vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Adds all the compiler flags to optimize Llama to be more performant. Also makes METAL non conditional

Browse source
This commit is contained in:
Marc Terns 2023-09-12 16:18:31 -05:00
parent 89e89599fd
commit 5faa33de27
1 changed files with 69 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

95
Package.swift
View file

@ -2,30 +2,13 @@
import PackageDescription import PackageDescription
#if arch(arm) || arch(arm64) let package = Package(
let platforms: [SupportedPlatform]? = [ name: "llama",
.macOS(.v11), platforms: [.macOS(.v11),
.iOS(.v14), .iOS(.v14),
.watchOS(.v4), .watchOS(.v4),
.tvOS(.v14) .tvOS(.v14)
] ],
let exclude: [String] = []
let additionalSources: [String] = ["ggml-metal.m"]
let additionalSettings: [CSetting] = [
.unsafeFlags(["-fno-objc-arc"]),
.define("GGML_SWIFT"),
.define("GGML_USE_METAL")
]
#else
let platforms: [SupportedPlatform]? = nil
let exclude: [String] = ["ggml-metal.metal"]
let additionalSources: [String] = []
let additionalSettings: [CSetting] = []
#endif
let package = Package(
name: "llama",
platforms: platforms,
products: [ products: [
.library(name: "llama", targets: ["llama"]), .library(name: "llama", targets: ["llama"]),
], ],
@ -33,21 +16,79 @@ let package = Package(
.target( .target(
name: "llama", name: "llama",
path: ".", path: ".",
exclude: exclude,
sources: [ sources: [
"ggml.c", "ggml.c",
"llama.cpp", "llama.cpp",
"ggml-alloc.c", "ggml-alloc.c",
"k_quants.c", "k_quants.c",
] + additionalSources, "ggml-metal.m"
],
publicHeadersPath: "spm-headers", publicHeadersPath: "spm-headers",
cSettings: [ cSettings: [
.unsafeFlags(["-Wno-shorten-64-to-32"]), .unsafeFlags(["-Wno-shorten-64-to-32",
"-Wall",
"-Wextra",
"-Wpedantic",
"-Wshadow",
"-Wcast-qual",
"-Wstrict-prototypes",
"-Wpointer-arith",
"-Wdouble-promotion",
"-Wno-unused-function",
"-Wmissing-prototypes",
"-Werror=vla",
"-mavx",
"-mavx2",
"-mfma",
"-mf16c",
"-O3",
"-DNDEBUG",
"-Wno-format",
"-mf16c"]),
.define("GGML_USE_K_QUANTS"), .define("GGML_USE_K_QUANTS"),
.define("GGML_USE_ACCELERATE") .define("GGML_USE_ACCELERATE"),
] + additionalSettings, .define("NDEBUG"),
.define("_XOPEN_SOURCE", to: "600"),
.define("_DARWIN_C_SOURCE"),
.unsafeFlags(["-fno-objc-arc"]),
.define("GGML_SWIFT"),
.define("GGML_USE_METAL")
],
cxxSettings: [
.unsafeFlags(["-Wno-shorten-64-to-32",
"-Wall",
"-Wextra",
"-Wpedantic",
"-Wshadow",
"-Wcast-qual",
"-Wstrict-prototypes",
"-Wpointer-arith",
"-Wdouble-promotion",
"-Wno-unused-function",
"-Wmissing-prototypes",
"-Werror=vla",
"-mavx",
"-mavx2",
"-mfma",
"-mf16c",
"-O3",
"-DNDEBUG",
"-Wno-format",
"-mf16c"]),
.define("GGML_USE_K_QUANTS"),
.define("GGML_USE_ACCELERATE"),
.unsafeFlags(["-fno-objc-arc"]),
.define("GGML_SWIFT"),
.define("GGML_USE_METAL"),
.define("NDEBUG"),
.define("_XOPEN_SOURCE", to: "600"),
.define("_DARWIN_C_SOURCE")
],
linkerSettings: [ linkerSettings: [
.linkedFramework("Accelerate") .linkedFramework("Accelerate"),
.linkedFramework("Foundation"),
.linkedFramework("Metal"),
.linkedFramework("MetalKit")
] ]
) )
], ],