From 7f2a3c0ce1f8706410ae81639c45f9cfeff5d184 Mon Sep 17 00:00:00 2001
From: David Anderson <davea@ssl.berkeley.edu>
Date: Tue, 13 Sep 2011 22:58:39 +0000
Subject: [PATCH] - client: get GPU available RAM at startup (only) - client:
 fix compile warning

svn path=/trunk/boinc/; revision=24188
---
 checkin_notes            | 12 ++++++++++++
 client/client_state.cpp  |  2 ++
 client/coproc_detect.cpp | 41 +++++++---------------------------------
 client/cpu_sched.cpp     |  8 --------
 client/work_fetch.cpp    | 14 +++++++-------
 client/work_fetch.h      |  2 +-
 lib/coproc.cpp           |  1 -
 7 files changed, 29 insertions(+), 51 deletions(-)

diff --git a/checkin_notes b/checkin_notes
index ff0ef9936f..0028141406 100644
--- a/checkin_notes
+++ b/checkin_notes
@@ -5957,3 +5957,15 @@ Rom    13 Sept 2011
         
     samples/vboxwrapper/
         vbox.cpp
+
+David  13 Sept 2011
+    - client: get GPU available RAM at startup (only)
+    - client: fix compile warning
+
+    lib/
+        coproc.cpp
+    client/
+        work_fetch.cpp,h
+        client_state.cpp
+        coproc_detect.cpp
+        cpu_sched.cpp
diff --git a/client/client_state.cpp b/client/client_state.cpp
index 1eebf885fd..7fa101d2b1 100644
--- a/client/client_state.cpp
+++ b/client/client_state.cpp
@@ -386,6 +386,7 @@ int CLIENT_STATE::init() {
             msg_printf(NULL, MSG_INFO, "NVIDIA GPU info taken from cc_config.xml");
         } else {
             coprocs.add(coprocs.nvidia);
+            coprocs.nvidia.print_available_ram();
         }
     }
     if (coprocs.have_ati()) {
@@ -393,6 +394,7 @@ int CLIENT_STATE::init() {
             msg_printf(NULL, MSG_INFO, "ATI GPU info taken from cc_config.xml");
         } else {
             coprocs.add(coprocs.ati);
+            coprocs.ati.print_available_ram();
         }
     }
     host_info._coprocs = coprocs;
diff --git a/client/coproc_detect.cpp b/client/coproc_detect.cpp
index 5972890130..054f750130 100644
--- a/client/coproc_detect.cpp
+++ b/client/coproc_detect.cpp
@@ -47,8 +47,6 @@
 using std::string;
 using std::vector;
 
-//#define MEASURE_AVAILABLE_RAM
-
 static bool in_vector(int n, vector<int>& v) {
     for (unsigned int i=0; i<v.size(); i++) {
         if (v[i] == n) return true;
@@ -104,10 +102,6 @@ cl_int (*__clGetDeviceInfo)(cl_device_id    /* device */,
 #endif
 
 void COPROC::print_available_ram() {
-#ifdef MEASURE_AVAILABLE_RAM
-    if (gstate.now - last_print_time < 60) return;
-    last_print_time = gstate.now;
-
     for (int i=0; i<count; i++) {
         if (available_ram_unknown[i]) {
             if (log_flags.coproc_debug) {
@@ -126,7 +120,6 @@ void COPROC::print_available_ram() {
             }
         }
     }
-#endif
 }
 
 
@@ -919,6 +912,8 @@ void COPROC_NVIDIA::get(
         return;
     }
 
+    get_available_ram();
+
     // identify the most capable non-ignored instance
     //
     bool first = true;
@@ -988,7 +983,6 @@ void COPROC_NVIDIA::fake(int driver_version, double ram, int n) {
 // If this fails, set "available_ram_unknown"
 //
 void COPROC_NVIDIA::get_available_ram() {
-#ifdef MEASURE_AVAILABLE_RAM
     int device, i, retval;
     unsigned int memfree, memtotal;
     unsigned int ctx;
@@ -1038,12 +1032,6 @@ void COPROC_NVIDIA::get_available_ram() {
         available_ram[i] = (double) memfree;
         available_ram_unknown[i] = false;
     }
-#else
-    for (int i=0; i<count; i++) {
-        available_ram_unknown[i] = false;
-        available_ram[i] = prop.totalGlobalMem;
-    }
-#endif
 }
 
 // check whether each GPU is running a graphics app (assume yes)
@@ -1417,6 +1405,8 @@ void COPROC_ATI::get(
         gpus.push_back(cc);
     }
 
+    get_available_ram();
+
     // shut down, otherwise Lenovo won't be able to switch to low-power GPU
     //
     retval = (*__calShutdown)();
@@ -1472,8 +1462,10 @@ void COPROC_ATI::fake(double ram, int n) {
 	set_peak_flops();
 }
 
