELEMWISE_MODE_ENABLE(FUSE_ADD_H_SWISH, cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(SILU_GRAD, cb)        \
-    MEGDNN_ELEMWISE_MODE_ENABLE(GELU_GRAD, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(GELU_GRAD, cb)        \
+    MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)            \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ASINH_GRAD, cb)       \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH_GRAD, cb)       \
+    MEGDNN_ELEMWISE_MODE_ENABLE(ATANH_GRAD, cb)       \
+    MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS_GRAD, cb)    \
+    MEGDNN_ELEMWISE_MODE_ENABLE(RELU6_GRAD, cb)       \
+    MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID_GRAD, cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_INT(cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(ABS_GRAD, cb)       \
@@ -86,15 +109,19 @@
     MEGDNN_ELEMWISE_MODE_ENABLE(FUSE_ADD_RELU, cb)  \
     MEGDNN_ELEMWISE_MODE_ENABLE(SHL, cb)            \
     MEGDNN_ELEMWISE_MODE_ENABLE(SHR, cb)            \
-    MEGDNN_ELEMWISE_MODE_ENABLE(RMULH, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(RMULH, cb)          \
+    MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_TERNARY_BOOL(cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_TERNARY_FLOAT(cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(COND_LEQ_MOV, cb)      \
     MEGDNN_ELEMWISE_MODE_ENABLE(COND_LT_MOV, cb)       \
-    MEGDNN_ELEMWISE_MODE_ENABLE(FUSE_MUL_ADD3, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(FUSE_MUL_ADD3, cb)     \
+    MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)              \
+    MEGDNN_ELEMWISE_MODE_ENABLE(PRELU_GRAD, cb)
 
 #define MEGDNN_FOREACH_ELEMWISE_MODE_TERNARY_INT(cb) \
     MEGDNN_ELEMWISE_MODE_ENABLE(COND_LEQ_MOV, cb)    \
-    MEGDNN_ELEMWISE_MODE_ENABLE(COND_LT_MOV, cb)
+    MEGDNN_ELEMWISE_MODE_ENABLE(COND_LT_MOV, cb)     \
+    MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)

dnn/src/common/elemwise/kern_defs.cuh

@@ -154,11 +154,18 @@ struct ElemwiseKern;
 
 // int and float
 DEF_KERN_ALL(NEGATE, -x);
+DEF_KERN_ALL(SQUARE, x* x);
 #if defined(__HIP_PLATFORM_HCC__) && !defined(__HIP_PLATFORM_NVCC__)
 DEF_KERN_INT(RELU, x <= ctype(0) ? ctype(0) : x);
+DEF_KERN_INT(RELU6, x <= ctype(0) ? ctype(0) : (x <= ctype(6) ? x : ctype(6)));
+DEF_KERN_INT(SIGN, x < ctype(0) ? ctype(-1) : (x > ctype(0) ? ctype(1) : ctype(0)));
 DEF_KERN_FLOAT(RELU, x <= 0.f ? ctype(0) : x);
+DEF_KERN_FLOAT(RELU6, x <= 6.f ? ctype(0) : (x <= 6.f ? x : ctype(6)));
+DEF_KERN_FLOAT(SIGN, x < 0.f ? -1.f : (x > 0.f ? 1.f : 0.f));
 #else
 DEF_KERN_ALL(RELU, x <= ctype(0) ? ctype(0) : x);
+DEF_KERN_ALL(RELU6, x <= ctype(0) ? ctype(0) : (x <= ctype(6) ? x : ctype(6)));
+DEF_KERN_ALL(SIGN, x < ctype(0) ? ctype(-1) : (x > ctype(0) ? ctype(1) : ctype(0)));
 #endif
 DEF_KERN_INT(ABS, abs(int(x)));
 // DEF_KERN_INT(ABS, x > ctype(0) ? x : -x);
@@ -186,6 +193,18 @@ DEF_KERN_FLOAT(ERFCINV, erfcinvf(x));
 DEF_KERN_FLOAT(H_SWISH, x* min(max(x + 3, 0.f), 6.f) * (1.f / 6.f));
 DEF_KERN_FLOAT(SILU, x / (expf(-x) + 1.f));
 DEF_KERN_FLOAT(GELU, x* normcdf(x));
+DEF_KERN_FLOAT(SINH, sinhf(x));
+DEF_KERN_FLOAT(COSH, coshf(x));
+DEF_KERN_FLOAT(ASINH, asinhf(x));
+DEF_KERN_FLOAT(ACOSH, acoshf(x));
+DEF_KERN_FLOAT(ATANH, atanhf(x));
+DEF_KERN_FLOAT(TAN, tanf(x));
+DEF_KERN_FLOAT(SOFTPLUS, log1pf(expf(-fabsf(x))) + (x <= ctype(0) ? ctype(0) : x));
+DEF_KERN_FLOAT(
+        HSIGMOID,
+        x <= ctype(-3) ? ctype(0) : (x >= ctype(3) ? ctype(1) : ((x + 3.f) / 6.f)));
+DEF_KERN_FLOAT(SQRT, sqrtf(x));
+DEF_KERN_FLOAT(LOGSIGMOID, -log1pf(expf(-fabsf(x))) + (x >= ctype(0) ? ctype(0) : x));
 
