[nvptx, libgomp] Add GOMP_NVPTX_JIT=-O[0-4] in nvptx libgomp plugin

2018-04-26  Tom de Vries  <tom@codesourcery.com>

	PR libgomp/84020
	* plugin/cuda/cuda.h (CUjit_option): Add CU_JIT_OPTIMIZATION_LEVEL.
	* plugin/plugin-nvptx.c (_GNU_SOURCE): Define.
	(process_GOMP_NVPTX_JIT): New function.
	(link_ptx): Use process_GOMP_NVPTX_JIT.

From-SVN: r259678

diff --git a/libgomp/ChangeLog b/libgomp/ChangeLog
index e498a0813e6b25cf472bc53e1417ac3201e82a24..0a612a359307ea2fc89a753d1168f25225238d03 100644 (file)
--- a/libgomp/ChangeLog
+++ b/libgomp/ChangeLog
@@ -1,3 +1,11 @@
+2018-04-26  Tom de Vries  <tom@codesourcery.com>
+
+       PR libgomp/84020
+       * plugin/cuda/cuda.h (CUjit_option): Add CU_JIT_OPTIMIZATION_LEVEL.
+       * plugin/plugin-nvptx.c (_GNU_SOURCE): Define.
+       (process_GOMP_NVPTX_JIT): New function.
+       (link_ptx): Use process_GOMP_NVPTX_JIT.
+
 2018-04-26  Richard Biener <rguenther@suse.de>
            Tom de Vries  <tom@codesourcery.com>
 

diff --git a/libgomp/plugin/cuda/cuda.h b/libgomp/plugin/cuda/cuda.h
index edad4c60a750a10e2db0d858be4603e67154a086..4799825bda2b773f19e270d2afa9718d2cf46d0c 100644 (file)
--- a/libgomp/plugin/cuda/cuda.h
+++ b/libgomp/plugin/cuda/cuda.h
@@ -88,6 +88,7 @@ typedef enum {
   CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES = 4,
   CU_JIT_ERROR_LOG_BUFFER = 5,
   CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES = 6,
+  CU_JIT_OPTIMIZATION_LEVEL = 7,
   CU_JIT_LOG_VERBOSE = 12
 } CUjit_option;
 

diff --git a/libgomp/plugin/plugin-nvptx.c b/libgomp/plugin/plugin-nvptx.c
index 9ae60953a9a352d8e1b2d05a8f2411fe52d48097..2b875ae2b53472080fcdb13283cc90efa43c4e8b 100644 (file)
--- a/libgomp/plugin/plugin-nvptx.c
+++ b/libgomp/plugin/plugin-nvptx.c
@@ -31,6 +31,7 @@
    is not clear as to what that state might be.  Or how one might
    propagate it from one thread to another.  */
 
+#define _GNU_SOURCE
 #include "openacc.h"
 #include "config.h"
 #include "libgomp-plugin.h"
@@ -138,6 +139,8 @@ init_cuda_lib (void)
 # define init_cuda_lib() true
 #endif
 
+#include "secure_getenv.h"
+
 /* Convenience macros for the frequently used CUDA library call and
    error handling sequence as well as CUDA library calls that
    do the error checking themselves or don't do it at all.  */
@@ -876,12 +879,42 @@ notify_var (const char *var_name, const char *env_var)
     GOMP_PLUGIN_debug (0, "%s: '%s'\n", var_name, env_var);
 }
 
+static void
+process_GOMP_NVPTX_JIT (intptr_t *gomp_nvptx_o)
+{
+  const char *var_name = "GOMP_NVPTX_JIT";
+  const char *env_var = secure_getenv (var_name);
+  notify_var (var_name, env_var);
+
+  if (env_var == NULL)
+    return;
+
+  const char *c = env_var;
+  while (*c != '\0')
+    {
+      while (*c == ' ')
+       c++;
+
+      if (c[0] == '-' && c[1] == 'O'
+         && '0' <= c[2] && c[2] <= '4'
+         && (c[3] == '\0' || c[3] == ' '))
+       {
+         *gomp_nvptx_o = c[2] - '0';
+         c += 3;
+         continue;
+       }
+
+      GOMP_PLUGIN_error ("Error parsing %s", var_name);
+      break;
+    }
+}
+
 static bool
 link_ptx (CUmodule *module, const struct targ_ptx_obj *ptx_objs,
          unsigned num_objs)
 {
-  CUjit_option opts[6];
-  void *optvals[6];
+  CUjit_option opts[7];
+  void *optvals[7];
   float elapsed = 0.0;
   char elog[1024];
   char ilog[16384];
@@ -908,7 +941,24 @@ link_ptx (CUmodule *module, const struct targ_ptx_obj *ptx_objs,
   opts[5] = CU_JIT_LOG_VERBOSE;
   optvals[5] = (void *) 1;
 
-  CUDA_CALL (cuLinkCreate, 6, opts, optvals, &linkstate);
+  static intptr_t gomp_nvptx_o = -1;
+
+  static bool init_done = false;
+  if (!init_done)
+    {
+      process_GOMP_NVPTX_JIT (&gomp_nvptx_o);
+      init_done = true;
+  }
+
+  int nopts = 6;
+  if (gomp_nvptx_o != -1)
+    {
+      opts[nopts] = CU_JIT_OPTIMIZATION_LEVEL;
+      optvals[nopts] = (void *) gomp_nvptx_o;
+      nopts++;
+    }
+
+  CUDA_CALL (cuLinkCreate, nopts, opts, optvals, &linkstate);
 
   for (; num_objs--; ptx_objs++)
     {