+// get available RAM of ATI GPUs
+// NOTE: last time we checked, repeated calls to this crash the driver
+//
 void COPROC_ATI::get_available_ram() {
-#ifdef MEASURE_AVAILABLE_RAM
     CALdevicestatus st;
     CALdevice dev;
     int i, retval;
@@ -1491,18 +1483,6 @@ void COPROC_ATI::get_available_ram() {
     for (i=0; i<count; i++) {
         available_ram[i] = 0;
         available_ram_unknown[i] = true;
-    }
-    retval = (*__calInit)();
-    if (retval) {
-        if (log_flags.coproc_debug) {
-            msg_printf(0, MSG_INFO,
-                "[coproc] calInit() returned %d", retval
-            );
-        }
-        return;
-    }
-
-    for (i=0; i<count; i++) {
         int devnum = device_nums[i];
         retval = (*__calDeviceOpen)(&dev, devnum);
         if (retval) {
@@ -1528,13 +1508,6 @@ void COPROC_ATI::get_available_ram() {
         available_ram_unknown[i] = false;
         (*__calDeviceClose)(dev);
     }
-    (*__calShutdown)();
-#else
-    for (int i=0; i<count; i++) {
-        available_ram_unknown[i] = false;
-        available_ram[i] = attribs.localRAM*MEGA;
-    }
-#endif
 }
 
 bool COPROC_ATI::matches(OPENCL_DEVICE_PROP& OpenCLprop) {
diff --git a/client/cpu_sched.cpp b/client/cpu_sched.cpp
index 81a4292893..6a934c9c3c 100644
--- a/client/cpu_sched.cpp
+++ b/client/cpu_sched.cpp
@@ -1357,17 +1357,9 @@ static inline void assign_coprocs(vector<RESULT*>& jobs) {
 
     coprocs.clear_usage();
     if (coprocs.have_nvidia()) {
-        coprocs.nvidia.get_available_ram();
-        if (log_flags.coproc_debug) {
-            coprocs.nvidia.print_available_ram();
-        }
         copy_available_ram(coprocs.nvidia, GPU_TYPE_NVIDIA);
     }
     if (coprocs.have_ati()) {
-        coprocs.ati.get_available_ram();
-        if (log_flags.coproc_debug) {
-            coprocs.ati.print_available_ram();
-        }
         copy_available_ram(coprocs.ati, GPU_TYPE_ATI);
     }
 
diff --git a/client/work_fetch.cpp b/client/work_fetch.cpp
index 5e3dce5b1e..16dde7c4a3 100644
--- a/client/work_fetch.cpp
+++ b/client/work_fetch.cpp
@@ -298,16 +298,16 @@ PROJECT* RSC_WORK_FETCH::choose_project(int criterion) {
     switch (criterion) {
     case FETCH_IF_IDLE_INSTANCE:
     case FETCH_IF_MAJOR_SHORTFALL:
-        set_request(pbest, true);
+        set_request(pbest);
         break;
     case FETCH_IF_PROJECT_STARVED:
-        set_request(pbest, false);
+        set_request(pbest);
         break;
     case FETCH_IF_MINOR_SHORTFALL:
         // in this case, potentially request work for all resources
         //
         if (project_priority(pbest) < 0) {
-            set_request(pbest, true);
+            set_request(pbest);
         } else {
             work_fetch.set_all_requests(pbest);
         }
@@ -339,7 +339,7 @@ PROJECT* RSC_WORK_FETCH::choose_project(int criterion) {
 // request this project's share of shortfall and instances.
 // don't request anything if project is overworked or backed off.
 //
-void RSC_WORK_FETCH::set_request(PROJECT* p, bool allow_overworked) {
+void RSC_WORK_FETCH::set_request(PROJECT* p) {
     if (dont_fetch(p, rsc_type)) return;
 
     // if backup project, fetch 1 job per idle instance
@@ -474,13 +474,13 @@ void RSC_WORK_FETCH::supplement(PROJECT* pp) {
         }
     }
     // didn't find a better project; ask for work
-    set_request(pp, true);
+    set_request(pp);
 }
 
 void WORK_FETCH::set_all_requests_hyst(PROJECT* p, int rsc_type) {
     for (int i=0; i<coprocs.n_rsc; i++) {
         if (i == rsc_type) {
-            rsc_work_fetch[i].set_request(p, true);
+            rsc_work_fetch[i].set_request(p);
         } else {
             if (i==0 || gpus_usable) {
                 rsc_work_fetch[i].supplement(p);
@@ -492,7 +492,7 @@ void WORK_FETCH::set_all_requests_hyst(PROJECT* p, int rsc_type) {
 void WORK_FETCH::set_all_requests(PROJECT* p) {
     for (int i=0; i<coprocs.n_rsc; i++) {
         if (i==0 || gpus_usable) {
-            rsc_work_fetch[i].set_request(p, false);
+            rsc_work_fetch[i].set_request(p);
         }
     }
 }
diff --git a/client/work_fetch.h b/client/work_fetch.h
index b875580ea2..84a1479382 100644
--- a/client/work_fetch.h
+++ b/client/work_fetch.h
@@ -189,7 +189,7 @@ struct RSC_WORK_FETCH {
     RSC_PROJECT_WORK_FETCH& project_state(PROJECT*);
     void print_state(const char*);
     void clear_request();
-    void set_request(PROJECT*, bool allow_overworked);
+    void set_request(PROJECT*);
     bool may_have_work(PROJECT*);
     RSC_WORK_FETCH() {
         rsc_type = 0;
diff --git a/lib/coproc.cpp b/lib/coproc.cpp
index 183bc8e4b8..8ceaf54451 100644
--- a/lib/coproc.cpp
+++ b/lib/coproc.cpp
@@ -365,7 +365,6 @@ void COPROC_NVIDIA::write_xml(MIOFILE& f, bool include_request) {
         prop.multiProcessorCount
     );
 
-
     if (have_opencl) {
         opencl_write_xml(f);
     }