hashcat/OpenCL/inc_vendor.h

/**
 * Author......: See docs/credits.txt
 * License.....: MIT
 */

#ifndef _INC_VENDOR_H
#define _INC_VENDOR_H

#ifdef _CPU_OPENCL_EMU_H
#define CONSTANT_AS
#define GLOBAL_AS
#define LOCAL_AS
#define KERNEL_FQ
#else
#define CONSTANT_AS __constant
#define GLOBAL_AS   __global
#define LOCAL_AS    __local
#define KERNEL_FQ   __kernel
#endif

#ifndef MAYBE_UNUSED
#define MAYBE_UNUSED
#endif

/**
 * device type
 */

#define DEVICE_TYPE_CPU   2
#define DEVICE_TYPE_GPU   4
#define DEVICE_TYPE_ACCEL 8

#if   DEVICE_TYPE == DEVICE_TYPE_CPU
#define IS_CPU
#elif DEVICE_TYPE == DEVICE_TYPE_GPU
#define IS_GPU
#elif DEVICE_TYPE == DEVICE_TYPE_ACCEL
#define IS_ACCEL
#endif

/**
 * vendor specific
 */

#if   VENDOR_ID == (1 << 0)
#define IS_AMD
#elif VENDOR_ID == (1 << 1)
#define IS_APPLE
#define IS_GENERIC
#elif VENDOR_ID == (1 << 2)
#define IS_INTEL_BEIGNET
#define IS_GENERIC
#elif VENDOR_ID == (1 << 3)
#define IS_INTEL_SDK
#define IS_GENERIC
#elif VENDOR_ID == (1 << 4)
#define IS_MESA
#define IS_GENERIC
#elif VENDOR_ID == (1 << 5)
#define IS_NV
#elif VENDOR_ID == (1 << 6)
#define IS_POCL
#define IS_GENERIC
#else
#define IS_GENERIC
#endif

#define LOCAL_MEM_TYPE_LOCAL  1
#define LOCAL_MEM_TYPE_GLOBAL 2

#if LOCAL_MEM_TYPE == LOCAL_MEM_TYPE_LOCAL
#define REAL_SHM
#endif

#ifdef REAL_SHM
#define SHM_TYPE LOCAL_AS
#else
#define SHM_TYPE CONSTANT_AS
#endif

/**
 * function declarations can have a large influence depending on the opencl runtime
 */

#if defined IS_CPU
#define DECLSPEC inline
#elif defined IS_GPU
#if defined IS_AMD
#define DECLSPEC inline
#else
#define DECLSPEC
#endif
#else
#define DECLSPEC
#endif

/**
 * AMD specific
 */

#ifdef IS_AMD
#if defined(cl_amd_media_ops)
#pragma OPENCL EXTENSION cl_amd_media_ops  : enable
#endif
#if defined(cl_amd_media_ops2)
#pragma OPENCL EXTENSION cl_amd_media_ops2 : enable
#endif
#endif

/**
 * Unrolling is generally enabled, for all device types and hash modes
 * There's a few exception when it's better not to unroll
 * Some algorithms run into too much register pressure due to loop unrolling
 */

// generic vendors: those algos have shown that they produce better results on both amd and nv when not unrolled
// so we can assume they will produce better results on other vendors as well

#ifdef NO_UNROLL
#undef _unroll
#endif

#endif
Initial commit 2015-12-04 14:47:52 +00:00			`/**`
Added docs/credits.txt Added docs/team.txt 2016-09-11 20:20:15 +00:00			`* Author......: See docs/credits.txt`
Initial commit 2015-12-04 14:47:52 +00:00			`* License.....: MIT`
			`*/`

Rename some functions in inc_common.cl to avoid conflicts with bitops.c 2019-03-23 21:15:38 +00:00			`#ifndef _INC_VENDOR_H`
			`#define _INC_VENDOR_H`

Replace __kernel, __constant, __global and __local qualifiers with macro for better control 2019-03-22 21:27:58 +00:00			`#ifdef _CPU_OPENCL_EMU_H`
			`#define CONSTANT_AS`
			`#define GLOBAL_AS`
			`#define LOCAL_AS`
			`#define KERNEL_FQ`
			`#else`
			`#define CONSTANT_AS __constant`
			`#define GLOBAL_AS __global`
			`#define LOCAL_AS __local`
			`#define KERNEL_FQ __kernel`
			`#endif`

