[36/52] target-m68k: inline shift ops
diff mbox

Message ID 1462396135-20925-4-git-send-email-laurent@vivier.eu
State New
Headers show

Commit Message

Laurent Vivier May 4, 2016, 9:08 p.m. UTC
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
---
 target-m68k/translate.c | 211 ++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 176 insertions(+), 35 deletions(-)

Comments

Richard Henderson May 6, 2016, 8:53 p.m. UTC | #1
On 05/04/2016 11:08 AM, Laurent Vivier wrote:
> -    if (count == 0) {
> -        count = 8;
> -    }
> +    count = ((count - 1) & 0x7) + 1; /* 1..8 */

How is that clearer, or even simpler?

> +                tcg_gen_setcond_i32(TCG_COND_EQ, QREG_CC_V, reg, zero);
> +                /* adjust V: (1,0) -> (0,-1) */
> +                tcg_gen_subi_i32(QREG_CC_V, QREG_CC_V, 1);

Better to use NE and neg.

> +                TCGv t0 = tcg_temp_new();
> +                TCGv t1 = tcg_const_i32(bits - 1 - count);
> +
> +                tcg_gen_shr_i32(QREG_CC_V, reg, t1);
> +                tcg_gen_sar_i32(t0, reg, t1);
> +                tcg_temp_free(t1);
> +                tcg_gen_not_i32(t0, t0);
> +
> +                tcg_gen_setcond_i32(TCG_COND_EQ, QREG_CC_V, QREG_CC_V, zero);
> +                tcg_gen_setcond_i32(TCG_COND_EQ, t0, t0, zero);
> +                tcg_gen_or_i32(QREG_CC_V, QREG_CC_V, t0); /* V is !V here */
> +
> +                tcg_temp_free(t0);
> +
> +                /* adjust V: (1,0) -> (0,-1) */
> +                tcg_gen_subi_i32(QREG_CC_V, QREG_CC_V, 1);

Can this be done more simply by creating a mask of sign bits, i.e.

     tcg_gen_sari_i32(t0, reg, 31);
     tcg_gen_shri_i32(t0, t0, bits - count);
     tcg_gen_shri_i32(t1, reg, bits - count);
     tcg_gen_setcond_i32(TCG_COND_NE, QREG_CC_V, t0, t1);
     tcg_gen_neg_i32(QREG_CC_V, QREG_CC_V);


r~

Patch
diff mbox

diff --git a/target-m68k/translate.c b/target-m68k/translate.c
index d183a3c..d48ab66 100644
--- a/target-m68k/translate.c
+++ b/target-m68k/translate.c
@@ -551,7 +551,7 @@  static inline void gen_ext(TCGv res, TCGv val, int opsize, int sign)
     }
 }
 
-static TCGv gen_extend(TCGv val, int opsize, int sign)
+static inline TCGv gen_extend(TCGv val, int opsize, int sign)
 {
     TCGv tmp;
 
@@ -2615,19 +2615,51 @@  DISAS_INSN(addx_mem)
     gen_store(s, opsize, addr_dest, QREG_CC_N);
 }
 
-DISAS_INSN(shift_im)
+static inline void shift_im(DisasContext *s, uint16_t insn, int opsize)
 {
-    TCGv reg = DREG(insn, 0);
     int count = (insn >> 9) & 7;
     int logical = insn & 8;
+    int left = insn & 0x100;
+    int bits = opsize_bytes(opsize) * 8;
+    TCGv reg = gen_extend(DREG(insn, 0), opsize, !logical);
+    TCGv zero;
 
-    if (count == 0) {
-        count = 8;
-    }
+    count = ((count - 1) & 0x7) + 1; /* 1..8 */
 
