macos.md

macOS report

This file is generated by make render from reports/macos/*.json. Do not edit by hand - edits will be overwritten on the next render.

How to read this. These numbers measure function-pointer loading overhead and Vulkan context setup across different API loaders, comparing the stock Vulkan loader against the patched one (see the README for what the patch does). Lower is better.

Do not compare across hosts. Hardware, drivers, OS, and compiler flags all differ between the Linux, MinGW, and macOS reports. The meaningful comparisons are within a single report: loader vs. loader and patched vs. unpatched on the same hardware.

At a glance

Lower is better. Winner = best patched average on this host for each task.

Task	Winner	Patched avg	Unpatched avg	Patch speedup
Load instance functions	Gloam (enabled-list)	0.83µs	1.13µs	1.4×
Load device functions	Gloam (enabled-list)	3.42µs	5.43µs	1.6×
Init + load all functions	Gloam (enabled-list)	5.00µs	7.20µs	1.4×
Full VK context (libvulkan persistent)	Volk	1379.73µs	1657.27µs	1.2×
Full VK context (libvulkan transient)	Gloam (enabled-list)	2247.83µs	1944.57µs	0.9×

Task detail

Load instance functions

Loader	Unpatched	Patched	Patch speedup	vs. fastest
Gloam (enabled-list)	1.13µs	0.83µs	1.4×	1.0×
GLAD (tycho)	4.90µs	2.57µs	1.9×	3.1×
Volk	9.66µs	4.64µs	2.1×	5.6×
Gloam (discover)	181.21µs	19.00µs	9.5×	22.9×
GLAD (dav1dde)	400.87µs	222.95µs	1.8×	269×

Load device functions

Loader	Unpatched	Patched	Patch speedup	vs. fastest
Gloam (enabled-list)	5.43µs	3.42µs	1.6×	1.0×
GLAD (tycho)	12.69µs	7.08µs	1.8×	2.1×
Volk	16.99µs	10.62µs	1.6×	3.1×
Gloam (discover)	17.41µs	11.72µs	1.5×	3.4×
GLAD (dav1dde)	762.92µs	597.76µs	1.3×	175×

Init + load all functions

Loader	Unpatched	Patched	Patch speedup	vs. fastest
Gloam (enabled-list)	7.20µs	5.00µs	1.4×	1.0×
Volk	28.40µs	13.97µs	2.0×	2.8×
Gloam (discover)	422.80µs	86.17µs	4.9×	17.2×
GLAD (tycho)	427.90µs	89.13µs	4.8×	17.8×
GLAD (dav1dde)	1555.03µs	1083.13µs	1.4×	217×

Full VK context (libvulkan persistent)

Loader	Unpatched	Patched	Patch speedup	vs. fastest
Volk	1657.27µs	1379.73µs	1.2×	1.0×
Gloam (enabled-list)	1651.93µs	1403.43µs	1.2×	1.0×
Gloam (discover)	1893.13µs	1436.73µs	1.3×	1.0×
GLAD (tycho)	1883.83µs	1487.80µs	1.3×	1.1×
GLAD (dav1dde)	3236.33µs	2522.37µs	1.3×	1.8×

Full VK context (libvulkan transient)

Loader	Unpatched	Patched	Patch speedup	vs. fastest
Gloam (enabled-list)	1944.57µs	2247.83µs	0.9×	1.0×
Volk	1968.47µs	2257.23µs	0.9×	1.0×
GLAD (tycho)	2159.80µs	2282.53µs	0.9×	1.0×
Gloam (discover)	2176.73µs	2283.10µs	1.0×	1.0×
GLAD (dav1dde)	3556.23µs	3352.43µs	1.1×	1.5×

Binary sizes

All sizes in bytes. Sorted by stripped binary size. Section values come from size; Mach-O binaries report BSS as zero because the Mach-O segment model folds zero-init into __DATA.

Loader	Loader .o	Binary	text	data	bss
GLAD (tycho)	74,024	70,608	49,152	16,384	0
Gloam (discover)	54,320	87,568	65,536	16,384	0
Gloam (enabled-list)	54,320	87,568	65,536	16,384	0
Volk	265,936	136,312	114,688	16,384	0
GLAD (dav1dde)	275,592	161,856	114,688	16,384	0

Test host details

Host

Field	Value
OS	Darwin 25.6.0
Architecture	arm64
CPU	Apple M5

Toolchain

Field	Value
CC	clang (21.0.0)
CXX	clang++ (21.0.0)
OPTFLAGS	-O2 -fno-unroll-loops -mcpu=native -g0
CFLAGS	-std=c17
CXXFLAGS	-std=c++20

Project versions

Project	Version
GLAD (dav1dde)	2.0.8-10-ge3c14f9
GLAD (tycho)	2.0.8-92-g1002eae
gloam	0.7.0
Volk	1.4.357.0-7-g2e19a77
xxHash	0.7.4-1023-gc0b5ea9
Vulkan-Headers	1.4.360

vulkaninfo

Devices:
========
GPU0:
	apiVersion         = 1.4.357
	driverVersion      = 0.2.2210
	vendorID           = 0x106b
	deviceID           = 0x1a06020a
	deviceType         = PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
	deviceName         = Apple M5
	driverID           = DRIVER_ID_MOLTENVK
	driverName         = MoltenVK
	driverInfo         = 1.4.2
	conformanceVersion = 1.4.4.0
	deviceUUID         = 0000106b-1a06-020a-0000-000000000000
	driverUUID         = 4d564b00-0000-28a2-1a06-020a00000000