From da52cffcfca12375a1a93506d34d4c4ad5f1d3d2 Mon Sep 17 00:00:00 2001
From: Fabian Bieler <der.fabe@gmx.net>
Date: Tue, 22 Mar 2011 12:27:30 +0100
Subject: [PATCH] mesa: Extend register live intervals to the beginning of loops if they are referenced outside of the loop

---
 src/mesa/program/prog_optimize.c |   49 ++++++++++++++++++++++++++++++++-----
 1 files changed, 42 insertions(+), 7 deletions(-)

diff --git a/src/mesa/program/prog_optimize.c b/src/mesa/program/prog_optimize.c
index 96971f2..b819b00 100644
--- a/src/mesa/program/prog_optimize.c
+++ b/src/mesa/program/prog_optimize.c
@@ -932,9 +932,10 @@ struct loop_info
  * instruction 'ic'.
  */
 static void
-update_interval(GLint intBegin[], GLint intEnd[],
+update_interval(GLint intBegin[], GLint intEnd[], GLint intLoopStart[],
 		struct loop_info *loopStack, GLuint loopStackDepth,
-		GLuint index, GLuint ic)
+		GLuint index, GLuint ic, GLuint regsInLoop[],
+		GLuint *numRegsInLoop)
 {
    int i;
 
@@ -952,9 +953,22 @@ update_interval(GLint intBegin[], GLint intEnd[],
    if (intBegin[index] == -1) {
       ASSERT(intEnd[index] == -1);
       intBegin[index] = intEnd[index] = ic;
+      ASSERT(intLoopStart[index] == -1);
+      if (loopStackDepth) {
+         regsInLoop[*numRegsInLoop] = index;
+         (*numRegsInLoop)++;
+      }
    }
    else {
       intEnd[index] = ic;
+      /* If a register is assigned to in a loop and referenced outside of the
+       * loop, add the entire loop to the live interval so that instructions
+       * before the assignement don't overwrite it on subsequent iterations.
+       */
+      if (intLoopStart[index] != -1) {
+         ASSERT(intLoopStart[index] <= intBegin[index]);
+         intBegin[index] = intLoopStart[index];
+      }
    }
 }
 
@@ -971,9 +985,16 @@ _mesa_find_temp_intervals(const struct prog_instruction *instructions,
    struct loop_info loopStack[MAX_LOOP_NESTING];
    GLuint loopStackDepth = 0;
    GLuint i;
+   /* If a register was first used in a (nested) loop this variable holds the
+    * start of the loop that contains said usage and was closed most recently
+    */
+   GLint intLoopStart[REG_ALLOCATE_MAX_PROGRAM_TEMPS];
+   /* Temporary registers inside current top level loop */
+   GLuint regsInLoop[REG_ALLOCATE_MAX_PROGRAM_TEMPS];
+   GLuint numRegsInLoop = 0;
 
    for (i = 0; i < REG_ALLOCATE_MAX_PROGRAM_TEMPS; i++){
-      intBegin[i] = intEnd[i] = -1;
+      intBegin[i] = intEnd[i] = intLoopStart[i] = -1;
    }
 
    /* Scan instructions looking for temporary registers */
@@ -985,7 +1006,19 @@ _mesa_find_temp_intervals(const struct prog_instruction *instructions,
          loopStackDepth++;
       }
       else if (inst->Opcode == OPCODE_ENDLOOP) {
+         GLuint j;
          loopStackDepth--;
+         for (j = 0; j < numRegsInLoop; j++) {
+            GLuint index = regsInLoop[j];
+            ASSERT(intBegin[index] != -1);
+            if (intBegin[index] > loopStack[loopStackDepth].Start &&
+                intBegin[index] < loopStack[loopStackDepth].End) {
+               intLoopStart[index] = loopStack[loopStackDepth].Start;
+            }
+         }
+         if (loopStackDepth == 0) {
+            numRegsInLoop = 0;
+         }
       }
       else if (inst->Opcode == OPCODE_CAL) {
          return GL_FALSE;
@@ -998,16 +1031,18 @@ _mesa_find_temp_intervals(const struct prog_instruction *instructions,
                const GLuint index = inst->SrcReg[j].Index;
                if (inst->SrcReg[j].RelAddr)
                   return GL_FALSE;
-               update_interval(intBegin, intEnd, loopStack, loopStackDepth,
-			       index, i);
+               update_interval(intBegin, intEnd, intLoopStart, loopStack,
+                               loopStackDepth, index, i, regsInLoop,
+                               &numRegsInLoop);
             }
          }
          if (inst->DstReg.File == PROGRAM_TEMPORARY) {
             const GLuint index = inst->DstReg.Index;
             if (inst->DstReg.RelAddr)
                return GL_FALSE;
-            update_interval(intBegin, intEnd, loopStack, loopStackDepth,
-			    index, i);
+            update_interval(intBegin, intEnd, intLoopStart, loopStack,
+                            loopStackDepth, index, i, regsInLoop,
+                            &numRegsInLoop);
          }
       }
    }
-- 
1.7.4.1