Re: [Qemu-devel] [5281] Use the new concat_i32_i64 op for std and stda

[Paul Brook <paul@codesourcery.com>]

On Sunday 21 September 2008, Blue Swirl wrote:
> +    r_low = tcg_temp_new(TCG_TYPE_I32);
> +    tcg_gen_trunc_tl_i32(r_low, cpu_tmp0);
> +    tcg_gen_trunc_tl_i32(cpu_tmp32, hi);
> +    tcg_gen_concat_i32_i64(cpu_tmp64, r_low, cpu_tmp32);
> +    tcg_temp_free(r_low);

It's more efficient to not use concat_i32_i64 on 64-bit targets.
See patch below. I'll let you decide if you want to apply or ignore it.

Signed-off-by: Paul Brook <paul@codesourcery.com>

Index: target-sparc/helper.h
===================================================================
--- target-sparc/helper.h	(revision 5280)
+++ target-sparc/helper.h	(working copy)
@@ -44,7 +44,6 @@ DEF_HELPER(void, helper_restore, (void))
 DEF_HELPER(void, helper_flush, (target_ulong addr))
 DEF_HELPER(target_ulong, helper_udiv, (target_ulong a, target_ulong b))
 DEF_HELPER(target_ulong, helper_sdiv, (target_ulong a, target_ulong b))
-DEF_HELPER(uint64_t, helper_pack64, (target_ulong high, target_ulong low))
 DEF_HELPER(void, helper_stdf, (target_ulong addr, int mem_idx))
 DEF_HELPER(void, helper_lddf, (target_ulong addr, int mem_idx))
 DEF_HELPER(void, helper_ldqf, (target_ulong addr, int mem_idx))
Index: target-sparc/op_helper.c
===================================================================
--- target-sparc/op_helper.c	(revision 5280)
+++ target-sparc/op_helper.c	(working copy)
@@ -2281,11 +2281,6 @@ target_ulong helper_sdiv(target_ulong a,
     }
 }
 
-uint64_t helper_pack64(target_ulong high, target_ulong low)
-{
-    return ((uint64_t)high << 32) | (uint64_t)(low & 0xffffffff);
-}
-
 void helper_stdf(target_ulong addr, int mem_idx)
 {
     helper_check_align(addr, 7);
Index: target-sparc/translate.c
===================================================================
--- target-sparc/translate.c	(revision 5280)
+++ target-sparc/translate.c	(working copy)
@@ -1619,6 +1619,31 @@ static inline void gen_clear_float_excep
     tcg_gen_helper_0_0(helper_clear_float_exceptions);
 }
 
+/* May clobber low.  */
+static void gen_pack64(TCGv dest, TCGv high, TCGv low)
+{
+#ifdef TARGET_SPARC64
+#if HOST_LONG_BITS == 64
+    tcg_gen_shli_i64(dest, high, 32);
+    tcg_gen_ext32u_i64(low, low);
+    tcg_gen_or_i64(dest, dest, low);
+# else
+    /* We could omit the truncation if we assumed that 64-bit TCG variables
+       can be used in 32-bit ops.  Hopefully TCG will be smart enough to
+       eliminate the copy.  */
+    TCGv low32 = tcg_temp_new(TCG_TYPE_I32);
+    TCGv high32 = tcg_temp_new(TCG_TYPE_I32);
+    tcg_gen_trunc_i64_i32(low32, low);
+    tcg_gen_trunc_i64_i32(high32, high);
+    tcg_gen_concat_i32_i64(dest, low32, high32);
+    tcg_temp_free(high32);
+    tcg_temp_free(low32);
+# endif
+#else /* !TARGET_SPARC64 */
+    tcg_gen_concat_i32_i64(dest, low, high);
+#endif
+}
+
 /* asi moves */
 #ifdef TARGET_SPARC64
 static inline TCGv gen_get_asi(int insn, TCGv r_addr)
@@ -1719,8 +1744,7 @@ static inline void gen_stda_asi(TCGv hi,
 
     r_temp = tcg_temp_new(TCG_TYPE_TL);
     gen_movl_reg_TN(rd + 1, r_temp);
-    tcg_gen_helper_1_2(helper_pack64, cpu_tmp64, hi,
-                       r_temp);
+    gen_pack64(cpu_tmp64, hi, r_temp);
     tcg_temp_free(r_temp);
     r_asi = gen_get_asi(insn, addr);
     r_size = tcg_const_i32(8);
@@ -1821,7 +1845,7 @@ static inline void gen_stda_asi(TCGv hi,
 
     r_temp = tcg_temp_new(TCG_TYPE_TL);
     gen_movl_reg_TN(rd + 1, r_temp);
-    tcg_gen_helper_1_2(helper_pack64, cpu_tmp64, hi, r_temp);
+    gen_pack64(cpu_tmp64, hi, r_temp);
     tcg_temp_free(r_temp);
     r_asi = tcg_const_i32(GET_FIELD(insn, 19, 26));
     r_size = tcg_const_i32(8);
@@ -4484,8 +4508,7 @@ static void disas_sparc_insn(DisasContex
                         tcg_temp_free(r_const);
                         r_low = tcg_temp_new(TCG_TYPE_TL);
                         gen_movl_reg_TN(rd + 1, r_low);
-                        tcg_gen_helper_1_2(helper_pack64, cpu_tmp64, cpu_val,
-                                           r_low);
+                        gen_pack64(cpu_tmp64, cpu_val, r_low);
                         tcg_temp_free(r_low);
                         tcg_gen_qemu_st64(cpu_tmp64, cpu_addr, dc->mem_idx);
                     }