From 73ea99a28b77c2601b19dd5574a2cb267859cbd7 Mon Sep 17 00:00:00 2001
From: Wilson Fung <wlfung@gmail.com>
Date: Mon, 3 Sep 2012 01:36:47 -0800
Subject: Fixing multi-element texel access.  Now texture with RGBA components
 are read properly.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 13949]
---
 CHANGES                      |  2 ++
 src/cuda-sim/instructions.cc | 19 ++++++++++++-------
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/CHANGES b/CHANGES
index 1a05fbc..52017bf 100644
--- a/CHANGES
+++ b/CHANGES
@@ -26,6 +26,8 @@ Version 3.1.1+edits (development branch) versus 3.1.1
     - Fixed incorrect counting of predicated instructions.  Now instructions
       are counted into gpu_sim_insn regardless of predication outcome.  (Bug
       #15 external)
+    - Added support for cudaReadModeNormalizedFloat (a texture read mode). 
+    - Fixed mult-element texel access (e.g. texel with RGBA components).  
 
 Version 3.1.1 versus 3.1.0
 - Add checks to top level makefile to ensure setup_environment is run and checks to
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 65d3af8..e91ddd4 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -3831,27 +3831,32 @@ void tex_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    case B32_TYPE: 
    case S8_TYPE:
    case S16_TYPE:
-   case S32_TYPE:
+   case S32_TYPE: {
+      unsigned long long elementOffset = 0; // offset into the next element 
       mem->read( tex_array_index, cuArray->desc.x/8, &data1.u32);
+      elementOffset += cuArray->desc.x/8;  
       if (cuArray->desc.y) {
-         mem->read( tex_array_index+4, cuArray->desc.y/8, &data2.u32);
+         mem->read( tex_array_index + elementOffset, cuArray->desc.y/8, &data2.u32);
+         elementOffset += cuArray->desc.y/8; 
          if (cuArray->desc.z) {
-            mem->read( tex_array_index+8, cuArray->desc.z/8, &data3.u32);
+            mem->read( tex_array_index + elementOffset, cuArray->desc.z/8, &data3.u32);
+            elementOffset += cuArray->desc.z/8; 
             if (cuArray->desc.w) 
-               mem->read( tex_array_index+12, cuArray->desc.w/8, &data4.u32);
+               mem->read( tex_array_index + elementOffset, cuArray->desc.w/8, &data4.u32);
          }
       }
       break;
+   }
    case B64_TYPE:
    case U64_TYPE:
    case S64_TYPE:
       mem->read( tex_array_index, 8, &data1.u64);
       if (cuArray->desc.y) {
-         mem->read( tex_array_index+4, 8, &data2.u64);
+         mem->read( tex_array_index+8, 8, &data2.u64);
          if (cuArray->desc.z) {
-            mem->read( tex_array_index+8, 8, &data3.u64);
+            mem->read( tex_array_index+16, 8, &data3.u64);
             if (cuArray->desc.w) 
-               mem->read( tex_array_index+12, 8, &data4.u64);
+               mem->read( tex_array_index+24, 8, &data4.u64);
          }
       }
       break;
-- 
cgit v1.3