compiler: use src register as dest2

Also try to use the src register as the second destination register for
opcodes with 2 results. Saves one unnecessary mov for splitX opcodes.

diff --git a/orc/orccompiler.c b/orc/orccompiler.c
index 46ae8fe..3618f5b 100644 (file)
--- a/orc/orccompiler.c
+++ b/orc/orccompiler.c
@@ -919,10 +919,14 @@ orc_compiler_rewrite_vars2 (OrcCompiler *compiler)
      */
     if (compiler->insns[j].flags & ORC_INSN_FLAG_INVARIANT) continue;
 
-    if (!(compiler->insns[j].opcode->flags & ORC_STATIC_OPCODE_ACCUMULATOR)
-        && compiler->insns[j].opcode->dest_size[1] == 0) {
+    if (!(compiler->insns[j].opcode->flags & ORC_STATIC_OPCODE_ACCUMULATOR)) {
       int src1 = compiler->insns[j].src_args[0];
-      int dest = compiler->insns[j].dest_args[0];
+      int dest;
+
+      if (compiler->insns[j].opcode->dest_size[1] == 0)
+        dest = compiler->insns[j].dest_args[0];
+      else
+        dest = compiler->insns[j].dest_args[1];
 
       if (compiler->vars[src1].last_use == j) {
         if (compiler->vars[src1].first_use == j) {

diff --git a/orc/orcrules-mmx.c b/orc/orcrules-mmx.c
index 82f7c20..3a79628 100644 (file)
--- a/orc/orcrules-mmx.c
+++ b/orc/orcrules-mmx.c
@@ -1737,10 +1737,11 @@ mmx_rule_splitql (OrcCompiler *p, void *user, OrcInstruction *insn)
   int dest2 = p->vars[insn->dest_args[1]].alloc;
 
 #ifndef MMX
-  orc_mmx_emit_pshufd (p, ORC_MMX_SHUF(2,0,2,0), src, dest2);
   orc_mmx_emit_pshufd (p, ORC_MMX_SHUF(3,1,3,1), src, dest1);
+  orc_mmx_emit_pshufd (p, ORC_MMX_SHUF(2,0,2,0), src, dest2);
 #else
-  orc_mmx_emit_movq (p, src, dest2);
+  if (dest2 != src)
+    orc_mmx_emit_movq (p, src, dest2);
   orc_mmx_emit_pshufw (p, ORC_MMX_SHUF(3,2,3,2), src, dest1);
 #endif
 }
@@ -1757,9 +1758,8 @@ mmx_rule_splitlw (OrcCompiler *p, void *user, OrcInstruction *insn)
   orc_mmx_emit_psrad_imm (p, 16, dest1);
   orc_mmx_emit_packssdw (p, dest1, dest1);
 
-  if (dest2 != src) {
+  if (dest2 != src)
     orc_mmx_emit_movq (p, src, dest2);
-  }
   orc_mmx_emit_pslld_imm (p, 16, dest2);
   orc_mmx_emit_psrad_imm (p, 16, dest2);
   orc_mmx_emit_packssdw (p, dest2, dest2);
@@ -1779,9 +1779,8 @@ mmx_rule_splitwb (OrcCompiler *p, void *user, OrcInstruction *insn)
   orc_mmx_emit_psraw_imm (p, 8, dest1);
   orc_mmx_emit_packsswb (p, dest1, dest1);
 
-  if (dest2 != src) {
+  if (dest2 != src)
     orc_mmx_emit_movq (p, src, dest2);
-  }
 
 #if 0
   orc_mmx_emit_psllw_imm (p, 8, dest2);

diff --git a/orc/orcrules-sse.c b/orc/orcrules-sse.c
index ea9a5c6..c2c13d5 100644 (file)
--- a/orc/orcrules-sse.c
+++ b/orc/orcrules-sse.c
@@ -1737,8 +1737,8 @@ sse_rule_splitql (OrcCompiler *p, void *user, OrcInstruction *insn)
   int dest2 = p->vars[insn->dest_args[1]].alloc;
 
 #ifndef MMX
-  orc_sse_emit_pshufd (p, ORC_SSE_SHUF(2,0,2,0), src, dest2);
   orc_sse_emit_pshufd (p, ORC_SSE_SHUF(3,1,3,1), src, dest1);
+  orc_sse_emit_pshufd (p, ORC_SSE_SHUF(2,0,2,0), src, dest2);
 #else
   orc_sse_emit_movdqa (p, src, dest2);
   orc_sse_emit_pshufw (p, ORC_SSE_SHUF(3,2,3,2), src, dest1);
@@ -1757,9 +1757,8 @@ sse_rule_splitlw (OrcCompiler *p, void *user, OrcInstruction *insn)
   orc_sse_emit_psrad_imm (p, 16, dest1);
   orc_sse_emit_packssdw (p, dest1, dest1);
 
-  if (dest2 != src) {
+  if (dest2 != src)
     orc_sse_emit_movdqa (p, src, dest2);
-  }
   orc_sse_emit_pslld_imm (p, 16, dest2);
   orc_sse_emit_psrad_imm (p, 16, dest2);
   orc_sse_emit_packssdw (p, dest2, dest2);
@@ -1774,14 +1773,14 @@ sse_rule_splitwb (OrcCompiler *p, void *user, OrcInstruction *insn)
   int dest2 = p->vars[insn->dest_args[1]].alloc;
   int tmp = orc_compiler_get_constant (p, 2, 0xff);
 
+  ORC_DEBUG ("got tmp %d", tmp);
   /* FIXME slow */
 
   orc_sse_emit_psraw_imm (p, 8, dest1);
   orc_sse_emit_packsswb (p, dest1, dest1);
 
-  if (dest2 != src) {
+  if (dest2 != src)
     orc_sse_emit_movdqa (p, src, dest2);
-  }
 
 #if 0
   orc_sse_emit_psllw_imm (p, 8, dest2);