metal : consolidate unary ops (#19490)

This commit is contained in:
Georgi Gerganov
2026-02-11 07:51:12 +02:00
committed by GitHub
parent 2cce9fddb7
commit ceaa89b786
7 changed files with 377 additions and 671 deletions
+56 -19
View File
@@ -80,7 +80,8 @@
#define FC_SSM_CONV 900
#define FC_SOLVE_TRI 1000
#define FC_COUNT_EQUAL 1100
#define FC_BIN 1200
#define FC_UNARY 1200
#define FC_BIN 1300
// op-specific constants
#define OP_FLASH_ATTN_EXT_NQPSG 8
@@ -89,6 +90,35 @@
#define OP_FLASH_ATTN_EXT_VEC_NQPSG 1
#define OP_FLASH_ATTN_EXT_VEC_NCPSG 32
#define OP_UNARY_NUM_SCALE 10
#define OP_UNARY_NUM_FILL 11
#define OP_UNARY_NUM_CLAMP 12
#define OP_UNARY_NUM_SQR 13
#define OP_UNARY_NUM_SQRT 14
#define OP_UNARY_NUM_SIN 15
#define OP_UNARY_NUM_COS 16
#define OP_UNARY_NUM_LOG 17
#define OP_UNARY_NUM_LEAKY_RELU 18
#define OP_UNARY_NUM_TANH 100
#define OP_UNARY_NUM_RELU 101
#define OP_UNARY_NUM_SIGMOID 102
#define OP_UNARY_NUM_GELU 103
#define OP_UNARY_NUM_GELU_ERF 104
#define OP_UNARY_NUM_GELU_QUICK 105
#define OP_UNARY_NUM_SILU 106
#define OP_UNARY_NUM_ELU 107
#define OP_UNARY_NUM_NEG 108
#define OP_UNARY_NUM_ABS 109
#define OP_UNARY_NUM_SGN 110
#define OP_UNARY_NUM_STEP 111
#define OP_UNARY_NUM_HARDSWISH 112
#define OP_UNARY_NUM_HARDSIGMOID 113
#define OP_UNARY_NUM_EXP 114
#define OP_UNARY_NUM_SOFTPLUS 115
#define OP_UNARY_NUM_EXPM1 116
// kernel argument structs
//
// - element counters (e.g. ne00) typically use int32_t to reduce register usage
@@ -124,6 +154,31 @@ typedef struct {
int32_t dim;
} ggml_metal_kargs_concat;
typedef struct {
int32_t ne00;
int32_t ne01;
int32_t ne02;
int32_t ne03;
uint64_t nb00;
uint64_t nb01;
uint64_t nb02;
uint64_t nb03;
int32_t ne0;
int32_t ne1;
int32_t ne2;
int32_t ne3;
uint64_t nb0;
uint64_t nb1;
uint64_t nb2;
uint64_t nb3;
float slope;
float scale;
float bias;
float val;
float min;
float max;
} ggml_metal_kargs_unary;
typedef struct {
int32_t ne00;
int32_t ne01;
@@ -181,20 +236,6 @@ typedef struct {
uint64_t nb3;
} ggml_metal_kargs_repeat;
typedef struct {
float scale;
float bias;
} ggml_metal_kargs_scale;
typedef struct {
float val;
} ggml_metal_kargs_fill;
typedef struct {
float min;
float max;
} ggml_metal_kargs_clamp;
typedef struct {
int64_t nk0;
int64_t ne00;
@@ -881,10 +922,6 @@ typedef struct {
int max_period;
} ggml_metal_kargs_timestep_embedding;
typedef struct {
float slope;
} ggml_metal_kargs_leaky_relu;
typedef struct {
int32_t ne00;
int32_t ne01;