Fix and tidy up missing_gcc_functions.c (#1529)

1 files changed, 160 insertions, 123 deletions

diff --git a/src/gcc_fix/missing_gcc_functions.c b/src/gcc_fix/missing_gcc_functions.c
index 4c572a764..c62418c7b 100644
--- a/src/gcc_fix/missing_gcc_functions.c
+++ b/src/gcc_fix/missing_gcc_functions.c
@@ -7,7 +7,9 @@
 
 #include "global.h"
 
-// Self-hosted memcmp.
+// Self-hosted libc memory functions, gcc assumes these exist even in a freestanding
+// environment and there is no way to tell it otherwise.
+
 int memcmp(void* s1, const void* s2, size_t n) {
     u8* m1 = (u8*)s1;
     u8* m2 = (u8*)s2;
@@ -35,177 +37,212 @@ void* memset(void* str, s32 c, size_t n) {
     return str;
 }
 
-// These functions convert c to an unsigned integer, rounding toward zero. Negative values
-// all become zero.
-u32 __fixunssfdi(f32 a) {
-    if (a < 0.0f) {
-        a = 0.0f;
-    }
+// Conversions involving 64-bit integer types required by the O32 MIPS ABI.
 
-    return (u32)a;
-}
+// f32 -> u64, negative values become 0
+u64 __fixunssfdi(f32 a) {
+    if (a > 0.0f) {
+        register union {
+            f64 f;
+            u64 i;
+        } m;
 
-u32 __fixunsdfdi(f64 a) {
-    if (a < 0.0) {
-        a = 0.0;
+        __asm__ ("cvt.l.s %0, %1" : "=f"(m.f) : "f"(a));
+        return m.i;
     }
+    return 0;
+}
+
+// f64 -> u64, negative values become 0
+u64 __fixunsdfdi(f64 a) {
+    if (a > 0.0) {
+        register union {
+            f64 f;
+            u64 i;
+        } m;
 
-    return (u32)a;
+        __asm__ ("cvt.l.d %0, %1" : "=f"(m.f) : "f"(a));
+        return m.i;
+    }
+    return 0;
 }
 
-// These functions convert c to a signed integer, rounding toward zero.
-s32 __fixsfdi(f32 c) {
-    return (s32)c;
+// f32 -> s64
+s64 __fixsfdi(f32 c) {
+    register union {
+        f64 f;
+        s64 i;
+    } m;
+
+    __asm__ ("cvt.l.s %0, %1" : "=f"(m.f) : "f"(c));
+    return m.i;
 }
 
-s32 __fixdfdi(f64 c) {
-    return (s32)c;
+// f64 -> s64
+s64 __fixdfdi(f64 c) {
+    register union {
+        f64 f;
+        s64 i;
+    } m;
+
+    __asm__ ("cvt.l.d %0, %1" : "=f"(m.f) : "f"(c));
+    return m.i;
 }
 
-// These functions convert c, a signed integer, to floating point.
-f32 __floatdisf(s32 c) {
-    return (f32)c;
+// s64 -> f32
+f32 __floatdisf(s64 c) {
+    register union {
+        f64 f;
+        s64 i;
+    } m;
+    register f32 v;
+
+    m.i = c;
+    __asm__ ("cvt.s.l %0, %1" : "=f"(v) : "f"(m.f));
+    return v;
 }
 
-f64 __floatdidf(s32 c) {
-    return (f64)c;
+// s64 -> f64
+f64 __floatdidf(s64 c) {
+    register union {
+        f64 f;
+        s64 i;
+    } m;
+    register f64 v;
+
+    m.i = c;
+    __asm__ ("cvt.d.l %0, %1" : "=f"(v) : "f"(m.f));
+    return v;
 }
 
-// These functions convert c, an unsigned integer, to floating point.
-f32 __floatundisf(u32 c) {
-    return (f32)c;
+// u64 -> f32
+f32 __floatundisf(u64 c) {
+    register union {
+        f64 f;
+        u64 i;
+    } m;
+    register f32 v;
+
+    m.i = c;
+    __asm__ ("cvt.s.l %0, %1" : "=f"(v) : "f"(m.f));
+    if ((s64)c < 0) {
+        // cvt.s.l assumes signed input, adjust output
+        v += 4294967296.0f; // 2^32
+    }
+    return v;
 }
 