-    if (insn & 0x100) {
-        tcg_gen_shri_i32(QREG_CC_C, reg, 31 - count);
+    zero = tcg_const_i32(0);
+    if (left) {
+        tcg_gen_shri_i32(QREG_CC_C, reg, bits - count);
         tcg_gen_shli_i32(QREG_CC_N, reg, count);
+
+        /* Note that ColdFire always clears V,
+           while M68000 sets if the most significant bit is changed at
+           any time during the shift operation */
+        tcg_gen_mov_i32(QREG_CC_V, zero);
+        if (!logical && m68k_feature(s->env, M68K_FEATURE_M68000)) {
+            /* if shift count >= bits, V is (reg != 0) */
+            if (count >= bits) {
+                tcg_gen_setcond_i32(TCG_COND_EQ, QREG_CC_V, reg, zero);
+                /* adjust V: (1,0) -> (0,-1) */
+                tcg_gen_subi_i32(QREG_CC_V, QREG_CC_V, 1);
+            } else {
+                TCGv t0 = tcg_temp_new();
+                TCGv t1 = tcg_const_i32(bits - 1 - count);
+
+                tcg_gen_shr_i32(QREG_CC_V, reg, t1);
+                tcg_gen_sar_i32(t0, reg, t1);
+                tcg_temp_free(t1);
+                tcg_gen_not_i32(t0, t0);
+
+                tcg_gen_setcond_i32(TCG_COND_EQ, QREG_CC_V, QREG_CC_V, zero);
+                tcg_gen_setcond_i32(TCG_COND_EQ, t0, t0, zero);
+                tcg_gen_or_i32(QREG_CC_V, QREG_CC_V, t0); /* V is !V here */
+
+                tcg_temp_free(t0);
+
+                /* adjust V: (1,0) -> (0,-1) */
+                tcg_gen_subi_i32(QREG_CC_V, QREG_CC_V, 1);
+            }
+        }
     } else {
         tcg_gen_shri_i32(QREG_CC_C, reg, count - 1);
         if (logical) {
@@ -2635,30 +2667,28 @@  DISAS_INSN(shift_im)
         } else {
             tcg_gen_sari_i32(QREG_CC_N, reg, count);
         }
+        tcg_gen_mov_i32(QREG_CC_V, zero);
     }
+
+    gen_ext(QREG_CC_N, QREG_CC_N, opsize, 1);
     tcg_gen_andi_i32(QREG_CC_C, QREG_CC_C, 1);
     tcg_gen_mov_i32(QREG_CC_Z, QREG_CC_N);
     tcg_gen_mov_i32(QREG_CC_X, QREG_CC_C);
 
-    /* Note that ColdFire always clears V, while M68000 sets it for
-       a change in the sign bit.  */
-    if (!logical && m68k_feature(s->env, M68K_FEATURE_M68000)) {
-        tcg_gen_xor_i32(QREG_CC_V, QREG_CC_N, reg);
-    } else {
-        tcg_gen_movi_i32(QREG_CC_V, 0);
-    }
-
-    tcg_gen_mov_i32(reg, QREG_CC_N);
+    gen_partset_reg(opsize, DREG(insn, 0), QREG_CC_N);
     set_cc_op(s, CC_OP_FLAGS);
 }
 
-DISAS_INSN(shift_reg)
+static inline void shift_reg(DisasContext *s, uint16_t insn, int opsize)
 {
-    TCGv reg, s32;
-    TCGv_i64 t64, s64;
     int logical = insn & 8;
+    int left = insn & 0x100;
+    int bits = opsize_bytes(opsize) * 8;
+    TCGv reg = gen_extend(DREG(insn, 0), opsize, !logical);
+    TCGv s32;
+    TCGv_i64 t64, s64;
+    TCGv zero;
 
-    reg = DREG(insn, 0);
     t64 = tcg_temp_new_i64();
     s64 = tcg_temp_new_i64();
     s32 = tcg_temp_new();
@@ -2669,44 +2699,148 @@  DISAS_INSN(shift_reg)
     tcg_gen_andi_i32(s32, DREG(insn, 9), 63);
     tcg_gen_extu_i32_i64(s64, s32);
 
-    /* Non-arithmetic shift clears V.  Use it as a source zero here.  */
-    tcg_gen_movi_i32(QREG_CC_V, 0);
+    zero = tcg_const_i32(0);
 