 // int only
 DEF_KERN(dt_bool, NOT, x ^ 1);
@@ -240,6 +259,12 @@ DEF_KERN_FLOAT(FUSE_ADD_RELU, (x + y) <= 0.f ? ctype(0) : (x + y));
 #else
 DEF_KERN_ALL(FUSE_ADD_RELU, (x + y) <= ctype(0) ? ctype(0) : (x + y));
 #endif
+#if defined(__HIP_PLATFORM_HCC__) && !defined(__HIP_PLATFORM_NVCC__)
+DEF_KERN_INT(PRELU, x > ctype(0) ? x : (x * y));
+DEF_KERN_FLOAT(PRELU, x > 0.f ? x : (x * y));
+#else
+DEF_KERN_ALL(PRELU, x > ctype(0) ? x : (x * y));
+#endif
 
 // float only
 DEF_KERN_FLOAT(TRUE_DIV, x / y);
@@ -259,6 +284,14 @@ DEF_KERN_FLOAT(
 DEF_KERN_FLOAT(FUSE_ADD_H_SWISH, fuse_add_hswish(x, y));
 DEF_KERN_FLOAT(SILU_GRAD, silu_grad(x, y));
 DEF_KERN_FLOAT(GELU_GRAD, gelu_grad(x, y));
+DEF_KERN_FLOAT(ASINH_GRAD, y / sqrt(x * x + 1.f));
+DEF_KERN_FLOAT(ACOSH_GRAD, y / sqrt(x * x - 1.f));
+DEF_KERN_FLOAT(ATANH_GRAD, y / (1.f - x * x));
+DEF_KERN_FLOAT(SOFTPLUS_GRAD, y* expf(x) / (1.f + expf(x)));
+DEF_KERN_FLOAT(RELU6_GRAD, x <= ctype(0) ? ctype(0) : (x >= ctype(6) ? ctype(0) : y));
+DEF_KERN_FLOAT(
+        HSIGMOID_GRAD,
+        x <= ctype(-3) ? ctype(0) : (x >= ctype(3) ? ctype(0) : (y / 6.f)));
 #undef KERN_SIG
 
 /* ================== ternary kernels ================== */
@@ -268,6 +301,8 @@ DEF_KERN_FLOAT(GELU_GRAD, gelu_grad(x, y));
 DEF_KERN_ALL(COND_LEQ_MOV, x <= y ? z : ctype(0));
 DEF_KERN_ALL(COND_LT_MOV, x < y ? z : ctype(0));
 DEF_KERN_ALL(FUSE_MUL_ADD3, x* y + z);
+DEF_KERN_ALL(CLIP, x <= y ? y : (x <= z ? x : z));
+DEF_KERN_FLOAT(PRELU_GRAD, x >= 0.f ? y : (y * z));
 
 #undef KERN_SIG
 

dnn/src/common/elemwise/opr_impl.cpp

@@ -62,6 +62,9 @@ const ModeTrait& ModeTrait::from_mode(Mode mode) {
         MEGDNN_FOREACH_ELEMWISE_MODE_UNARY_FLOAT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_FLOAT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_TERNARY_FLOAT(cb);
+        cb(NEQ);
+        cb(ISNAN);
+        cb(ISINF);
 #undef cb
 