Make it easier to include OpenCL kernels into modules 2019-04-04 18:01:37 +00:00			`#ifndef MAYBE_UNUSED`
			`#define MAYBE_UNUSED`
			`#endif`

Prepare for a more dynamic #pragma unroll use 2016-05-09 19:32:12 +00:00			`/**`
			`* device type`
			`*/`

			`#define DEVICE_TYPE_CPU 2`
			`#define DEVICE_TYPE_GPU 4`
			`#define DEVICE_TYPE_ACCEL 8`

			`#if DEVICE_TYPE == DEVICE_TYPE_CPU`
			`#define IS_CPU`
			`#elif DEVICE_TYPE == DEVICE_TYPE_GPU`
			`#define IS_GPU`
			`#elif DEVICE_TYPE == DEVICE_TYPE_ACCEL`
			`#define IS_ACCEL`
Test fix for broken AMD support after https://github.com/hashcat/oclHashcat/pull/303 2016-04-20 19:19:15 +00:00			`#endif`

Initial commit 2015-12-04 14:47:52 +00:00			`/**`
			`* vendor specific`
			`*/`

Prepare for a more dynamic #pragma unroll use 2016-05-09 19:32:12 +00:00			`#if VENDOR_ID == (1 << 0)`
Initial commit 2015-12-04 14:47:52 +00:00			`#define IS_AMD`
Added SIMD code for WPA/WPA2 2016-05-14 17:45:51 +00:00			`#elif VENDOR_ID == (1 << 1)`
			`#define IS_APPLE`
			`#define IS_GENERIC`
			`#elif VENDOR_ID == (1 << 2)`
			`#define IS_INTEL_BEIGNET`
			`#define IS_GENERIC`
			`#elif VENDOR_ID == (1 << 3)`
			`#define IS_INTEL_SDK`
			`#define IS_GENERIC`
			`#elif VENDOR_ID == (1 << 4)`
			`#define IS_MESA`
			`#define IS_GENERIC`
			`#elif VENDOR_ID == (1 << 5)`
Initial commit 2015-12-04 14:47:52 +00:00			`#define IS_NV`
Added SIMD code for WPA/WPA2 2016-05-14 17:45:51 +00:00			`#elif VENDOR_ID == (1 << 6)`
			`#define IS_POCL`
			`#define IS_GENERIC`
Added support to utilize multiple different OpenCL platforms in parallel, ex: AMD + NV 2016-01-13 16:10:40 +00:00			`#else`
Fix for -m 1500, -m 3000: Missing "defined" keyword Fix for -m 1000: MD4_H1 copy/paste error Fix for -m 8900, -m 9300: Invalid value for device_processor_cores for CPU devices Fix for -m 9100: Variable salt2 initializer needed some clauses for clearness Temporary limit gpu_accel for CPU devices to 1 for development phase Mark pocl as too bleeding edge for production use, recommend native drivers Remove workarounds for pocl Rename VENDOR_ID_UNKNOWN to VENDOR_ID_GENERIC in host code Rename IS_UNKNOWN to IS_GENERIC in kernel code 2016-01-07 19:14:34 +00:00			`#define IS_GENERIC`
Prepare for allow other OpenCL platforms thans AMD and NVidia 2016-01-04 12:17:20 +00:00			`#endif`

OpenCL Device: Do a real query on OpenCL local memory type instead of just assuming it 2018-11-20 09:06:34 +00:00			`#define LOCAL_MEM_TYPE_LOCAL 1`
			`#define LOCAL_MEM_TYPE_GLOBAL 2`

			`#if LOCAL_MEM_TYPE == LOCAL_MEM_TYPE_LOCAL`
Remove SCR_TYPE macro from OpenCL code Disable REAL_SHM access to AMD platform devices 2018-08-13 10:10:03 +00:00			`#define REAL_SHM`
			`#endif`

			`#ifdef REAL_SHM`
Replace __kernel, __constant, __global and __local qualifiers with macro for better control 2019-03-22 21:27:58 +00:00			`#define SHM_TYPE LOCAL_AS`
Remove SCR_TYPE macro from OpenCL code Disable REAL_SHM access to AMD platform devices 2018-08-13 10:10:03 +00:00			`#else`
Replace __kernel, __constant, __global and __local qualifiers with macro for better control 2019-03-22 21:27:58 +00:00			`#define SHM_TYPE CONSTANT_AS`
Remove SCR_TYPE macro from OpenCL code Disable REAL_SHM access to AMD platform devices 2018-08-13 10:10:03 +00:00			`#endif`

