From 9f9353542801538b36e30ec3471a2e48716110ea Mon Sep 17 00:00:00 2001
From: David Anderson <davea@ssl.berkeley.edu>
Date: Wed, 14 Oct 2009 19:11:11 +0000
Subject: [PATCH]     - client: bug fixes to the above.         Don't fetch
 work for an unable resource.

svn path=/trunk/boinc/; revision=19302
---
 checkin_notes           |  9 +++++++++
 client/cpu_sched.cpp    | 18 +++++++++++-------
 client/cs_scheduler.cpp |  3 +++
 client/work_fetch.cpp   | 23 +++++++++++++----------
 4 files changed, 36 insertions(+), 17 deletions(-)

diff --git a/checkin_notes b/checkin_notes
index 13dd8f875a..b860aec8a5 100644
--- a/checkin_notes
+++ b/checkin_notes
@@ -8656,3 +8656,12 @@ David  14 Oct 2009
         coproc.cpp,h
     sched/
         sched_locality.cpp
+
+David  14 Oct 2009
+    - client: bug fixes to the above.
+        Don't fetch work for an unable resource.
+
+    client/
+        cpu_sched.cpp
+        cs_scheduler.cpp
+        work_fetch.cpp
diff --git a/client/cpu_sched.cpp b/client/cpu_sched.cpp
index a15ba17532..a74102a244 100644
--- a/client/cpu_sched.cpp
+++ b/client/cpu_sched.cpp
@@ -191,35 +191,39 @@ struct PROC_RESOURCES {
 // see whether there's been a change in coproc usability;
 // if so set or clear "coproc_missing" flags and return true.
 //
+#include "filesys.h"
 bool check_coproc_usable(COPROC* cp) {
-    int i;
+    unsigned int i;
+    bool is_cuda = (cp==coproc_cuda);
+    bool new_usable = cp->is_usable();
+    //bool new_usable = !boinc_file_exists("unusable");
     if (cp->usable) {
-        if (!cp->is_usable()) {
+        if (!new_usable) {
             cp->usable = false;
             for (i=0; i<gstate.results.size(); i++) {
                 RESULT* rp = gstate.results[i];
-                if (rp->avp->ncudas) {
+                if (is_cuda?rp->avp->ncudas:rp->avp->natis) {
                     rp->coproc_missing = true;
                 }
             }
             msg_printf(NULL, MSG_INFO,
                 "%s GPU has become unusable; disabling tasks",
-                cp==coproc_cuda?"NVIDIA":"ATI"
+                is_cuda?"NVIDIA":"ATI"
             );
             return true;
         }
     } else {
-        if (cp->is_usable()) {
+        if (new_usable) {
             cp->usable = true;
             for (i=0; i<gstate.results.size(); i++) {
                 RESULT* rp = gstate.results[i];
-                if (rp->avp->ncudas) {
+                if (is_cuda?rp->avp->ncudas:rp->avp->natis) {
                     rp->coproc_missing = false;
                 }
             }
             msg_printf(NULL, MSG_INFO,
                 "%s GPU has become usable; enabling tasks",
-                cp==coproc_cuda?"NVIDIA":"ATI"
+                is_cuda?"NVIDIA":"ATI"
             );
             return true;
         }
diff --git a/client/cs_scheduler.cpp b/client/cs_scheduler.cpp
index 9d85669c5c..01ac42e560 100644
--- a/client/cs_scheduler.cpp
+++ b/client/cs_scheduler.cpp
@@ -833,8 +833,11 @@ int CLIENT_STATE::handle_scheduler_reply(PROJECT* project, char* scheduler_url)
             rp->set_state(RESULT_NEW, "handle_scheduler_reply");
             if (rp->avp->ncudas) {
                 est_cuda_duration += rp->estimated_duration(false);
+                coproc_cuda->usable = true;
+                    // trigger a check of whether GPU is actually usable
             } else if (rp->avp->natis) {
                 est_ati_duration += rp->estimated_duration(false);
+                coproc_ati->usable = true;
             } else {
                 est_cpu_duration += rp->estimated_duration(false);
             }
diff --git a/client/work_fetch.cpp b/client/work_fetch.cpp
index 62250f37cf..22dc1e02e3 100644
--- a/client/work_fetch.cpp
+++ b/client/work_fetch.cpp
@@ -284,10 +284,10 @@ PROJECT* RSC_WORK_FETCH::choose_project(int criterion) {
 
 void WORK_FETCH::set_shortfall_requests(PROJECT* p) {
     cpu_work_fetch.set_shortfall_request(p);
-    if (coproc_cuda) {
+    if (coproc_cuda && coproc_cuda->usable) {
         cuda_work_fetch.set_shortfall_request(p);
     }
-    if (coproc_ati) {
+    if (coproc_ati && coproc_ati->usable) {
         ati_work_fetch.set_shortfall_request(p);
     }
 }
@@ -480,28 +480,31 @@ PROJECT* WORK_FETCH::choose_project() {
     gstate.rr_simulation();
     set_overall_debts();
 
-    if (coproc_cuda) {
+    bool cuda_usable = coproc_cuda && coproc_cuda->usable;
+    bool ati_usable = coproc_ati && coproc_ati->usable;
+
+    if (cuda_usable) {
         p = cuda_work_fetch.choose_project(FETCH_IF_IDLE_INSTANCE);
     }
-    if (coproc_ati) {
+    if (ati_usable) {
         p = ati_work_fetch.choose_project(FETCH_IF_IDLE_INSTANCE);
     }
     if (!p) {
         p = cpu_work_fetch.choose_project(FETCH_IF_IDLE_INSTANCE);
     }
-    if (!p && coproc_cuda) {
+    if (!p && cuda_usable) {
         p = cuda_work_fetch.choose_project(FETCH_IF_MAJOR_SHORTFALL);
     }
-    if (!p && coproc_ati) {
+    if (!p && ati_usable) {
         p = ati_work_fetch.choose_project(FETCH_IF_MAJOR_SHORTFALL);
     }
     if (!p) {
         p = cpu_work_fetch.choose_project(FETCH_IF_MAJOR_SHORTFALL);
     }
-    if (!p && coproc_cuda) {
+    if (!p && cuda_usable) {
         p = cuda_work_fetch.choose_project(FETCH_IF_MINOR_SHORTFALL);
     }
-    if (!p && coproc_ati) {
+    if (!p && ati_usable) {
         p = ati_work_fetch.choose_project(FETCH_IF_MINOR_SHORTFALL);
     }
     if (!p) {
@@ -511,10 +514,10 @@ PROJECT* WORK_FETCH::choose_project() {
     // don't try to maintain GPU work for all projects,
     // since we don't use round-robin scheduling for GPUs
     //
-    if (!p && coproc_cuda) {
+    if (!p && cuda_usable) {
         p = cuda_work_fetch.choose_project(FETCH_IF_PROJECT_STARVED);
     }
-    if (!p && coproc_ati) {
+    if (!p && ati_usable) {
         p = ati_work_fetch.choose_project(FETCH_IF_PROJECT_STARVED);
     }
 #endif