[Qemu-devel] [PATCH] target-ppc: fix evmergelo and evmergelohi

[Qemu-devel] [PATCH] target-ppc: fix evmergelo and evmergelohi

[Nathan Froyd]

For 32-bit PPC targets, we translated:

evmergelo rX, rX, rY

as:

rX-lo = rY-lo
rX-hi = rX-lo

which is wrong, because we should be transferring rX-lo first.  This
problem is fixed by swapping the order in which we write the parts of
rX.

Similarly, we translated:

evmergelohi rX, rX, rY

as:

rX-lo = rY-hi
rX-hi = rX-lo

In this case, we can't swap the assignment statements, because that
would just cause problems for:

evmergelohi rX, rY, rX

Instead, we detect the first case and save rX-lo in a temporary
variable:

tmp = rX-lo
rX-lo = rY-hi
rX-hi = tmp

These problems don't occur on PPC64 targets because we don't split the
SPE registers into hi/lo parts for such targets.

Signed-off-by: Nathan Froyd <froydnj@codesourcery.com>
---
 target-ppc/translate.c |   14 +++++++++++---
 1 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/target-ppc/translate.c b/target-ppc/translate.c
index 24c78d1..c61667a 100644
--- a/target-ppc/translate.c
+++ b/target-ppc/translate.c
@@ -7018,8 +7018,8 @@ static always_inline void gen_evmergelo (DisasContext *ctx)
     tcg_temp_free(t0);
     tcg_temp_free(t1);
 #else
-    tcg_gen_mov_i32(cpu_gpr[rD(ctx->opcode)], cpu_gpr[rB(ctx->opcode)]);
     tcg_gen_mov_i32(cpu_gprh[rD(ctx->opcode)], cpu_gpr[rA(ctx->opcode)]);
+    tcg_gen_mov_i32(cpu_gpr[rD(ctx->opcode)], cpu_gpr[rB(ctx->opcode)]);
 #endif
 }
 static always_inline void gen_evmergehilo (DisasContext *ctx)
@@ -7056,8 +7056,16 @@ static always_inline void gen_evmergelohi (DisasContext *ctx)
     tcg_temp_free(t0);
     tcg_temp_free(t1);
 #else
-    tcg_gen_mov_i32(cpu_gpr[rD(ctx->opcode)], cpu_gprh[rB(ctx->opcode)]);
-    tcg_gen_mov_i32(cpu_gprh[rD(ctx->opcode)], cpu_gpr[rA(ctx->opcode)]);
+    if (rD(ctx->opcode) == rA(ctx->opcode)) {
+        TCGv_i32 tmp = tcg_temp_new_i32();
+        tcg_gen_mov_i32(tmp, cpu_gpr[rA(ctx->opcode)]);
+        tcg_gen_mov_i32(cpu_gpr[rD(ctx->opcode)], cpu_gprh[rB(ctx->opcode)]);
+        tcg_gen_mov_i32(cpu_gprh[rD(ctx->opcode)], tmp);
+        tcg_temp_free_i32(tmp);
+    } else {
+        tcg_gen_mov_i32(cpu_gpr[rD(ctx->opcode)], cpu_gprh[rB(ctx->opcode)]);
+        tcg_gen_mov_i32(cpu_gprh[rD(ctx->opcode)], cpu_gpr[rA(ctx->opcode)]);
+    }
 #endif
 }
 static always_inline void gen_evsplati (DisasContext *ctx)
-- 
1.6.3.2

Re: [Qemu-devel] [PATCH] target-ppc: fix evmergelo and evmergelohi

[Nathan Froyd]

On Thu, Jun 04, 2009 at 01:46:41PM -0700, Nathan Froyd wrote:
> For 32-bit PPC targets, we translated:
> 
> evmergelo rX, rX, rY
> 
> as:
> 
> rX-lo = rY-lo
> rX-hi = rX-lo
> 
> which is wrong, because we should be transferring rX-lo first.  This
> problem is fixed by swapping the order in which we write the parts of
> rX.
> 
> Similarly, we translated:
> 
> evmergelohi rX, rX, rY
> 
> as:
> 
> rX-lo = rY-hi
> rX-hi = rX-lo
> 
> In this case, we can't swap the assignment statements, because that
> would just cause problems for:
> 
> evmergelohi rX, rY, rX
> 
> Instead, we detect the first case and save rX-lo in a temporary
> variable:
> 
> tmp = rX-lo
> rX-lo = rY-hi
> rX-hi = tmp
> 
> These problems don't occur on PPC64 targets because we don't split the
> SPE registers into hi/lo parts for such targets.

Ping.

-Nathan

Re: [Qemu-devel] [PATCH] target-ppc: fix evmergelo and evmergelohi

[Nathan Froyd]

On Thu, Jun 04, 2009 at 01:46:41PM -0700, Nathan Froyd wrote:
> For 32-bit PPC targets, we translated:
> 
> evmergelo rX, rX, rY
> 
> as:
> 
> rX-lo = rY-lo
> rX-hi = rX-lo
> 
> which is wrong, because we should be transferring rX-lo first.  This
> problem is fixed by swapping the order in which we write the parts of
> rX.
> 
> Similarly, we translated:
> 
> evmergelohi rX, rX, rY
> 
> as:
> 
> rX-lo = rY-hi
> rX-hi = rX-lo
> 
> In this case, we can't swap the assignment statements, because that
> would just cause problems for:
> 
> evmergelohi rX, rY, rX
> 
> Instead, we detect the first case and save rX-lo in a temporary
> variable:
> 
> tmp = rX-lo
> rX-lo = rY-hi
> rX-hi = tmp
> 
> These problems don't occur on PPC64 targets because we don't split the
> SPE registers into hi/lo parts for such targets.

Ping again.

-Nathan

Re: [Qemu-devel] [PATCH] target-ppc: fix evmergelo and evmergelohi

[Aurelien Jarno]

On Thu, Jun 04, 2009 at 01:46:41PM -0700, Nathan Froyd wrote:
> For 32-bit PPC targets, we translated:
> 
> evmergelo rX, rX, rY
> 
> as:
> 
> rX-lo = rY-lo
> rX-hi = rX-lo
> 
> which is wrong, because we should be transferring rX-lo first.  This
> problem is fixed by swapping the order in which we write the parts of
> rX.
> 
> Similarly, we translated:
> 
> evmergelohi rX, rX, rY
> 
> as:
> 
> rX-lo = rY-hi
> rX-hi = rX-lo
> 
> In this case, we can't swap the assignment statements, because that
> would just cause problems for:
> 
> evmergelohi rX, rY, rX
> 
> Instead, we detect the first case and save rX-lo in a temporary
> variable:
> 
> tmp = rX-lo
> rX-lo = rY-hi
> rX-hi = tmp
> 
> These problems don't occur on PPC64 targets because we don't split the
> SPE registers into hi/lo parts for such targets.
> 

Thanks, applied. 

Sorry for the delay I was away from QEMU for a few months. I am now
(slowly) start to work again on it. Don't hesitate to ping for other
patches I may have forget.

-- 
Aurelien Jarno	                        GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net