-    if (insn & 0x100) {
-        tcg_gen_extu_i32_i64(t64, reg);
+    tcg_gen_extu_i32_i64(t64, reg);
+    if (left) {
+        tcg_gen_shli_i64(t64, t64, 32 - bits);
         tcg_gen_shl_i64(t64, t64, s64);
         tcg_temp_free_i64(s64);
         tcg_gen_extr_i64_i32(QREG_CC_N, QREG_CC_C, t64);
         tcg_temp_free_i64(t64);
+        tcg_gen_sari_i32(QREG_CC_N, QREG_CC_N, 32 - bits);
         tcg_gen_andi_i32(QREG_CC_C, QREG_CC_C, 1);
+
+        /* Note that ColdFire always clears V,
+           while M68000 sets if the most significant bit is changed at
+           any time during the shift operation */
+        tcg_gen_mov_i32(QREG_CC_V, zero);
+        if (!logical && m68k_feature(s->env, M68K_FEATURE_M68000)) {
+
+            TCGv t1 = tcg_const_i32(bits - 1);
+            TCGv t0 = tcg_temp_new();
+
+            tcg_gen_sub_i32(t0, t1, s32);
+            tcg_gen_shr_i32(QREG_CC_V, reg, t0);
+            tcg_gen_sar_i32(t0, reg, t0);
+            tcg_gen_not_i32(t0, t0);
+
+            tcg_gen_setcond_i32(TCG_COND_EQ, QREG_CC_V, QREG_CC_V, zero);
+            tcg_gen_setcond_i32(TCG_COND_EQ, t0, t0, zero);
+            tcg_gen_or_i32(QREG_CC_V, QREG_CC_V, t0); /* V is !V here */
+
+            /* if shift count >= bits, V is (reg != 0) */
+            tcg_gen_setcond_i32(TCG_COND_EQ, t0, reg, zero);
+            tcg_gen_movcond_i32(TCG_COND_GT, QREG_CC_V, s32, t1, t0, QREG_CC_V);
+
+            tcg_temp_free(t0);
+            tcg_temp_free(t1);
+
+            /* adjust V: (1,0) -> (0,-1) */
+            tcg_gen_subi_i32(QREG_CC_V, QREG_CC_V, 1);
+
+            /* if shift count is zero, V is 0 */
+            tcg_gen_movcond_i32(TCG_COND_NE, QREG_CC_V, s32, zero,
+                                QREG_CC_V, zero);
+        }
     } else {
-        tcg_gen_extu_i32_i64(t64, reg);
-        tcg_gen_shli_i64(t64, t64, 32);
+        tcg_gen_shli_i64(t64, t64, 64 - bits);
         if (logical) {
+            tcg_gen_shri_i64(t64, t64, 32 - bits);
             tcg_gen_shr_i64(t64, t64, s64);
         } else {
+            tcg_gen_sari_i64(t64, t64, 32 - bits);
             tcg_gen_sar_i64(t64, t64, s64);
         }
         tcg_temp_free_i64(s64);
         tcg_gen_extr_i64_i32(QREG_CC_C, QREG_CC_N, t64);
         tcg_temp_free_i64(t64);
+        gen_ext(QREG_CC_N, QREG_CC_N, opsize, 1);
         tcg_gen_shri_i32(QREG_CC_C, QREG_CC_C, 31);
+        tcg_gen_mov_i32(QREG_CC_V, zero);
     }
     tcg_gen_mov_i32(QREG_CC_Z, QREG_CC_N);
 
-    /* Note that X = C, but only if the shift count was non-zero.  */
-    tcg_gen_movcond_i32(TCG_COND_NE, QREG_CC_X, s32, QREG_CC_V,
+    /* C is cleared if shift count was zero */
+    tcg_gen_movcond_i32(TCG_COND_NE, QREG_CC_C, s32, zero,
+                        QREG_CC_C, zero);
+
+    /* X = C, but only if the shift count was non-zero.  */
+    tcg_gen_movcond_i32(TCG_COND_NE, QREG_CC_X, s32, zero,
                         QREG_CC_C, QREG_CC_X);
+    tcg_temp_free(zero);
     tcg_temp_free(s32);
 
