]> Git Repo - qemu.git/blobdiff - tcg/tcg-op.c
tcg/arm: Try pc-relative addresses for movi
[qemu.git] / tcg / tcg-op.c
index 2b520c19536239f2287d2327f4a00adc472581bd..87f673ef49c6fb569fa03444916bc1eb6b95ad0e 100644 (file)
@@ -497,33 +497,27 @@ void tcg_gen_ctz_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
         tcg_gen_extrl_i64_i32(ret, t1);
         tcg_temp_free_i64(t1);
         tcg_temp_free_i64(t2);
-    } else if (TCG_TARGET_HAS_clz_i32) {
-        TCGv_i32 t1 = tcg_temp_new_i32();
-        TCGv_i32 t2 = tcg_temp_new_i32();
-        tcg_gen_neg_i32(t1, arg1);
-        tcg_gen_xori_i32(t2, arg2, 31);
-        tcg_gen_and_i32(t1, t1, arg1);
-        tcg_gen_clz_i32(ret, t1, t2);
-        tcg_temp_free_i32(t1);
-        tcg_temp_free_i32(t2);
-        tcg_gen_xori_i32(ret, ret, 31);
-    } else if (TCG_TARGET_HAS_clz_i64) {
-        TCGv_i32 t1 = tcg_temp_new_i32();
-        TCGv_i32 t2 = tcg_temp_new_i32();
-        TCGv_i64 x1 = tcg_temp_new_i64();
-        TCGv_i64 x2 = tcg_temp_new_i64();
-        tcg_gen_neg_i32(t1, arg1);
-        tcg_gen_xori_i32(t2, arg2, 63);
-        tcg_gen_and_i32(t1, t1, arg1);
-        tcg_gen_extu_i32_i64(x1, t1);
-        tcg_gen_extu_i32_i64(x2, t2);
-        tcg_temp_free_i32(t1);
-        tcg_temp_free_i32(t2);
-        tcg_gen_clz_i64(x1, x1, x2);
-        tcg_gen_extrl_i64_i32(ret, x1);
-        tcg_temp_free_i64(x1);
-        tcg_temp_free_i64(x2);
-        tcg_gen_xori_i32(ret, ret, 63);
+    } else if (TCG_TARGET_HAS_ctpop_i32
+               || TCG_TARGET_HAS_ctpop_i64
+               || TCG_TARGET_HAS_clz_i32
+               || TCG_TARGET_HAS_clz_i64) {
+        TCGv_i32 z, t = tcg_temp_new_i32();
+
+        if (TCG_TARGET_HAS_ctpop_i32 || TCG_TARGET_HAS_ctpop_i64) {
+            tcg_gen_subi_i32(t, arg1, 1);
+            tcg_gen_andc_i32(t, t, arg1);
+            tcg_gen_ctpop_i32(t, t);
+        } else {
+            /* Since all non-x86 hosts have clz(0) == 32, don't fight it.  */
+            tcg_gen_neg_i32(t, arg1);
+            tcg_gen_and_i32(t, t, arg1);
+            tcg_gen_clzi_i32(t, t, 32);
+            tcg_gen_xori_i32(t, t, 31);
+        }
+        z = tcg_const_i32(0);
+        tcg_gen_movcond_i32(TCG_COND_EQ, ret, arg1, z, arg2, t);
+        tcg_temp_free_i32(t);
+        tcg_temp_free_i32(z);
     } else {
         gen_helper_ctz_i32(ret, arg1, arg2);
     }
@@ -531,9 +525,47 @@ void tcg_gen_ctz_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
 
 void tcg_gen_ctzi_i32(TCGv_i32 ret, TCGv_i32 arg1, uint32_t arg2)
 {
-    TCGv_i32 t = tcg_const_i32(arg2);
-    tcg_gen_ctz_i32(ret, arg1, t);
-    tcg_temp_free_i32(t);
+    if (!TCG_TARGET_HAS_ctz_i32 && TCG_TARGET_HAS_ctpop_i32 && arg2 == 32) {
+        /* This equivalence has the advantage of not requiring a fixup.  */
+        TCGv_i32 t = tcg_temp_new_i32();
+        tcg_gen_subi_i32(t, arg1, 1);
+        tcg_gen_andc_i32(t, t, arg1);
+        tcg_gen_ctpop_i32(ret, t);
