Add a PowerPC branch detector

PowerPCs have static instruction lengths, so writing an "in-between" brl detection is quite simple on this architecture. You are welcome to write something like this for any other platform, but if a compile doesn't trigger build errors on PowerPC, it shouldn't on other platforms either, as PowerPC has quite advanced branch instructions.

This patch is not mandatory, makes debugging a lot easier though.

Index: qemu-snapshot-2008-01-15_05/dyngen.c
===================================================================

dyngen.c +0 -0

@@ -1488,6 +1488,16 @@ void gen_code(const char *name, host_ulo
         if (get32((uint32_t *)p) != 0x4e800020)
             error("blr expected at the end of %s", name);
         copy_size = p - p_start;

Quoted 23 lines · Alexander Graf

PowerPCs have static instruction lengths, so writing an "in-between" brl detection is quite simple on this architecture. You are welcome to write something like this for any other platform, but if a compile doesn't trigger build errors on PowerPC, it shouldn't on other platforms either, as PowerPC has quite advanced branch instructions.

This patch is not mandatory, makes debugging a lot easier though.

Index: qemu-snapshot-2008-01-15_05/dyngen.c
===================================================================

dyngen.c +5 -0

@@ -1488,6 +1488,16 @@ void gen_code(const char *name, host_ulo
         if (get32((uint32_t *)p) != 0x4e800020)
             error("blr expected at the end of %s", name);
         copy_size = p - p_start;
+
+/* blr check for inline returns */
+
+        if(strstart(name, "op_", NULL) && !strstart(name, "op_exit", NULL)) {
+            for(p=p_start; p < p_end - 4; p+=4) {

dyngen.c void gen_code(const char *name, host_ulo

Quoted 6 lines

dyngen.c +5 -0

+                if ((get32((uint32_t *)p) & 0xfc00fff0) == 0x4c000020) {
+                    error("Inline blr detected in %s. Please append FORCE_RET to the function.", name);
+                }
+            }
+        }
 

Is check_ops.sh not enough for debugging micro-ops?

Quoted 2 lines

Is check_ops.sh not enough for debugging micro-ops?

Basically it should be. PowerPC branching can be (regex) b..rl. Honestly I did not know about this script though and as it was not in the makefile, it did not tell me that something wrong was going on. This check costs near no time and has to be passed in order to build successfully.

So either check_ops should be fixed (not only brl) and put into the Makefile.target or a check like this is good to have.

Alexander Graf wrote:
[snip]

Quoted 8 lines

Quoted 2 lines

Is check_ops.sh not enough for debugging micro-ops?

Basically it should be. PowerPC branching can be (regex) b..rl. Honestly I did not know about this script though and as it was not in the makefile, it did not tell me that something wrong was going on. This check costs near no time and has to be passed in order to build successfully.

Quoted 2 lines

So either check_ops should be fixed (not only brl) and put into the Makefile.target or a check like this is good to have.

Fixing check_ops is IMO preferable, OTOH I hope the whole problem goes away with the new code generator which is in the works.

Hi,

Quoted 10 lines · Thiemo Seufer

Alexander Graf wrote:
[snip]

Quoted 8 lines

Quoted 2 lines

Is check_ops.sh not enough for debugging micro-ops?

Basically it should be. PowerPC branching can be (regex) b..rl. Honestly I did not know about this script though and as it was not in the makefile, it did not tell me that something wrong was going on. This check costs near no time and has to be passed in order to build successfully.

Quoted 5 lines

Quoted 2 lines

So either check_ops should be fixed (not only brl) and put into the Makefile.target or a check like this is good to have.

Fixing check_ops is IMO preferable, OTOH I hope the whole problem goes away with the new code generator which is in the works.

The code generated by that new code generator is substantially slower than what we have right now, so I am not so enthusiastic.

Quoted 23 lines · Johannes Schindelin

Hi,

Quoted 18 lines · Thiemo Seufer

Alexander Graf wrote:
[snip]

Quoted 12 lines

Quoted 3 lines

Is check_ops.sh not enough for debugging micro-ops?

Basically it should be. PowerPC branching can be (regex) b..rl. Honestly I did not know about this script though and as it was not in the makefile, it did not tell me that something wrong was going on. This check costs near no time and has to be passed in order to build successfully.

So either check_ops should be fixed (not only brl) and put into the Makefile.target or a check like this is good to have.

Fixing check_ops is IMO preferable, OTOH I hope the whole problem goes away with the new code generator which is in the works.

Quoted 3 lines

The code generated by that new code generator is substantially slower than what we have right now, so I am not so enthusiastic.

I meant not qops but something Fabrice is working on. Which apparently has about the same speed.

Quoted 5 lines

Quoted 2 lines

I hope the whole problem goes
away with the new code generator which is in the works.

The code generated by that new code generator is substantially slower than what we have right now, so I am not so enthusiastic.

I meant not qops but something Fabrice is working on. Which apparently has about the same speed.