[compiler-rt][builtins] Use c[tl]zsi macro instead of __builtin_c[tl]z

[atrosinenko]

Summary

__builtin_c[tl]z accepts unsigned int argument that is not always the
same as uint32_t. For example, unsigned int is uint16_t on MSP430.

[uabelho]

We've run with changes like this for a long time to make the routines work for our out of tree target where int is also 16 bits so thumbs up from me even if I'm not sure I can formally LGTM it.

[atrosinenko]

Ping.

[MaskRay]

The (aWidth - 1) - clzsi(a) change is correct, but why is the ctz change?

[atrosinenko]

In D86547#2284095, @MaskRay wrote:

The (aWidth - 1) - clzsi(a) change is correct, but why is the ctz change?

d.s.low and d.s.high are su_int. While some helpers from libgcc are documented with int, long, etc. types for simplicity and use machine modes under the hood, the __builtin_ctz() function (as well as clz), on the other hand, seems to really accept an int-sized argument:

$ clang --version
clang version 10.0.0-4ubuntu1 
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
$ clang -S -O3 -o- -x c - <<< "int f() { return __builtin_ctz(0x100000); }"
        .text
        .file   "-"
        .globl  f                       # -- Begin function f
        .p2align        4, 0x90
        .type   f,@function
f:                                      # @f
        .cfi_startproc
# %bb.0:
        movl    $20, %eax
        retq
.Lfunc_end0:
        .size   f, .Lfunc_end0-f
        .cfi_endproc
                                        # -- End function
        .ident  "clang version 10.0.0-4ubuntu1 "
        .section        ".note.GNU-stack","",@progbits
        .addrsig
$ clang -S -O3 -target msp430 -o- -x c - <<< "int f() { return __builtin_ctz(0x100000); }"
<stdin>:1:32: warning: implicit conversion from 'long' to 'unsigned int' changes value from 1048576 to 0 [-Wconstant-conversion]
int f() { return __builtin_ctz(0x100000); }
                 ~~~~~~~~~~~~~ ^~~~~~~~
        .text
        .file   "-"
        .globl  f                       ; -- Begin function f
        .p2align        1
        .type   f,@function
f:                                      ; @f
; %bb.0:
        ret
.Lfunc_end0:
        .size   f, .Lfunc_end0-f
                                        ; -- End function
        .ident  "clang version 10.0.0-4ubuntu1 "
        .section        ".note.GNU-stack","",@progbits
        .addrsig
1 warning generated.

[aykevl]

I checked this, and it seems good to me although I would like someone else to also take a look.

In D86547#2287353, @atrosinenko wrote:

In D86547#2284095, @MaskRay wrote:

The (aWidth - 1) - clzsi(a) change is correct, but why is the ctz change?

d.s.low and d.s.high are su_int. While some helpers from libgcc are documented with int, long, etc. types for simplicity and use machine modes under the hood, the __builtin_ctz() function (as well as clz), on the other hand, seems to really accept an int-sized argument:

I can confirm this. This is why I introduced the clzsi and ctzsi macros in D78662. They are not intended for use with plain int types, but for use with si_int / su_int types (that are always 32-bit, unlike int).

[aykevl]

Looks good to me.

compiler-rt/lib/builtins/fp_extend.h

32–39

Perhaps more reliable would be the following:

#if ULONG_MAX == 0xFFFFFFFFFFFFFFFF
  return __builtin_clzl(a);
#elif ULLONG_MAX == 0xFFFFFFFFFFFFFFFF
  return __builtin_clzll(a);
#else
  #error Unsupported platform
#endif

This is what I've also used in int_types.h to detect clzsi etc. It probably would need to be tested on some more architectures though (32 and 64 bit).

[aykevl]

compiler-rt/lib/builtins/fp_extend.h

32–39

In fact, it might be possible to use __builtin_clzll for all platforms (it maps to 'long long', which I think is 64-bit practically everywhere).

compiler-rt/lib/builtins/floatsisf.c

COMPILER_RT_ABI fp_t __floatsisf(si_int a) {
   // All other cases begin by extracting the sign and absolute value of a
   rep_t sign = 0;
   if (a < 0) {
     sign = signBit;
     a = -a;
   }
 
   // Exponent of (fp_t)a is the width of abs(a).
-  const int exponent = (aWidth - 1) - __builtin_clz(a);
+  const int exponent = (aWidth - 1) - clzsi(a);
   rep_t result;
 
   // Shift a into the significand field, rounding if it is a right-shift
   if (exponent <= significandBits) {
     const int shift = significandBits - exponent;
     result = (rep_t)a << shift ^ implicitBit;
   } else {
     const int shift = exponent - significandBits;

compiler-rt/lib/builtins/floatsitf.c

COMPILER_RT_ABI fp_t __floatsitf(si_int a) {
   rep_t sign = 0;
   su_int aAbs = (su_int)a;
   if (a < 0) {
     sign = signBit;
     aAbs = ~(su_int)a + (su_int)1U;
   }
 
