Skip to contentSkip to conversation

target/sh4: Fix SUBV opcode

RFC PATCH Superseded Philippe Mathieu-Daudé 2 participants 4 messages Apr 30, 2024

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 target/sh4: Fix SUBV opcode

The documentation says:

SUBV Rm, Rn Rn - Rm -> Rn, underflow -> T

While correctly performing the substraction, the underflow is not detected.

While we can check the high xored bit for overflow, for underflow we need to check the xored value is not negative.

Fixes ad8d25a11f ("target-sh4: implement addv and subv using TCG")

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2318

---
 target/sh4/translate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/sh4/translate.c +1 -1
index 4a1dd0d1f4..1c48d8ebea 100644
@@ -937,7 +937,7 @@ static void _decode_opc(DisasContext * ctx)
             t2 = tcg_temp_new();
             tcg_gen_xor_i32(t2, REG(B11_8), REG(B7_4));
             tcg_gen_and_i32(t1, t1, t2);
-            tcg_gen_shri_i32(cpu_sr_t, t1, 31);
+            tcg_gen_setcondi_i32(TCG_COND_GE, cpu_sr_t, t1, 0);
             tcg_gen_mov_i32(REG(B11_8), t0);
         }
         return;
Signature · 1 line
2.41.0
#1 lore raw

Hi Philippe,

If I'm not mistaken, the overflow / underflow can be calculated like this:

T = ((Rn ^ Rm) & (Result ^ Rn)) >> 31

Looking at what Qemu does (before this patch), it was doing this:
T = ((Rn ^ Rm) & (Result ^ Rm)) >> 31

I changed line 936 to this, and overflow / underflow with SUBV now seem to work fine:

tcg_gen_xor_i32(t1, t0, REG(B11_8));

So a change from REG(B7_B4) to REG(B11_8).

Cheers,
-Paul

Le mardi 30 avril 2024 à 14:06 +0200, Philippe Mathieu-Daudé a écrit :

Quoted 32 lines

The documentation says:

  SUBV Rm, Rn        Rn - Rm -> Rn, underflow -> T

While correctly performing the substraction, the underflow is not detected.

While we can check the high xored bit for overflow, for underflow we need to check the xored value is not negative.

Fixes ad8d25a11f ("target-sh4: implement addv and subv using TCG")

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2318

---
 target/sh4/translate.c | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)
target/sh4/translate.c +1 -1
index 4a1dd0d1f4..1c48d8ebea 100644
@@ -937,7 +937,7 @@ static void _decode_opc(DisasContext * ctx)
              t2 = tcg_temp_new();
              tcg_gen_xor_i32(t2, REG(B11_8), REG(B7_4));
              tcg_gen_and_i32(t1, t1, t2);
-            tcg_gen_shri_i32(cpu_sr_t, t1, 31);
+            tcg_gen_setcondi_i32(TCG_COND_GE, cpu_sr_t, t1, 0);
              tcg_gen_mov_i32(REG(B11_8), t0);
          }
          return;
#2 lore raw
Quoted 10 lines · Paul Cercueil

Hi Philippe,

If I'm not mistaken, the overflow / underflow can be calculated like this:

T = ((Rn ^ Rm) & (Result ^ Rn)) >> 31

Looking at what Qemu does (before this patch), it was doing this:
T = ((Rn ^ Rm) & (Result ^ Rm)) >> 31

Quoted 6 lines

I changed line 936 to this, and overflow / underflow with SUBV now seem to work fine:

tcg_gen_xor_i32(t1, t0, REG(B11_8));

So a change from REG(B7_B4) to REG(B11_8).

Correct, thanks!

Quoted 38 lines

Cheers,
-Paul

Le mardi 30 avril 2024 à 14:06 +0200, Philippe Mathieu-Daudé a écrit :

Quoted 32 lines

The documentation says:

  SUBV Rm, Rn        Rn - Rm -> Rn, underflow -> T

While correctly performing the substraction, the underflow is not detected.

While we can check the high xored bit for overflow, for underflow we need to check the xored value is not negative.

Fixes ad8d25a11f ("target-sh4: implement addv and subv using TCG")

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2318

---
  target/sh4/translate.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
target/sh4/translate.c +1 -1
index 4a1dd0d1f4..1c48d8ebea 100644
@@ -937,7 +937,7 @@ static void _decode_opc(DisasContext * ctx)
              t2 = tcg_temp_new();
              tcg_gen_xor_i32(t2, REG(B11_8), REG(B7_4));
              tcg_gen_and_i32(t1, t1, t2);
-            tcg_gen_shri_i32(cpu_sr_t, t1, 31);
+            tcg_gen_setcondi_i32(TCG_COND_GE, cpu_sr_t, t1, 0);
              tcg_gen_mov_i32(REG(B11_8), t0);
          }
          return;
#3 lore raw
In reply to Philippe Mathieu-Daudé

Hi Philippe,

Le mardi 30 avril 2024 à 14:06 +0200, Philippe Mathieu-Daudé a écrit :

Quoted 9 lines

The documentation says:

  SUBV Rm, Rn        Rn - Rm -> Rn, underflow -> T

While correctly performing the substraction, the underflow is not detected.

While we can check the high xored bit for overflow, for underflow we need to check the xored value is not negative.

This fix still does not work properly; it will incorrectly set the underflow bit when the sign changes from positive to negative.

e.g. Rn == 0 and Rm == 2, the result will be Rn == -2, without any underflow.

Cheers,
-Paul

Quoted 23 lines
Fixes ad8d25a11f ("target-sh4: implement addv and subv using TCG")

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2318

---
 target/sh4/translate.c | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)
target/sh4/translate.c +1 -1
index 4a1dd0d1f4..1c48d8ebea 100644
@@ -937,7 +937,7 @@ static void _decode_opc(DisasContext * ctx)
              t2 = tcg_temp_new();
              tcg_gen_xor_i32(t2, REG(B11_8), REG(B7_4));
              tcg_gen_and_i32(t1, t1, t2);
-            tcg_gen_shri_i32(cpu_sr_t, t1, 31);
+            tcg_gen_setcondi_i32(TCG_COND_GE, cpu_sr_t, t1, 0);
              tcg_gen_mov_i32(REG(B11_8), t0);
          }
          return;