diff --git a/rt/loader32.asm b/rt/loader32.asm
index 13bdb8f..e4a98d7 100644
--- a/rt/loader32.asm
+++ b/rt/loader32.asm
@@ -2,6 +2,11 @@
 
 %include "rtld.inc"
 
+%ifndef HASH_END_TYP
+%warning "W: HASH_END_TYP not defined, falling back to 16-bit!"
+%define HASH_END_TYP word
+%endif
+
 %define R10_BIAS (0x178)
 
 %ifdef ELF_TYPE
@@ -186,7 +191,7 @@ _smol_start:
 %ifdef USE_JMP_BYTES
             inc edi ; skip 0xE9 (jmp) offset
 %endif
-            cmp word [edi], 0
+            cmp HASH_END_TYP [edi], 0
             jne short .next_hash
 
 ; if USE_DNLOAD_LOADER
@@ -300,7 +305,7 @@ repne scasd
 %ifdef USE_JMP_BYTES
                 inc edi
 %endif
-                cmp word [edi], 0
+                cmp HASH_END_TYP [edi], 0
                 jne short .next_hash
 
        pop eax ; get rid of leftover ecx on stack
diff --git a/rt/loader64.asm b/rt/loader64.asm
index d704685..336ef65 100644
--- a/rt/loader64.asm
+++ b/rt/loader64.asm
@@ -1,5 +1,10 @@
 ; vim: set ft=nasm:
 
+%ifndef HASH_END_TYP
+%warning "W: HASH_END_TYP not defined, falling back to 16-bit!"
+%define HASH_END_TYP word
+%endif
+
 ;%define R10_BIAS (0x2B4)
 %define R10_BIAS (0x2B4+0x40)
 
@@ -153,7 +158,7 @@ _smol_start:
 %endif
 %endif
           stosq
-            cmp word [rdi], 0
+            cmp HASH_END_TYP [rdi], 0
 %ifdef IFUNC_SUPPORT
 %ifdef SKIP_ZERO_VALUE
             jne .next_hash;short .next_hash
@@ -311,7 +316,7 @@ repne scasd ; technically, scasq should be used, but meh. this is 1 byte smaller
 ; IFUNC_SUPPORT
 %endif
           stosq ; *phash = finaladdr
-            cmp word [rdi], 0
+            cmp HASH_END_TYP [rdi], 0
             jne short .next_hash
             ; } while (1)
 ;       jmp short .next_hash
diff --git a/smol/emit.py b/smol/emit.py
index 55e7e40..56aae42 100644
--- a/smol/emit.py
+++ b/smol/emit.py
@@ -4,6 +4,23 @@
 
 from .shared import *
 
+fetch_width_from_bits = { 8: 'byte', 16: 'word', 32: 'dword', 64: 'qword' }
+
+def get_min_check_width(libraries, hashfn):
+    minv = 8 # can't go lower
+    for k, v in libraries.items():
+        for sym in v:
+            hv = hashfn(sym[0]) # sym == (name, reloc)
+            if (hv & 0xffffffff) == 0:
+                # this should (hopefully) NEVER happen
+                error("Aiee, all-zero hash for sym '%s'!" % sym)
+            elif (hv & 0xFFFF) == 0:
+                minv = max(minv, 32) # need at least 32 bits
+            elif (hv & 0xFF) == 0:
+                minv = max(minv, 16) # need at least 16 bits
+
+    return minv
+
 def sort_imports(libraries, hashfn):
     #eprintf("in: " + str(libraries))
 
@@ -29,6 +46,9 @@ def output_x86(libraries, nx, h16, outf, det):
     hashfn = hash_bsd2 if h16 else hash_djb2
     if det: libraries = sort_imports(libraries, hashfn)
 
+    outf.write('%%define HASH_END_TYP %s\n' %
+               fetch_width_from_bits[get_min_check_width(libraries, hashfn)])
+
     usedrelocs = set({})
     for library, symrels in libraries.items():
         for sym, reloc in symrels: usedrelocs.add(reloc)
@@ -74,7 +94,7 @@ def output_x86(libraries, nx, h16, outf, det):
         for sym, reloc in symrels:
             # meh
             if reloc != 'R_386_PC32' and reloc != 'R_386_GOT32X':
-                eprintf('Relocation type ' + reloc + ' of symbol ' + sym + ' unsupported!')
+                eprintf('Relocation type %s of symbol %s unsupported!' % (reloc, sym))
                 sys.exit(1)
 
             if nx:
@@ -118,6 +138,9 @@ def output_amd64(libraries, nx, h16, outf, det):
     hashfn = hash_djb2 #hash_bsd2 if h16 else hash_djb2
     if det: libraries = sort_imports(libraries, hashfn)
 
+    outf.write('%%define HASH_END_TYP %s\n' %
+               fetch_width_from_bits[get_min_check_width(libraries, hashfn)])
+
     outf.write('; vim: set ft=nasm:\n')
     outf.write('bits 64\n')
 
@@ -153,7 +176,7 @@ def output_amd64(libraries, nx, h16, outf, det):
         for sym, reloc in symrels:
             if reloc not in ['R_X86_64_PLT32', 'R_X86_64_GOTPCRELX', \
                              'R_X86_64_REX_GOTPCRELX', 'R_X86_64_GOTPCREL']:
-                error('Relocation type ' + reloc + ' of symbol ' + sym + ' unsupported!')
+                error('Relocation type %s of symbol %s unsupported!' % (reloc, sym))
 
             if reloc in ['R_X86_64_GOTPCRELX', 'R_X86_64_REX_GOTPCRELX', \
                          'R_X86_64_GOTPCREL']:
@@ -189,5 +212,5 @@ def output(arch, libraries, nx, h16, outf, det):
     if arch == 'i386': output_x86(libraries, nx, h16, outf, det)
     elif arch == 'x86_64': output_amd64(libraries, nx, h16, outf, det)
     else:
-        error("E: cannot emit for arch '" + str(arch) + "'")
+        error("E: cannot emit for arch '%s'" % str(arch))