From 1597316b97cbf45ac0a313f9292efca4199c7905 Mon Sep 17 00:00:00 2001
From: David Anderson <davea@einstein-mac1.atlas.local>
Date: Mon, 5 Feb 2024 03:10:16 +0100
Subject: [PATCH] client: debug Apple Silicon GPU detection

---
 client/gpu_detect.cpp | 38 +++++++++++++++++++++++++-------------
 client/gpu_opencl.cpp | 11 ++++-------
 lib/coproc.h          |  4 ++--
 3 files changed, 31 insertions(+), 22 deletions(-)

diff --git a/client/gpu_detect.cpp b/client/gpu_detect.cpp
index 56a3077e0c..b6fb79c180 100644
--- a/client/gpu_detect.cpp
+++ b/client/gpu_detect.cpp
@@ -212,7 +212,8 @@ void COPROCS::detect_gpus(vector<string> &warnings) {
     catch (...) {
         gpu_warning(warnings, "Caught SIGSEGV in OpenCL detection");
     }
-#else
+#else // non-Windows
+
     void (*old_sig)(int) = signal(SIGSEGV, segv_handler);
     if (setjmp(resume)) {
         gpu_warning(warnings, "Caught SIGSEGV in NVIDIA GPU detection");
@@ -227,6 +228,8 @@ void COPROCS::detect_gpus(vector<string> &warnings) {
     } else {
         ati.get(warnings);
     }
+#else
+    apple_gpu.get(warnings);
 #endif
     if (setjmp(resume)) {
         gpu_warning(warnings, "Caught SIGSEGV in INTEL GPU detection");
@@ -848,24 +851,33 @@ void COPROC_APPLE::get(vector<string>&) {
     if (retval) return;
     FILE* f = fopen("temp", "r");
     if (!f) return;
-    char buf[256], model[256];
-    int n, metal;
-    bool have_model=false, have_ncores=false, have_metal=false;
+    char buf[256], chipset_model[256];
+    int n, metalv;
+    bool have_model=false, have_ncores=false, have_metalv=false;
     while (fgets(buf, sizeof(buf), f)) {
-        if (sscanf(buf, "%*[ ]Chipset Model: %[^\n]", model) == 1) {
+        if (sscanf(buf, "%*[ ]Chipset Model: %[^\n]", chipset_model) == 1) {
             have_model = true;
-        } else if (sscanf(buf, "%*[ ]Total Number of Cores: %d", n) == 1) {
+        } else if (sscanf(buf, "%*[ ]Total Number of Cores: %d", &n) == 1) {
             have_ncores = true;
-        } else if (sscanf(buf, "%*[ ]Metal Support: %[^\n]", metal) == 1) {
-            have_metal = true;
+        } else if (sscanf(buf, "%*[ ]Metal Support: Metal %d", &metalv) == 1) {
+            have_metalv = true;
         }
     }
     fclose(f);
-    if (have_model && have_ncores && have_metal) {
-        count = 1;
-        safe_strcpy(chipset_model, model);
-        ncores = n;
-        metal_support = metal;
+    if (have_model && have_ncores && have_metalv) {
+        COPROC_APPLE ca;
+        ca.count = 1;
+        safe_strcpy(ca.model, chipset_model);
+        ca.ncores = n;
+        ca.metal_support = metalv;
+        ca.have_metal = true;
+        apple_gpus.push_back(ca);
     }
 }
+
+void COPROC_APPLE::correlate(bool, vector<int> &ignore_devs) {
+    if (!ignore_devs.empty() && ignore_devs[0]==0) return;
+    if (apple_gpus.empty()) return;
+    *this = apple_gpus[0];
+}
 #endif
diff --git a/client/gpu_opencl.cpp b/client/gpu_opencl.cpp
index f774fe110b..010c2e524e 100644
--- a/client/gpu_opencl.cpp
+++ b/client/gpu_opencl.cpp
@@ -717,12 +717,6 @@ void COPROCS::get_opencl(
                 prop.opencl_available_ram = prop.global_mem_size;
 
                 apple_gpu_opencls.push_back(prop);
-
-                // At present Apple GPUs only support OpenCL
-                // and do not have a native GPGPU framework,
-                // so treat each detected GPU as a native device.
-                //
-                apple_gpus.push_back(c);
             } else {
                 //////////// OTHER GPU OR ACCELERATOR //////////////
                 // Put each coprocessor instance into a separate other_opencls element
@@ -857,9 +851,12 @@ void COPROCS::correlate_opencl(
                 use_all, apple_gpu_opencls,
                 ignore_gpu_instance[PROC_TYPE_APPLE_GPU]
             );
-            apple_gpu.available_ram = apple_gpu.opencl_prop.global_mem_size;
             safe_strcpy(apple_gpu.model, apple_gpu.opencl_prop.name);
         }
+        apple_gpu.set_peak_flops();
+        if (!apple_gpu.available_ram) {
+            apple_gpu.available_ram = apple_gpu.opencl_prop.global_mem_size;
+        }
     }
 }
 
diff --git a/lib/coproc.h b/lib/coproc.h
index e0c5538ad7..37622f7581 100644
--- a/lib/coproc.h
+++ b/lib/coproc.h
@@ -382,8 +382,8 @@ struct COPROC_APPLE : public COPROC {
 #endif
     COPROC_APPLE(int): COPROC() {}
     COPROC_APPLE(): COPROC() {clear();}
-    void get(std::vector<std::string>& ) {};
-    void correlate(bool , std::vector<int>& ) {};
+    void get(std::vector<std::string>&);
+    void correlate(bool , std::vector<int>&);
     void clear();
     int parse(XML_PARSER&);
     void set_peak_flops();