-    /* Note that ColdFire always clears V (which we have done above),
-       while M68000 sets it for a change in the sign bit.  */
-    if (!logical && m68k_feature(s->env, M68K_FEATURE_M68000)) {
-        tcg_gen_xor_i32(QREG_CC_V, QREG_CC_N, reg);
+    /* Write back the result.  */
+    gen_partset_reg(opsize, DREG(insn, 0), QREG_CC_N);
+    set_cc_op(s, CC_OP_FLAGS);
+}
+
+DISAS_INSN(shift8_im)
+{
+    shift_im(s, insn, OS_BYTE);
+}
+
+DISAS_INSN(shift16_im)
+{
+    shift_im(s, insn, OS_WORD);
+}
+
+DISAS_INSN(shift_im)
+{
+    shift_im(s, insn, OS_LONG);
+}
+
+DISAS_INSN(shift8_reg)
+{
+    shift_reg(s, insn, OS_BYTE);
+}
+
+DISAS_INSN(shift16_reg)
+{
+    shift_reg(s, insn, OS_WORD);
+}
+
+DISAS_INSN(shift_reg)
+{
+    shift_reg(s, insn, OS_LONG);
+}
+
+DISAS_INSN(shift_mem)
+{
+    int logical = insn & 8;
+    int left = insn & 0x100;
+    TCGv src;
+    TCGv addr;
+
+    SRC_EA(env, src, OS_WORD, !logical, &addr);
+    tcg_gen_movi_i32(QREG_CC_V, 0);
+    if (left) {
+        tcg_gen_shri_i32(QREG_CC_C, src, 15);
+        tcg_gen_shli_i32(QREG_CC_N, src, 1);
+
+        /* Note that ColdFire always clears V,
+           while M68000 sets if the most significant bit is changed at
+           any time during the shift operation */
+        if (!logical && m68k_feature(s->env, M68K_FEATURE_M68000)) {
+            src = gen_extend(src, OS_WORD, 1);
+            tcg_gen_xor_i32(QREG_CC_V, QREG_CC_N, src);
+        }
+    } else {
+        tcg_gen_mov_i32(QREG_CC_C, src);
+        if (logical) {
+            tcg_gen_shri_i32(QREG_CC_N, src, 1);
+        } else {
+            tcg_gen_sari_i32(QREG_CC_N, src, 1);
+        }
     }
 
-    /* Write back the result.  */
-    tcg_gen_mov_i32(reg, QREG_CC_N);
+    gen_ext(QREG_CC_N, QREG_CC_N, OS_WORD, 1);
+    tcg_gen_andi_i32(QREG_CC_C, QREG_CC_C, 1);
+    tcg_gen_mov_i32(QREG_CC_Z, QREG_CC_N);
+    tcg_gen_mov_i32(QREG_CC_X, QREG_CC_C);
+
+    DEST_EA(env, insn, OS_WORD, QREG_CC_N, &addr);
     set_cc_op(s, CC_OP_FLAGS);
 }
 
@@ -4451,6 +4585,13 @@  void register_m68k_insns (CPUM68KState *env)
     INSN(adda,      d0c0, f0c0, M68000);
     INSN(shift_im,  e080, f0f0, CF_ISA_A);
     INSN(shift_reg, e0a0, f0f0, CF_ISA_A);
+    INSN(shift8_im, e000, f0f0, M68000);
+    INSN(shift16_im, e040, f0f0, M68000);
+    INSN(shift_im,  e080, f0f0, M68000);
+    INSN(shift8_reg, e020, f0f0, M68000);
+    INSN(shift16_reg, e060, f0f0, M68000);
+    INSN(shift_reg, e0a0, f0f0, M68000);
+    INSN(shift_mem, e0c0, fcc0, M68000);
     INSN(rotate_im, e090, f0f0, M68000);
     INSN(rotate8_im, e010, f0f0, M68000);
     INSN(rotate16_im, e050, f0f0, M68000);