OpenCL Kernels: Add general function declaration keyword (static inline) 2018-02-06 18:12:24 +00:00			`/**`
			`* function declarations can have a large influence depending on the opencl runtime`
			`*/`

Some syntax error 2018-02-08 08:49:59 +00:00			`#if defined IS_CPU`
Re-enable CPU optimizations and some CPU case in thread management 2018-02-16 17:56:21 +00:00			`#define DECLSPEC inline`
Some syntax error 2018-02-08 08:49:59 +00:00			`#elif defined IS_GPU`
			`#if defined IS_AMD`
DECLSPEC for CPU 2018-02-08 08:42:59 +00:00			`#define DECLSPEC inline`
			`#else`
NV JiT doesn't like static inline keywords 2018-02-06 21:05:15 +00:00			`#define DECLSPEC`
DECLSPEC for CPU 2018-02-08 08:42:59 +00:00			`#endif`
NV JiT doesn't like static inline keywords 2018-02-06 21:05:15 +00:00			`#else`
DECLSPEC for CPU 2018-02-08 08:42:59 +00:00			`#define DECLSPEC`
NV JiT doesn't like static inline keywords 2018-02-06 21:05:15 +00:00			`#endif`
OpenCL Kernels: Add general function declaration keyword (static inline) 2018-02-06 18:12:24 +00:00
Initial commit 2015-12-04 14:47:52 +00:00			`/**`
			`* AMD specific`
			`*/`

			`#ifdef IS_AMD`
Rename some hashcat specific OpenCL functions to avoid conflicts with existing OpenCL functions from OpenCL runtime 2018-07-22 10:20:20 +00:00			`#if defined(cl_amd_media_ops)`
Prepare for allow other OpenCL platforms thans AMD and NVidia 2016-01-04 12:17:20 +00:00			`#pragma OPENCL EXTENSION cl_amd_media_ops : enable`
Rename some hashcat specific OpenCL functions to avoid conflicts with existing OpenCL functions from OpenCL runtime 2018-07-22 10:20:20 +00:00			`#endif`
			`#if defined(cl_amd_media_ops2)`
Initial commit 2015-12-04 14:47:52 +00:00			`#pragma OPENCL EXTENSION cl_amd_media_ops2 : enable`
			`#endif`
Rename some hashcat specific OpenCL functions to avoid conflicts with existing OpenCL functions from OpenCL runtime 2018-07-22 10:20:20 +00:00			`#endif`
Initial commit 2015-12-04 14:47:52 +00:00
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor 2015-12-15 11:04:22 +00:00			`/**`
Prepare for a more dynamic #pragma unroll use 2016-05-09 19:32:12 +00:00			`* Unrolling is generally enabled, for all device types and hash modes`
			`* There's a few exception when it's better not to unroll`
Add a shared section in inc_vendor.cl 2016-11-13 19:58:28 +00:00			`* Some algorithms run into too much register pressure due to loop unrolling`
- Dropped all vector code since new GPU's are all scalar, makes the code much easier - Some performance on low-end GPU may drop because of that, but only for a few hash-modes - Dropped scalar code (aka warp) since we do not have any vector datatypes anymore - Renamed C++ overloading functions memcat32_9 -> memcat_c32_w4x4_a3x4 - Still need to fix kernels to new function names, needs to be done manually - Temperature Management needs to be rewritten partially because of conflicting datatypes names - Added code to create different codepaths for NV on AMD in runtime in host (see data.vendor_id) - Added code to create different codepaths for NV on AMD in runtime in kernels (see IS_NV and IS_AMD) - First tests working for -m 0, for example - Great performance increases in general for NV so far - Tested amp_* and markov_* kernel - Migrated special NV optimizations for rule processor 2015-12-15 11:04:22 +00:00			`*/`
Initial commit 2015-12-04 14:47:52 +00:00
Add a shared section in inc_vendor.cl 2016-11-13 19:58:28 +00:00			`// generic vendors: those algos have shown that they produce better results on both amd and nv when not unrolled`
			`// so we can assume they will produce better results on other vendors as well`
Prepare for a more dynamic #pragma unroll use 2016-05-09 19:32:12 +00:00
Migrate unroll handling for -m 8000 from OpenCL/inc_vendor.cl to module_jit_build_options() 2019-03-03 10:39:18 +00:00			`#ifdef NO_UNROLL`
			`#undef _unroll`
			`#endif`
Rename some functions in inc_common.cl to avoid conflicts with bitops.c 2019-03-23 21:15:38 +00:00
			`#endif`