/src/llama.cpp/ggml/src/ggml-cpu/common.h

Source
#pragma once

#include "ggml.h"
#include "traits.h"
#include "ggml-cpu-impl.h"
#include "ggml-impl.h"
#include "simd-mappings.h"

#ifdef __cplusplus

#include <utility>

// convenience functions/macros for use in template calls
// note: these won't be required after the 'traits' lookup table is used.
static inline ggml_fp16_t f32_to_f16(float x) {
    return GGML_CPU_FP32_TO_FP16(x);
}

static inline float f16_to_f32(ggml_fp16_t x) {
    return GGML_CPU_FP16_TO_FP32(x);
}

static inline ggml_bf16_t f32_to_bf16(float x) {
    return GGML_FP32_TO_BF16(x);
}

static inline float bf16_to_f32(ggml_bf16_t x) {
    return GGML_BF16_TO_FP32(x);
}

static inline float i32_to_f32(int32_t x) {
    return x;
}

static inline int32_t f32_to_i32(float x) {
    return x;
}

static inline float f32_to_f32(float x) {
    return x;
}

// TODO - merge this into the traits table, after using row-based conversions
template <class T>
struct type_conversion_table;

template <>
struct type_conversion_table<ggml_fp16_t> {
    static constexpr float (*to_f32)(ggml_fp16_t) = f16_to_f32;
    static constexpr ggml_fp16_t (*from_f32)(float) = f32_to_f16;
};

template <>
struct type_conversion_table<float> {
    static constexpr float (*to_f32)(float) = f32_to_f32;
    static constexpr float (*from_f32)(float) = f32_to_f32;
};

template <>
struct type_conversion_table<ggml_bf16_t> {
    static constexpr float (*to_f32)(ggml_bf16_t) = bf16_to_f32;
    static constexpr ggml_bf16_t (*from_f32)(float) = f32_to_bf16;
};

template <>
struct type_conversion_table<int32_t> {
    static constexpr float (*to_f32)(int32_t) = i32_to_f32;
    static constexpr int32_t (*from_f32)(float) = f32_to_i32;
};

static std::pair<int64_t, int64_t> get_thread_range(const struct ggml_compute_params * params, const struct ggml_tensor * src0) {
    const int64_t ith = params->ith;
    const int64_t nth = params->nth;

    const int64_t nr  = ggml_nrows(src0);

    // rows per thread
    const int64_t dr = (nr + nth - 1)/nth;

    // row range for this thread
    const int64_t ir0 = dr*ith;
    const int64_t ir1 = MIN(ir0 + dr, nr);

