[Guile-commits] 264/437: ARM: Do not use ldrt/strt by default

[Andy Wingo]

wingo pushed a commit to branch lightning
in repository guile.

commit 2d4bac43a9b746ec93fa02832fa4cc8ae7ea0a53
Author: pcpa <address@hidden>
Date:   Tue Oct 8 12:26:52 2013 -0300

    ARM: Do not use ldrt/strt by default
    
        * include/lightning/jit_arm.h, lib/jit_arm-cpu.c: Do not use
        by default load/store instructions that map to ldrt/strt.
        There is already the long displacement version for positive
        offsets, and when using a (shorter) negative offset it does
        not map to ldrt/strt. At least on qemu strt may cause
        reproducible, but unexpected SIGILL.
---
 ChangeLog                   |  9 +++++++++
 include/lightning/jit_arm.h |  6 ++++++
 lib/jit_arm-cpu.c           | 16 ++++++++--------
 3 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 1fe7608..13aa358 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,14 @@
 2013-10-08 Paulo Andrade <address@hidden>
 
+       * include/lightning/jit_arm.h, lib/jit_arm-cpu.c: Do not use
+       by default load/store instructions that map to ldrt/strt.
+       There is already the long displacement version for positive
+       offsets, and when using a (shorter) negative offset it does
+       not map to ldrt/strt. At least on qemu strt may cause
+       reproducible, but unexpected SIGILL.
+
+2013-10-08 Paulo Andrade <address@hidden>
+
        * lib/jit_arm-vfp.c: Correct wrong load/store offset
        calculation when the displacement is constant but too
        large to use an instruction with an immediate offset.
diff --git a/include/lightning/jit_arm.h b/include/lightning/jit_arm.h
index 6c8ede8..1db7593 100644
--- a/include/lightning/jit_arm.h
+++ b/include/lightning/jit_arm.h
@@ -28,6 +28,7 @@
  */
 #define jit_swf_p()            (jit_cpu.vfp == 0)
 #define jit_hardfp_p()         jit_cpu.abi
+#define jit_ldrt_strt_p()      jit_cpu.ldrt_strt
 
 #define JIT_FP                 _R11
 typedef enum {
@@ -115,6 +116,11 @@ typedef struct {
     jit_uint32_t vfp           : 3;
     jit_uint32_t neon          : 1;
     jit_uint32_t abi           : 2;
+    /* use strt+offset instead of str.w?
+     * on special cases it causes a SIGILL at least on qemu, probably
+     * due to some memory ordering constraint not being respected, so,
+     * disable by default */
+    jit_uint32_t ldrt_strt     : 1;
 } jit_cpu_t;
 
 typedef jit_int64_t            jit_regset_t;
diff --git a/lib/jit_arm-cpu.c b/lib/jit_arm-cpu.c
index c91ac6e..c0419c3 100644
--- a/lib/jit_arm-cpu.c
+++ b/lib/jit_arm-cpu.c
@@ -2953,7 +2953,7 @@ _ldxi_c(jit_state_t *_jit, jit_int32_t r0, jit_int32_t 
r1, jit_word_t i0)
 {
     jit_int32_t                reg;
     if (jit_thumb_p()) {
-       if (i0 >= 0 && i0 <= 255)
+       if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_LDRSBI(r0, r1, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_LDRSBIN(r0, r1, -i0);
@@ -3036,7 +3036,7 @@ _ldxi_uc(jit_state_t *_jit, jit_int32_t r0, jit_int32_t 
r1, jit_word_t i0)
     if (jit_thumb_p()) {
        if ((r0|r1) < 8 && i0 >= 0 && i0 < 0x20)
            T1_LDRBI(r0, r1, i0);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_LDRBI(r0, r1, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_LDRBIN(r0, r1, -i0);
@@ -3117,7 +3117,7 @@ _ldxi_s(jit_state_t *_jit, jit_int32_t r0, jit_int32_t 
r1, jit_word_t i0)
 {
     jit_int32_t                reg;
     if (jit_thumb_p()) {
-       if (i0 >= 0 && i0 <= 255)
+       if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_LDRSHI(r0, r1, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_LDRSHIN(r0, r1, -i0);
@@ -3200,7 +3200,7 @@ _ldxi_us(jit_state_t *_jit, jit_int32_t r0, jit_int32_t 
r1, jit_word_t i0)
     if (jit_thumb_p()) {
        if ((r0|r1) < 8 && i0 >= 0 && !(i0 & 1) && (i0 >> 1) < 0x20)
            T1_LDRHI(r0, r1, i0 >> 1);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_LDRHI(r0, r1, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_LDRHIN(r0, r1, -i0);
@@ -3286,7 +3286,7 @@ _ldxi_i(jit_state_t *_jit, jit_int32_t r0, jit_int32_t 
r1, jit_word_t i0)
        else if (r1 == _R13_REGNO && r0 < 8 &&
                 i0 >= 0 && !(i0 & 3) && (i0 >> 2) <= 255)
            T1_LDRISP(r0, i0 >> 2);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_LDRI(r0, r1, i0);
        else if (i0 < 0 && i0 > -255)
            T2_LDRIN(r0, r1, -i0);
@@ -3369,7 +3369,7 @@ _stxi_c(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, 
jit_int32_t r1)
     if (jit_thumb_p()) {
        if ((r0|r1) < 8 && i0 >= 0 && i0 < 0x20)
            T1_STRBI(r1, r0, i0);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_STRBI(r1, r0, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_STRBIN(r1, r0, -i0);
@@ -3441,7 +3441,7 @@ _stxi_s(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, 
jit_int32_t r1)
     if (jit_thumb_p()) {
        if ((r0|r1) < 8 && i0 >= 0 && !(i0 & 1) && (i0 >> 1) < 0x20)
            T1_STRHI(r1, r0, i0 >> 1);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_STRHI(r1, r0, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_STRHIN(r1, r0, -i0);
@@ -3516,7 +3516,7 @@ _stxi_i(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, 
jit_int32_t r1)
        else if (r0 == _R13_REGNO && r1 < 8 &&
                 i0 >= 0 && !(i0 & 3) && (i0 >> 2) <= 255)
            T1_STRISP(r1, i0 >> 2);
-       else if (i0 >= 0 && i0 <= 255)
+       else if (jit_ldrt_strt_p() && i0 >= 0 && i0 <= 255)
            T2_STRI(r1, r0, i0);
        else if (i0 < 0 && i0 >= -255)
            T2_STRIN(r1, r0, -i0);