 #define cb(_m)                                                  \
@@ -84,11 +87,14 @@ const ModeTrait& ModeTrait::from_mode(Mode mode) {
         MEGDNN_FOREACH_ELEMWISE_MODE_UNARY_FLOAT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_UNARY_INT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_UNARY_BOOL(cb);
+        cb(ISNAN);
+        cb(ISINF);
 #undef _a
 #define _a 2
         MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_FLOAT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_INT(cb);
         MEGDNN_FOREACH_ELEMWISE_MODE_BINARY_BOOL(cb);
+        cb(NEQ);
 #undef _a
 #define _a 3
         MEGDNN_FOREACH_ELEMWISE_MODE_TERNARY_FLOAT(cb);
@@ -223,6 +229,28 @@ const ModeTrait& ModeTrait::from_mode(Mode mode) {
         CB_MODE(Mode::GELU);
         CB_MODE(Mode::GELU_GRAD);
         CB_MODE(Mode::COND_LT_MOV);
+        CB_MODE(Mode::SINH);
+        CB_MODE(Mode::COSH);
+        CB_MODE(Mode::ASINH);
+        CB_MODE(Mode::ACOSH);
+        CB_MODE(Mode::ATANH);
+        CB_MODE(Mode::TAN);
+        CB_MODE(Mode::ASINH_GRAD);
+        CB_MODE(Mode::ACOSH_GRAD);
+        CB_MODE(Mode::ATANH_GRAD);
+        CB_MODE(Mode::PRELU);
+        CB_MODE(Mode::PRELU_GRAD);
+        CB_MODE(Mode::CLIP);
+        CB_MODE(Mode::SOFTPLUS);
+        CB_MODE(Mode::SOFTPLUS_GRAD);
+        CB_MODE(Mode::RELU6);
+        CB_MODE(Mode::RELU6_GRAD);
+        CB_MODE(Mode::HSIGMOID);
+        CB_MODE(Mode::HSIGMOID_GRAD);
+        CB_MODE(Mode::LOGSIGMOID);
+        CB_MODE(Mode::SQRT);
+        CB_MODE(Mode::SQUARE);
+        CB_MODE(Mode::SIGN);
         default:
             megdnn_assert(
                     0,

dnn/src/cuda/elemwise/kimpl/ACOSH_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ACOSH_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ACOSH_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/ACOSH_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ACOSH_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ACOSH_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/ASINH_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ASINH_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ASINH_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/ASINH_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ASINH_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ASINH_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/ATANH_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ATANH_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ATANH_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/ATANH_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ATANH_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/ATANH_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/CLIP_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/CLIP_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/CLIP_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/CLIP_dt_int16.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_int16
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/CLIP_dt_int32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_int32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/CLIP_dt_int8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_int8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/CLIP_dt_uint8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_uint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/COSH_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(COSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/COSH_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(COSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/COSH_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(COSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/HSIGMOID_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/HSIGMOID_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/HSIGMOID_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/HSIGMOID_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/HSIGMOID_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/HSIGMOID_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(HSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/LOGSIGMOID_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(LOGSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/LOGSIGMOID_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(LOGSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/LOGSIGMOID_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(LOGSIGMOID, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU_GRAD, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/PRELU_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU_GRAD, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/PRELU_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU_GRAD, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/PRELU_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/PRELU_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_dt_int16.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_int16
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_dt_int32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_int32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_dt_int8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_int8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/PRELU_dt_uint8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(PRELU, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_uint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/RELU6_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/RELU6_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/RELU6_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/RELU6_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_dt_int16.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int16
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_dt_int32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_dt_int8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/RELU6_dt_uint8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(RELU6, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_uint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SIGN_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SIGN_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SIGN_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SIGN_dt_int16.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int16
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SIGN_dt_int32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SIGN_dt_int8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SIGN_dt_uint8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SIGN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_uint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SINH_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SINH_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SINH_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_GRAD_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_GRAD_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_GRAD_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SOFTPLUS_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SOFTPLUS, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQRT_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQRT, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SQRT_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQRT, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SQRT_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQRT, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_int16.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int16
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_int32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_int8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_int8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/SQUARE_dt_uint8.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(SQUARE, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_uint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise/kimpl/TAN_dt_bfloat16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(TAN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_bfloat16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/TAN_dt_float16.cu

@@ -0,0 +1,7 @@
+// generated by gen_elemwise_kern_impls.py
+#if !MEGDNN_DISABLE_FLOAT16
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(TAN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float16
+#include "../kern_impl.inl"
+#endif

dnn/src/cuda/elemwise/kimpl/TAN_dt_float32.cu

@@ -0,0 +1,5 @@
+// generated by gen_elemwise_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(TAN, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_CTYPE    dt_float32
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ACOSH_GRAD_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ACOSH_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ACOSH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ASINH_GRAD_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ASINH_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ASINH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ATANH_GRAD_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH_GRAD, cb)
+#define KERN_IMPL_ARITY    2
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/ATANH_dt_qint8_dt_qint8.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(ATANH, cb)
+#define KERN_IMPL_ARITY    1
+#define KERN_IMPL_STYPE    dt_qint8
+#define KERN_IMPL_DTYPE    dt_qint8
+#include "../kern_impl.inl"

dnn/src/cuda/elemwise_multi_type/kimpl/CLIP_dt_qint4_dt_qint4.cu

@@ -0,0 +1,6 @@
+// generated by gen_elemwise_multi_type_kern_impls.py
+#define KERN_IMPL_MODE(cb) MEGDNN_ELEMWISE_MODE_ENABLE(CLIP, cb)
+#define KERN_IMPL_ARITY    3
+#define KERN_IMPL_STYPE    dt_qint4
+#define KERN_IMPL_DTYPE    dt_qint4
+#include "../kern_impl_q4.inl"