summaryrefslogtreecommitdiff
path: root/src/cuda-sim
diff options
context:
space:
mode:
Diffstat (limited to 'src/cuda-sim')
-rw-r--r--src/cuda-sim/cuda-math.h4
-rw-r--r--src/cuda-sim/cuda-sim.cc6
-rw-r--r--src/cuda-sim/instructions.cc9
-rw-r--r--src/cuda-sim/ptx.y3
-rw-r--r--src/cuda-sim/ptx_ir.cc1
-rw-r--r--src/cuda-sim/ptx_ir.h7
-rw-r--r--src/cuda-sim/ptx_parser.cc4
-rw-r--r--src/cuda-sim/ptx_parser.h1
8 files changed, 26 insertions, 9 deletions
diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index 4721e8a..f88c526 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -67,6 +67,8 @@
#ifndef CUDA_MATH
#define CUDA_MATH
+#include <cmath>
+
// cuda math implementations
#undef max
#undef min
@@ -321,7 +323,7 @@ float __internal_accurate_fdividef(float a, float b)
float __saturatef(float a)
{
float b;
- if (isnan(a)) b = 0.0f;
+ if (std::isnan(a)) b = 0.0f;
else if (a >= 1.0f) b = 1.0f;
else if (a <= 0.0f) b = 0.0f;
else b = a;
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 2f166aa..9246613 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1159,13 +1159,13 @@ void function_info::finalize( memory_space *param_mem )
// copy the parameter over word-by-word so that parameter that crosses a memory page can be copied over
//Jin: copy parameter using aligned rules
const size_t word_size = 4;
- param_address = (param_address + size - 1) / size * size; //aligned with size
+ //param_address = (param_address + size - 1) / size * size; //aligned with size TODO: align not correct
for (size_t idx = 0; idx < size; idx += word_size) {
const char *pdata = reinterpret_cast<const char*>(param_value.pdata) + idx; // cast to char * for ptr arithmetic
param_mem->write(param_address + idx, word_size, pdata,NULL,NULL);
}
unsigned offset = p.get_offset();
- assert(offset == param_address);
+ //assert(offset == param_address);
param->set_address(param_address);
param_address += size;
}
@@ -1432,7 +1432,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
if ( (g_ptx_sim_num_insn % 100000) == 0 ) {
dim3 ctaid = get_ctaid();
dim3 tid = get_tid();
- printf("GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
+ DPRINTF(LIVENESS, "GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
g_ptx_sim_num_insn, ctaid.x,ctaid.y,ctaid.z,tid.x,tid.y,tid.z );
fflush(stdout);
}
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 011c285..e3b8970 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -33,6 +33,7 @@
#include "ptx.tab.h"
#include <stdlib.h>
#include <math.h>
+#include <cmath>
#include <fenv.h>
#include "cuda-math.h"
#include "../abstract_hardware_model.h"
@@ -1961,7 +1962,7 @@ ptx_reg_t d2d( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign,
y.f64 = x.f64;
break;
}
- if (isnan(y.f64)) {
+ if (std::isnan(y.f64)) {
y.u64 = 0xfff8000000000000ull;
} else if (saturation_mode) {
y.f64 = cuda_math::__saturatef(y.f64);
@@ -2086,7 +2087,7 @@ void ptx_round(ptx_reg_t& data, int rounding_mode, int type)
}
}
if ((type == F64_TYPE)||(type == FF64_TYPE)) {
- if (isnan(data.f64)) {
+ if (std::isnan(data.f64)) {
data.u64 = 0xfff8000000000000ull;
}
}
@@ -2648,12 +2649,12 @@ void mad_def( const ptx_instruction *pI, ptx_thread_info *thread, bool use_carry
bool isNaN(float x)
{
- return isnan(x);
+ return std::isnan(x);
}
bool isNaN(double x)
{
- return isnan(x);
+ return std::isnan(x);
}
void max_impl( const ptx_instruction *pI, ptx_thread_info *thread )
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index e00aa4b..c0c58a6 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -228,7 +228,8 @@ function_defn: function_decl { set_symtab($1); func_header(".skip"); } statement
block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {func_header_info_int(".maxntid", $2);
func_header_info_int(",", $4);
- func_header_info_int(",", $6); }
+ func_header_info_int(",", $6);
+ maxnt_id($2, $4, $6);}
| MINNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); }
| MAXNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); }
;
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 8ebdcf8..ee36957 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -222,6 +222,7 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
} else {
*func_info = new function_info(entry_point);
(*func_info)->set_name(name);
+ (*func_info)->set_maxnt_id(0);
m_function_info_lookup[key] = *func_info;
}
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 9ad1571..36ef3d5 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -1245,6 +1245,7 @@ public:
const struct gpgpu_ptx_sim_info* get_kernel_info () const
{
+ assert (m_kernel_info.maxthreads == maxnt_id);
return &m_kernel_info;
}
@@ -1252,6 +1253,8 @@ public:
m_kernel_info = info;
m_kernel_info.ptx_version = 10*get_ptx_version().ver();
m_kernel_info.sm_target = get_ptx_version().target();
+ // THIS DEPENDS ON ptxas being called after the PTX is parsed.
+ m_kernel_info.maxthreads = maxnt_id;
}
symbol_table *get_symtab()
{
@@ -1275,7 +1278,11 @@ public:
}
bool is_entry_point() const { return m_entry_point; }
+ void set_maxnt_id(unsigned maxthreads) { maxnt_id = maxthreads;}
+ unsigned get_maxnt_id() { return maxnt_id;}
+
private:
+ unsigned maxnt_id;
unsigned m_uid;
unsigned m_local_mem_framesize;
bool m_entry_point;
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index baa3bcd..e5731a8 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -969,6 +969,10 @@ void target_header3(char* a, char* b, char* c)
g_global_symbol_table->set_sm_target(a,b,c);
}
+void maxnt_id(int x, int y, int z) {
+ g_func_info->set_maxnt_id(x * y * z);
+}
+
void func_header(const char* a) {} //intentional dummy function
void func_header_info(const char* a) {} //intentional dummy function
void func_header_info_int(const char* a, int b) {} //intentional dummy function
diff --git a/src/cuda-sim/ptx_parser.h b/src/cuda-sim/ptx_parser.h
index 32f3903..13042e1 100644
--- a/src/cuda-sim/ptx_parser.h
+++ b/src/cuda-sim/ptx_parser.h
@@ -93,6 +93,7 @@ void change_double_operand_type( int addr_type );
void change_operand_neg( );
void set_immediate_operand_type( );
void version_header(double a);
+void maxnt_id(int x, int y, int z);
//Jin: handle instructino group for cdp
void start_inst_group();
generated by cgit v1.3 (git 2.53.0) at 2026-07-11 01:27:02 +0000