From ff96015f53cf0dab30a97756fe45b2213efc6378 Mon Sep 17 00:00:00 2001
From: Jens Steube <jens.steube@gmail.com>
Date: Wed, 14 Apr 2021 15:22:30 +0200
Subject: [PATCH] Add OPTS_TYPE_NATIVE_THREADS for use by plugin developer to
 enforce native thread count (useful for scrypt)

---
 include/types.h |  1 +
 src/backend.c   | 56 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 57 insertions(+)

diff --git a/include/types.h b/include/types.h
index 3dfa6f162..e6dd881a7 100644
--- a/include/types.h
+++ b/include/types.h
@@ -435,6 +435,7 @@ typedef enum opts_type
   OPTS_TYPE_DYNAMIC_SHARED    = (1ULL << 48), // use dynamic shared memory (note: needs special kernel changes)
   OPTS_TYPE_SELF_TEST_DISABLE = (1ULL << 49), // some algos use JiT in combinations with a salt or create too much startup time
   OPTS_TYPE_MP_MULTI_DISABLE  = (1ULL << 50), // do not multiply the kernel-accel with the multiprocessor count per device to allow more fine-tuned workload settings
+  OPTS_TYPE_NATIVE_THREADS    = (1ULL << 51), // forces "native" thread count: CPU=1, GPU-Intel=8, GPU-AMD=64 (wavefront), GPU-NV=32 (warps)
 
 } opts_type_t;
 
diff --git a/src/backend.c b/src/backend.c
index 0872e20d3..ef544858c 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -8166,6 +8166,62 @@ int backend_session_begin (hashcat_ctx_t *hashcat_ctx)
 
     const u32 device_processors = device_param->device_processors;
 
+    if (hashconfig->opts_type & OPTS_TYPE_MP_MULTI_DISABLE)
+    {
+      u32 native_accel = device_processors;
+
+      if ((native_accel >= device_param->kernel_accel_min) && (native_accel <= device_param->kernel_accel_max))
+      {
+        device_param->kernel_accel_min = native_accel;
+        device_param->kernel_accel_max = native_accel;
+      }
+    }
+
+    /**
+     * device threads
+     */
+
+    if (hashconfig->opts_type & OPTS_TYPE_NATIVE_THREADS)
+    {
+      u32 native_threads = 0;
+
+      if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)
+      {
+        native_threads = 1;
+      }
+      else if (device_param->opencl_device_type & CL_DEVICE_TYPE_GPU)
+      {
+        // for GPU we need to distinguish by vendor
+
+        if (device_param->opencl_device_vendor_id == VENDOR_ID_INTEL_SDK)
+        {
+          native_threads = 8;
+        }
+        else if (device_param->opencl_device_vendor_id == VENDOR_ID_AMD)
+        {
+          native_threads = 64;
+        }
+        else
+        {
+          native_threads = 32;
+        }
+      }
+      else
+      {
+        // abort?
+      }
+
+      if ((native_threads >= device_param->kernel_threads_min) && (native_threads <= device_param->kernel_threads_max))
+      {
+        device_param->kernel_threads_min = native_threads;
+        device_param->kernel_threads_max = native_threads;
+      }
+      else
+      {
+        // abort?
+      }
+    }
+
     /**
      * create context for each device
      */