   // Exponent of (fp_t)a is the width of abs(a).
-  const int exponent = (aWidth - 1) - __builtin_clz(aAbs);
+  const int exponent = (aWidth - 1) - clzsi(aAbs);
   rep_t result;
 
   // Shift a into the significand field and clear the implicit bit.
   const int shift = significandBits - exponent;
   result = (rep_t)aAbs << shift ^ implicitBit;
 
   // Insert the exponent
   result += (rep_t)(exponent + exponentBias) << significandBits;
   // Insert the sign bit and return
   return fromRep(result | sign);
 }
 
 #endif

compiler-rt/lib/builtins/floatunsisf.c

COMPILER_RT_ABI fp_t __floatunsisf(su_int a) {
 
   const int aWidth = sizeof a * CHAR_BIT;
 
   // Handle zero as a special case to protect clz
   if (a == 0)
     return fromRep(0);
 
   // Exponent of (fp_t)a is the width of abs(a).
-  const int exponent = (aWidth - 1) - __builtin_clz(a);
+  const int exponent = (aWidth - 1) - clzsi(a);
   rep_t result;
 
   // Shift a into the significand field, rounding if it is a right-shift
   if (exponent <= significandBits) {
     const int shift = significandBits - exponent;
     result = (rep_t)a << shift ^ implicitBit;
   } else {
     const int shift = exponent - significandBits;

compiler-rt/lib/builtins/floatunsitf.c

COMPILER_RT_ABI fp_t __floatunsitf(su_int a) {
 
   const int aWidth = sizeof a * CHAR_BIT;
 
   // Handle zero as a special case to protect clz
   if (a == 0)
     return fromRep(0);
 
   // Exponent of (fp_t)a is the width of abs(a).
-  const int exponent = (aWidth - 1) - __builtin_clz(a);
+  const int exponent = (aWidth - 1) - clzsi(a);
   rep_t result;
 
   // Shift a into the significand field and clear the implicit bit.
   const int shift = significandBits - exponent;
   result = (rep_t)a << shift ^ implicitBit;
 
   // Insert the exponent
   result += (rep_t)(exponent + exponentBias) << significandBits;
   return fromRep(result);
 }
 
 #endif

compiler-rt/lib/builtins/fp_extend.h

 #define src_rep_t_clz clzsi
 
 #elif defined SRC_DOUBLE
 typedef double src_t;
 typedef uint64_t src_rep_t;
 #define SRC_REP_C UINT64_C
 static const int srcSigBits = 52;
 static __inline int src_rep_t_clz(src_rep_t a) {
 #if defined __LP64__
   return __builtin_clzl(a);
 #else
   if (a & REP_C(0xffffffff00000000))
-    return __builtin_clz(a >> 32);
+    return clzsi(a >> 32);
   else
-    return 32 + __builtin_clz(a & REP_C(0xffffffff));
+    return 32 + clzsi(a & REP_C(0xffffffff));
 #endif
 }
 
 #elif defined SRC_HALF
 #ifdef COMPILER_RT_HAS_FLOAT16
 typedef _Float16 src_t;
 #else
 typedef uint16_t src_t;
 #endif

compiler-rt/lib/builtins/udivmoddi4.c

if (d.s.low == 0) {
     // ---
     // K 0
     if ((d.s.high & (d.s.high - 1)) == 0) /* if d is a power of 2 */ {
       if (rem) {
         r.s.low = n.s.low;
         r.s.high = n.s.high & (d.s.high - 1);
         *rem = r.all;
       }
-      return n.s.high >> __builtin_ctz(d.s.high);
+      return n.s.high >> ctzsi(d.s.high);
     }
     // K K
     // ---
     // K 0
     sr = clzsi(d.s.high) - clzsi(n.s.high);
     // 0 <= sr <= n_uword_bits - 2 or sr large
     if (sr > n_uword_bits - 2) {
       if (rem)
if (d.s.high == 0) {
       // K X
       // ---
       // 0 K
       if ((d.s.low & (d.s.low - 1)) == 0) /* if d is a power of 2 */ {
         if (rem)
           *rem = n.s.low & (d.s.low - 1);
         if (d.s.low == 1)
           return n.all;
-        sr = __builtin_ctz(d.s.low);
+        sr = ctzsi(d.s.low);
         q.s.high = n.s.high >> sr;
         q.s.low = (n.s.high << (n_uword_bits - sr)) | (n.s.low >> sr);
         return q.all;
       }
       // K X
       // ---
       // 0 K
       sr = 1 + n_uword_bits + clzsi(d.s.low) - clzsi(n.s.high);