-f64 __floatundidf(u32 c) {
-    return (f64)c;
+// u64 -> f64
+f64 __floatundidf(u64 c) {
+    register union {
+        f64 f;
+        u64 i;
+    } m;
+    register f64 v;
+
+    m.i = c;
+    __asm__ ("cvt.d.l %0, %1" : "=f"(v) : "f"(m.f));
+    if ((s64)c < 0) {
+        // cvt.d.l assumes signed input, adjust output
+        v += 18446744073709551616.0; // 2^64
+    }
+    return v;
 }
 
-f32 __powisf2(f32 a, s32 b) {
-    const s32 recip = b < 0;
-    f32 r = 1;
+// Compute x^m by binary exponentiation
 
-    while (1) {
-        if (b & 1) {
-            r *= a;
-        }
+f32 __powisf2(f32 x, s32 m) {
+    u32 n = (m < 0) ? -m : m;
+    f32 y = (n % 2 != 0) ? x : 1.0f;
 
-        b /= 2;
+    while (n >>= 1) {
+        x = x * x;
 
-        if (b == 0) {
-            break;
+        if (n % 2 != 0) {
+            y = y * x;
         }
-
-        a *= a;
     }
-    return recip ? 1 / r : r;
+    return (m < 0) ? (1.0f / y) : y;
 }
 
