Changeset 7b2c2c5f for src/libcfa

Timestamp:

Mar 9, 2017, 3:49:30 PM (8 years ago)

Author:

Thierry Delisle <tdelisle@…>

Branches:

ADT, aaron-thesis, arm-eh, ast-experimental, cleanup-dtors, deferred_resn, demangler, enum, forall-pointer-decay, jacob/cs343-translation, jenkins-sandbox, master, new-ast, new-ast-unique-expr, new-env, no_list, persistent-indexer, pthread-emulation, qualifiedEnum, resolv-new, with_gc

Children:

149d297

Parents:

5222605

Message:

Fixed saving and loading of control words for i386

Location:

src/libcfa/concurrency

Files:

• CtxSwitch-i386.S (modified) (3 diffs)
• CtxSwitch-x86_64.S (modified) (2 diffs)
• invoke.c (modified) (2 diffs)

src/libcfa/concurrency/CtxSwitch-i386.S

@@ -52 +52 @@
         movl 4(%esp),%eax
 
+        // Save floating & SSE control words on the stack.
+
+        sub    $8,%esp
+        stmxcsr 0(%esp)         // 4 bytes
+        fnstcw  4(%esp)         // 2 bytes
+
         // Save volatile registers on the stack.
 
@@ -69 +75 @@
         // argument is now at 8 + 12 = 20(%esp)
 
-        movl 20(%esp),%eax
+        movl 28(%esp),%eax
 
         // Load new context from the "to" area.
@@ -81 +87 @@
         popl %edi
         popl %ebx
+
+        // Load floating & SSE control words from the stack.
+
+        fldcw   4(%esp)
+        ldmxcsr 0(%esp)
+        add    $8,%esp
 
         // Return to thread.

src/libcfa/concurrency/CtxSwitch-x86_64.S

@@ -47 +47 @@
 CtxSwitch:
 
-        // Save volatile registers on the stack.
+        // Save floating & SSE control words on the stack.
 
         subq   $8,%rsp
         stmxcsr 0(%rsp)         // 4 bytes
         fnstcw  4(%rsp)         // 2 bytes
+
+        // Save volatile registers on the stack.
+
         pushq %r15
         pushq %r14
@@ -75 +78 @@
         popq %r14
         popq %r15
+
+        // Load floating & SSE control words from the stack.
+
         fldcw   4(%rsp)
         ldmxcsr 0(%rsp)
-        addq $8,%rsp
+        addq   $8,%rsp
 
         // Return to thread.

src/libcfa/concurrency/invoke.c

@@ -91 +91 @@
         struct FakeStack {
             void *fixedRegisters[3];                    // fixed registers ebx, edi, esi (popped on 1st uSwitch, values unimportant)
-            void *rturn;                                      // where to go on return from uSwitch
+            uint32_t mxcr;                              // SSE Status and Control bits (control bits are preserved across function calls)
+            uint16_t fcw;                               // X97 FPU control word (preserved across function calls)
+            void *rturn;                                // where to go on return from uSwitch
             void *dummyReturn;                          // fake return compiler would have pushed on call to uInvoke
             void *argument[3];                          // for 16-byte ABI, 16-byte alignment starts here
@@ -108 +110 @@
       struct FakeStack {
             void *fixedRegisters[5];                    // fixed registers rbx, r12, r13, r14, r15
-            uint32_t mxcr;                                  // SSE Status and Control bits (control bits are preserved across function calls)
-            uint16_t fcw;                                   // X97 FPU control word (preserved across function calls)
-            void *rturn;                                      // where to go on return from uSwitch
+            uint32_t mxcr;                              // SSE Status and Control bits (control bits are preserved across function calls)
+            uint16_t fcw;                               // X97 FPU control word (preserved across function calls)
+            void *rturn;                                // where to go on return from uSwitch
             void *dummyReturn;                          // NULL return address to provide proper alignment
       };