+        tcg_temp_free_i32(t);
+    } else {
+        TCGv_i32 t = tcg_const_i32(arg2);
+        tcg_gen_ctz_i32(ret, arg1, t);
+        tcg_temp_free_i32(t);
+    }
+}
+
+void tcg_gen_clrsb_i32(TCGv_i32 ret, TCGv_i32 arg)
+{
+    if (TCG_TARGET_HAS_clz_i32) {
+        TCGv_i32 t = tcg_temp_new_i32();
+        tcg_gen_sari_i32(t, arg, 31);
+        tcg_gen_xor_i32(t, t, arg);
+        tcg_gen_clzi_i32(t, t, 32);
+        tcg_gen_subi_i32(ret, t, 1);
+        tcg_temp_free_i32(t);
+    } else {
+        gen_helper_clrsb_i32(ret, arg);
+    }
+}
+
+void tcg_gen_ctpop_i32(TCGv_i32 ret, TCGv_i32 arg1)
+{
+    if (TCG_TARGET_HAS_ctpop_i32) {
+        tcg_gen_op2_i32(INDEX_op_ctpop_i32, ret, arg1);
+    } else if (TCG_TARGET_HAS_ctpop_i64) {
+        TCGv_i64 t = tcg_temp_new_i64();
+        tcg_gen_extu_i32_i64(t, arg1);
+        tcg_gen_ctpop_i64(t, t);
+        tcg_gen_extrl_i64_i32(ret, t);
+        tcg_temp_free_i64(t);
+    } else {
+        gen_helper_ctpop_i32(ret, arg1);
+    }
 }
 
 void tcg_gen_rotl_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
@@ -1813,16 +1845,24 @@ void tcg_gen_ctz_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2)
 {
     if (TCG_TARGET_HAS_ctz_i64) {
         tcg_gen_op3_i64(INDEX_op_ctz_i64, ret, arg1, arg2);
-    } else if (TCG_TARGET_HAS_clz_i64) {
-        TCGv_i64 t1 = tcg_temp_new_i64();
-        TCGv_i64 t2 = tcg_temp_new_i64();
-        tcg_gen_neg_i64(t1, arg1);
-        tcg_gen_xori_i64(t2, arg2, 63);
-        tcg_gen_and_i64(t1, t1, arg1);
-        tcg_gen_clz_i64(ret, t1, t2);
-        tcg_temp_free_i64(t1);
-        tcg_temp_free_i64(t2);
-        tcg_gen_xori_i64(ret, ret, 63);
+    } else if (TCG_TARGET_HAS_ctpop_i64 || TCG_TARGET_HAS_clz_i64) {
+        TCGv_i64 z, t = tcg_temp_new_i64();
+
+        if (TCG_TARGET_HAS_ctpop_i64) {
+            tcg_gen_subi_i64(t, arg1, 1);
+            tcg_gen_andc_i64(t, t, arg1);
+            tcg_gen_ctpop_i64(t, t);
+        } else {
+            /* Since all non-x86 hosts have clz(0) == 64, don't fight it.  */
+            tcg_gen_neg_i64(t, arg1);
+            tcg_gen_and_i64(t, t, arg1);
+            tcg_gen_clzi_i64(t, t, 64);
+            tcg_gen_xori_i64(t, t, 63);
+        }
+        z = tcg_const_i64(0);
+        tcg_gen_movcond_i64(TCG_COND_EQ, ret, arg1, z, arg2, t);
+        tcg_temp_free_i64(t);
+        tcg_temp_free_i64(z);
     } else {
         gen_helper_ctz_i64(ret, arg1, arg2);
     }
@@ -1839,6 +1879,15 @@ void tcg_gen_ctzi_i64(TCGv_i64 ret, TCGv_i64 arg1, uint64_t arg2)
         tcg_gen_ctz_i32(TCGV_LOW(ret), TCGV_LOW(arg1), t32);
         tcg_gen_movi_i32(TCGV_HIGH(ret), 0);
         tcg_temp_free_i32(t32);
+    } else if (!TCG_TARGET_HAS_ctz_i64
+               && TCG_TARGET_HAS_ctpop_i64
+               && arg2 == 64) {
+        /* This equivalence has the advantage of not requiring a fixup.  */
+        TCGv_i64 t = tcg_temp_new_i64();
+        tcg_gen_subi_i64(t, arg1, 1);