+// Compute division and modulo of 64-bit signed and unsigned integers
+
 __asm__("                                   \n\
     .set push                               \n\
-    .set noat                               \n\
     .set noreorder                          \n\
     .set gp=64                              \n\
                                             \n\
 .global __umoddi3                           \n\
 __umoddi3:                                  \n\
-    .type __umoddi3, @function              \n\
-    .ent __umoddi3                          \n\
-    sw    $a0, ($sp)                        \n\
-    sw    $a1, 4($sp)                       \n\
-    sw    $a2, 8($sp)                       \n\
-    sw    $a3, 0xc($sp)                     \n\
-    ld    $t7, 8($sp)                       \n\
-    ld    $t6, ($sp)                        \n\
-    ddivu $zero, $t6, $t7                   \n\
-    bnez  $t7, 1f                           \n\
-     nop                                    \n\
-    break 7                                 \n\
-1:                                          \n\
-    mfhi  $v0                               \n\
-    dsll32 $v1, $v0, 0                      \n\
-    dsra32 $v1, $v1, 0                      \n\
-    jr    $ra                               \n\
+  .type __umoddi3, @function                \n\
+  .ent __umoddi3                            \n\
+    sw      $a0, 0x0($sp)                   \n\
+    sw      $a1, 0x4($sp)                   \n\
+    sw      $a2, 0x8($sp)                   \n\
+    sw      $a3, 0xC($sp)                   \n\
+    ld      $t6, 0($sp)                     \n\
+    ld      $t7, 8($sp)                     \n\
+    dremu   $v0, $t6, $t7                   \n\
+    dsll32  $v1, $v0, 0                     \n\
+    dsra32  $v1, $v1, 0                     \n\
+    jr      $ra                             \n\
      dsra32 $v0, $v0, 0                     \n\
-    .end __umoddi3                          \n\
-    .size __umoddi3, . - __umoddi3          \n\
+  .end __umoddi3                            \n\
+  .size __umoddi3, . - __umoddi3            \n\
                                             \n\
 .global __udivdi3                           \n\
 __udivdi3:                                  \n\
-    .type __udivdi3, @function              \n\
-    .ent __udivdi3                          \n\
-    sw    $a0, ($sp)                        \n\
-    sw    $a1, 4($sp)                       \n\
-    sw    $a2, 8($sp)                       \n\
-    sw    $a3, 0xc($sp)                     \n\
-    ld    $t7, 8($sp)                       \n\
-    ld    $t6, ($sp)                        \n\
-    ddivu $zero, $t6, $t7                   \n\
-    bnez  $t7, 1f                           \n\
-     nop                                    \n\
-    break 7                                 \n\
-1:                                          \n\
-    mflo  $v0                               \n\
-    dsll32 $v1, $v0, 0                      \n\
-    dsra32 $v1, $v1, 0                      \n\
-    jr    $ra                               \n\
+  .type __udivdi3, @function                \n\
+  .ent __udivdi3                            \n\
+    sw      $a0, 0x0($sp)                   \n\
+    sw      $a1, 0x4($sp)                   \n\
+    sw      $a2, 0x8($sp)                   \n\
+    sw      $a3, 0xC($sp)                   \n\
+    ld      $t6, 0($sp)                     \n\
+    ld      $t7, 8($sp)                     \n\
+    ddivu   $v0, $t6, $t7                   \n\
+    dsll32  $v1, $v0, 0                     \n\
+    dsra32  $v1, $v1, 0                     \n\
+    jr      $ra                             \n\
      dsra32 $v0, $v0, 0                     \n\
-    .end __udivdi3                          \n\
-    .size __udivdi3, . - __udivdi3          \n\
+  .end __udivdi3                            \n\
+  .size __udivdi3, . - __udivdi3            \n\
                                             \n\
 .global __moddi3                            \n\
 __moddi3:                                   \n\
-    .type __moddi3, @function               \n\
-    .ent __moddi3                           \n\
-    sw    $a0, ($sp)                        \n\
-    sw    $a1, 4($sp)                       \n\
-    sw    $a2, 8($sp)                       \n\
-    sw    $a3, 0xc($sp)                     \n\
-    ld    $t7, 8($sp)                       \n\
-    ld    $t6, ($sp)                        \n\
-    ddivu $zero, $t6, $t7                   \n\
-    bnez  $t7, 1f                           \n\
-     nop                                    \n\
-    break 7                                 \n\
-1:                                          \n\
-    mfhi  $v0                               \n\
-    dsll32 $v1, $v0, 0                      \n\
-    dsra32 $v1, $v1, 0                      \n\
-    jr    $ra                               \n\
+  .type __moddi3, @function                 \n\
+  .ent __moddi3                             \n\
+    sw      $a0, 0x0($sp)                   \n\
+    sw      $a1, 0x4($sp)                   \n\
+    sw      $a2, 0x8($sp)                   \n\
+    sw      $a3, 0xC($sp)                   \n\
+    ld      $t6, 0($sp)                     \n\
+    ld      $t7, 8($sp)                     \n\
+    drem    $v0, $t6, $t7                   \n\
+    dsll32  $v1, $v0, 0                     \n\
+    dsra32  $v1, $v1, 0                     \n\
+    jr      $ra                             \n\
      dsra32 $v0, $v0, 0                     \n\
-    .end __moddi3                           \n\
-    .size __moddi3, . - __moddi3            \n\
+  .end __moddi3                             \n\
+  .size __moddi3, . - __moddi3              \n\
                                             \n\
 .global __divdi3                            \n\
 __divdi3:                                   \n\
-    .type __divdi3, @function               \n\
-    .ent __divdi3                           \n\
-    sw    $a0, ($sp)                        \n\
-    sw    $a1, 4($sp)                       \n\
-    sw    $a2, 8($sp)                       \n\
-    sw    $a3, 0xc($sp)                     \n\
+  .type __divdi3, @function                 \n\
+  .ent __divdi3                             \n\
+    sw    $a0, 0x0($sp)                     \n\
+    sw    $a1, 0x4($sp)                     \n\
+    sw    $a2, 0x8($sp)                     \n\
+    sw    $a3, 0xC($sp)                     \n\
+    ld    $t6, 0($sp)                       \n\
     ld    $t7, 8($sp)                       \n\
-    ld    $t6, ($sp)                        \n\
-    ddiv  $zero, $t6, $t7                   \n\
-    nop                                     \n\
-    bnez  $t7, 1f                           \n\
-     nop                                    \n\
-    break 7                                 \n\
-1:                                          \n\
-    daddiu $at, $zero, -1                   \n\
-    bne   $t7, $at, 2f                      \n\
-     daddiu $at, $zero, 1                   \n\
-    dsll32 $at, $at, 0x1f                   \n\
-    bne   $t6, $at, 2f                      \n\
-     nop                                    \n\
-    break 6                                 \n\
-2:                                          \n\
-    mflo  $v0                               \n\
+    ddiv  $v0, $t6, $t7                     \n\
     dsll32 $v1, $v0, 0                      \n\
     dsra32 $v1, $v1, 0                      \n\
     jr    $ra                               \n\
      dsra32 $v0, $v0, 0                     \n\
-    .end __divdi3                           \n\
-    .size __divdi3, . - __divdi3            \n\
+  .end __divdi3                             \n\
+  .size __divdi3, . - __divdi3              \n\
                                             \n\
     .set pop                                \n\
                                             \n");