    return {ir0, ir1};
}

#endif

Line	Count	Source
1		#pragma once
2
3		#include "ggml.h"
4		#include "traits.h"
5		#include "ggml-cpu-impl.h"
6		#include "ggml-impl.h"
7		#include "simd-mappings.h"
8
9		#ifdef __cplusplus
10
11		#include <utility>
12
13		// convenience functions/macros for use in template calls
14		// note: these won't be required after the 'traits' lookup table is used.
15	0	static inline ggml_fp16_t f32_to_f16(float x) {
16	0	return GGML_CPU_FP32_TO_FP16(x);
17	0	} Unexecuted instantiation: binary-ops.cpp:f32_to_f16(float) Unexecuted instantiation: unary-ops.cpp:f32_to_f16(float) Unexecuted instantiation: ops.cpp:f32_to_f16(float)
18
19	0	static inline float f16_to_f32(ggml_fp16_t x) {
20	0	return GGML_CPU_FP16_TO_FP32(x);
21	0	} Unexecuted instantiation: binary-ops.cpp:f16_to_f32(unsigned short) Unexecuted instantiation: unary-ops.cpp:f16_to_f32(unsigned short) Unexecuted instantiation: ops.cpp:f16_to_f32(unsigned short)
22
23	0	static inline ggml_bf16_t f32_to_bf16(float x) {
24	0	return GGML_FP32_TO_BF16(x);
25	0	} Unexecuted instantiation: binary-ops.cpp:f32_to_bf16(float) Unexecuted instantiation: unary-ops.cpp:f32_to_bf16(float) Unexecuted instantiation: ops.cpp:f32_to_bf16(float)
26
27	0	static inline float bf16_to_f32(ggml_bf16_t x) {
28	0	return GGML_BF16_TO_FP32(x);
29	0	} Unexecuted instantiation: binary-ops.cpp:bf16_to_f32(ggml_bf16_t) Unexecuted instantiation: unary-ops.cpp:bf16_to_f32(ggml_bf16_t) Unexecuted instantiation: ops.cpp:bf16_to_f32(ggml_bf16_t)
30
31	0	static inline float i32_to_f32(int32_t x) {
32	0	return x;
33	0	} Unexecuted instantiation: binary-ops.cpp:i32_to_f32(int) Unexecuted instantiation: unary-ops.cpp:i32_to_f32(int) Unexecuted instantiation: ops.cpp:i32_to_f32(int)
34
35	0	static inline int32_t f32_to_i32(float x) {
36	0	return x;
37	0	} Unexecuted instantiation: binary-ops.cpp:f32_to_i32(float) Unexecuted instantiation: unary-ops.cpp:f32_to_i32(float) Unexecuted instantiation: ops.cpp:f32_to_i32(float)
38
39	0	static inline float f32_to_f32(float x) {
40	0	return x;
41	0	} Unexecuted instantiation: binary-ops.cpp:f32_to_f32(float) Unexecuted instantiation: unary-ops.cpp:f32_to_f32(float) Unexecuted instantiation: ops.cpp:f32_to_f32(float)
42
43		// TODO - merge this into the traits table, after using row-based conversions
44		template <class T>
45		struct type_conversion_table;
46
47		template <>
48		struct type_conversion_table<ggml_fp16_t> {
49		static constexpr float (*to_f32)(ggml_fp16_t) = f16_to_f32;
50		static constexpr ggml_fp16_t (*from_f32)(float) = f32_to_f16;
51		};
52
53		template <>
54		struct type_conversion_table<float> {
55		static constexpr float (*to_f32)(float) = f32_to_f32;
56		static constexpr float (*from_f32)(float) = f32_to_f32;
57		};
58
59		template <>
60		struct type_conversion_table<ggml_bf16_t> {
61		static constexpr float (*to_f32)(ggml_bf16_t) = bf16_to_f32;
62		static constexpr ggml_bf16_t (*from_f32)(float) = f32_to_bf16;
63		};
64
65		template <>
66		struct type_conversion_table<int32_t> {
67		static constexpr float (*to_f32)(int32_t) = i32_to_f32;
68		static constexpr int32_t (*from_f32)(float) = f32_to_i32;
69		};
70
71	0	static std::pair<int64_t, int64_t> get_thread_range(const struct ggml_compute_params * params, const struct ggml_tensor * src0) {
72	0	const int64_t ith = params->ith;
73	0	const int64_t nth = params->nth;
74
75	0	const int64_t nr = ggml_nrows(src0);
76
77		// rows per thread
78	0	const int64_t dr = (nr + nth - 1)/nth;
79
80		// row range for this thread
81	0	const int64_t ir0 = dr*ith;
82	0	const int64_t ir1 = MIN(ir0 + dr, nr);
83
84	0	return {ir0, ir1};
85	0	} Unexecuted instantiation: binary-ops.cpp:get_thread_range(ggml_compute_params const, ggml_tensor const) Unexecuted instantiation: unary-ops.cpp:get_thread_range(ggml_compute_params const, ggml_tensor const) Unexecuted instantiation: ops.cpp:get_thread_range(ggml_compute_params const, ggml_tensor const)
86
87		#endif

Coverage Report

Created: 2025-11-24 06:10