+        tcg_gen_andc_i64(t, t, arg1);
+        tcg_gen_ctpop_i64(ret, t);
+        tcg_temp_free_i64(t);
     } else {
         TCGv_i64 t64 = tcg_const_i64(arg2);
         tcg_gen_ctz_i64(ret, arg1, t64);
@@ -1846,6 +1895,34 @@ void tcg_gen_ctzi_i64(TCGv_i64 ret, TCGv_i64 arg1, uint64_t arg2)
     }
 }
 
+void tcg_gen_clrsb_i64(TCGv_i64 ret, TCGv_i64 arg)
+{
+    if (TCG_TARGET_HAS_clz_i64 || TCG_TARGET_HAS_clz_i32) {
+        TCGv_i64 t = tcg_temp_new_i64();
+        tcg_gen_sari_i64(t, arg, 63);
+        tcg_gen_xor_i64(t, t, arg);
+        tcg_gen_clzi_i64(t, t, 64);
+        tcg_gen_subi_i64(ret, t, 1);
+        tcg_temp_free_i64(t);
+    } else {
+        gen_helper_clrsb_i64(ret, arg);
+    }
+}
+
+void tcg_gen_ctpop_i64(TCGv_i64 ret, TCGv_i64 arg1)
+{
+    if (TCG_TARGET_HAS_ctpop_i64) {
+        tcg_gen_op2_i64(INDEX_op_ctpop_i64, ret, arg1);
+    } else if (TCG_TARGET_REG_BITS == 32 && TCG_TARGET_HAS_ctpop_i32) {
+        tcg_gen_ctpop_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1));
+        tcg_gen_ctpop_i32(TCGV_LOW(ret), TCGV_LOW(arg1));
+        tcg_gen_add_i32(TCGV_LOW(ret), TCGV_LOW(ret), TCGV_HIGH(ret));
+        tcg_gen_movi_i32(TCGV_HIGH(ret), 0);
+    } else {
+        gen_helper_ctpop_i64(ret, arg1);
+    }
+}
+
 void tcg_gen_rotl_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2)
 {
     if (TCG_TARGET_HAS_rot_i64) {
@@ -2510,6 +2587,18 @@ void tcg_gen_goto_tb(unsigned idx)
     tcg_gen_op1i(INDEX_op_goto_tb, idx);
 }
 
+void tcg_gen_lookup_and_goto_ptr(TCGv addr)
+{
+    if (TCG_TARGET_HAS_goto_ptr && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
+        TCGv_ptr ptr = tcg_temp_new_ptr();
+        gen_helper_lookup_tb_ptr(ptr, tcg_ctx.tcg_env, addr);
+        tcg_gen_op1i(INDEX_op_goto_ptr, GET_TCGV_PTR(ptr));
+        tcg_temp_free_ptr(ptr);
+    } else {
+        tcg_gen_exit_tb(0);
+    }
+}
+
 static inline TCGMemOp tcg_canonicalize_memop(TCGMemOp op, bool is64, bool st)
 {
     /* Trigger the asserts within as early as possible.  */
@@ -2784,6 +2873,9 @@ void tcg_gen_atomic_cmpxchg_i64(TCGv_i64 retv, TCGv addr, TCGv_i64 cmpv,
 #endif
 #else
         gen_helper_exit_atomic(tcg_ctx.tcg_env);
+        /* Produce a result, so that we have a well-formed opcode stream
+           with respect to uses of the result in the (dead) code following.  */
+        tcg_gen_movi_i64(retv, 0);
 #endif /* CONFIG_ATOMIC64 */
     } else {
         TCGv_i32 c32 = tcg_temp_new_i32();
@@ -2889,6 +2981,9 @@ static void do_atomic_op_i64(TCGv_i64 ret, TCGv addr, TCGv_i64 val,
 #endif
 #else
         gen_helper_exit_atomic(tcg_ctx.tcg_env);
+        /* Produce a result, so that we have a well-formed opcode stream
+           with respect to uses of the result in the (dead) code following.  */
+        tcg_gen_movi_i64(ret, 0);
 #endif /* CONFIG_ATOMIC64 */
     } else {
         TCGv_i32 v32 = tcg_temp_new_i32();
This page took 0.029112 seconds and 4 git commands to generate.