From 6296350d7616f6b45546415caf7bc8e8b13717ef Mon Sep 17 00:00:00 2001
From: Joel Dice <joel.dice@gmail.com>
Date: Sat, 29 Jan 2011 18:09:47 -0700
Subject: [PATCH] fix ARM tails=true and continuations=true builds

---
 src/arm.cpp       | 29 ++++++++++++++++++++---------
 src/compile-arm.S | 44 +++++++++++---------------------------------
 2 files changed, 31 insertions(+), 42 deletions(-)

diff --git a/src/arm.cpp b/src/arm.cpp
index 8db2de4692..4abb6cf17e 100644
--- a/src/arm.cpp
+++ b/src/arm.cpp
@@ -1617,7 +1617,7 @@ argumentFootprint(unsigned footprint)
 }
 
 void
-nextFrame(ArchitectureContext* c UNUSED, uint32_t* start, unsigned size UNUSED,
+nextFrame(ArchitectureContext* c, uint32_t* start, unsigned size UNUSED,
           unsigned footprint, int32_t*, void* link, void*,
           unsigned targetParameterFootprint UNUSED, void** ip, void** stack)
 {
@@ -1640,7 +1640,8 @@ nextFrame(ArchitectureContext* c UNUSED, uint32_t* start, unsigned size UNUSED,
 
   if (instruction <= start + 2) {
     *ip = link;
-    *stack = reinterpret_cast<void**>(*stack) + offset;
+    *stack = static_cast<void**>(*stack) + offset;
+    return;
   }
 
   if (*instruction == 0xe12fff1e) { // return
@@ -1654,14 +1655,23 @@ nextFrame(ArchitectureContext* c UNUSED, uint32_t* start, unsigned size UNUSED,
         - StackAlignmentInWords;
     }
 
-    // todo: check for post-non-tail-call stack adjustment of the form
-    // "add sp, sp, #offset"
+    // check for post-non-tail-call stack adjustment of the form "add
+    // sp, sp, #offset":
+    if ((*instruction >> 12) == 0xe24dd) {
+      unsigned value = *instruction & 0xff;
+      unsigned rotation = (*instruction >> 8) & 0xf;
+      switch (rotation) {
+      case  0: offset -= value / BytesPerWord; break;
+      case 15: offset -= value; break;
+      default: abort(c);
+      }
+    }
 
-    // todo: use frameTable to check for and handle tail calls
+    // todo: check for and handle tail calls
   }
 
-  *ip = reinterpret_cast<void**>(*stack)[offset - 1];
-  *stack = reinterpret_cast<void**>(*stack) + offset;
+  *ip = static_cast<void**>(*stack)[offset - 1];
+  *stack = static_cast<void**>(*stack) + offset;
 }
 
 void
@@ -2204,14 +2214,15 @@ class MyAssembler: public Assembler {
 
     if (TailCalls) {
       if (offset) {
+        footprint += FrameHeaderSize;
+
         Register link(LinkRegister);
         Memory returnAddressSrc
           (StackRegister, (footprint - 1) * BytesPerWord);
         moveMR(&c, BytesPerWord, &returnAddressSrc, BytesPerWord, &link);
     
         Register stack(StackRegister);
-        ResolvedPromise footprintPromise
-          ((footprint - offset + 1) * BytesPerWord);
+        ResolvedPromise footprintPromise((footprint - offset) * BytesPerWord);
         Constant footprintConstant(&footprintPromise);
         addC(&c, BytesPerWord, &footprintConstant, &stack, &stack);
 
diff --git a/src/compile-arm.S b/src/compile-arm.S
index a7b0963a49..8af887fdbb 100644
--- a/src/compile-arm.S
+++ b/src/compile-arm.S
@@ -36,8 +36,6 @@
 #define CONTINUATION_FRAME_POINTER_OFFSET 24
 #define CONTINUATION_LENGTH 28
 #define CONTINUATION_BODY 32
-
-#define ARGUMENT_BASE (BYTES_PER_WORD * 2)
    
 .globl GLOBAL(vmInvoke)
 GLOBAL(vmInvoke):
@@ -117,12 +115,11 @@ GLOBAL(vmInvoke_safeStack):
    add  r7,r5,#CONTINUATION_BODY
 
    mov  r11,#0
-   add  r10,sp,#ARGUMENT_BASE
    b    LOCAL(vmInvoke_continuationTest)
 
 LOCAL(vmInvoke_continuationLoop):
    ldr  r9,[r7,r11]
-   str  r9,[r10,r11]
+   str  r9,[sp,r11]
    add  r11,r11,#4
 
 LOCAL(vmInvoke_continuationTest):
@@ -137,12 +134,6 @@ LOCAL(vmInvoke_getAddress):
    ldr  r11,[r11,r10]
    str  r11,[sp,r7]
 
-   ldr  r7,[r5,#CONTINUATION_FRAME_POINTER_OFFSET]
-   ldr  r11,[sp]
-   add  r7,r7,sp
-   str  r11,[r7]
-   str  r7,[sp]
-
    ldr  r7,[r5,#CONTINUATION_NEXT]
    str  r7,[r8,#THREAD_CONTINUATION]
 
@@ -198,40 +189,27 @@ GLOBAL(vmJumpAndInvoke):
 #ifdef AVIAN_CONTINUATIONS
    //      r0: thread
    //      r1: address
-   //      r2: (unused)
-   //      r3: stack
-   // [sp,#0]: argumentFootprint
-   // [sp,#4]: arguments
-   // [sp,#8]: frameSize
+   //      r2: stack
+   //      r3: argumentFootprint
+   // [sp,#0]: arguments
+   // [sp,#4]: frameSize
 
-   ldr  r4,[sp]
-   ldr  r5,[sp,#4]
-   ldr  r6,[sp,#8]
-   
-   // restore (pseudo)-stack pointer (we don't want to touch the real
-   // stack pointer, since we haven't copied the arguments yet)
-   ldr  r3,[r3]
-
-   // make everything between sp and r3 one big stack frame while we
-   // shuffle things around
-   str  r3,[sp]
+   ldr  r5,[sp,#0]
+   ldr  r6,[sp,#4]
 
    // allocate new frame, adding room for callee-saved registers
-   neg  r10,r6
-   add  r10,r10,#-80
-   mov  r2,r3
-	 str  r2,[r3,r10]!
+   sub  r2,r2,r6
+   sub  r2,r2,#80
    
    mov  r8,r0
 
    // copy arguments into place
    mov  r6,#0
-   add  r9,r3,#ARGUMENT_BASE
    b    LOCAL(vmJumpAndInvoke_argumentTest)
 
 LOCAL(vmJumpAndInvoke_argumentLoop):
    ldr  r12,[r5,r6]
-   str  r12,[r9,r6]
+   str  r12,[r2,r6]
    add  r6,r6,#4
 
 LOCAL(vmJumpAndInvoke_argumentTest):
@@ -240,7 +218,7 @@ LOCAL(vmJumpAndInvoke_argumentTest):
 
    // the arguments have been copied, so we can set the real stack
    // pointer now
-   mov  sp,r3
+   mov  sp,r2
    
    // set return address to vmInvoke_returnAddress
    ldr  r10,LOCAL(vmInvoke_returnAddress_word)