a bunch of vfor

14 files changed, 86 insertions, 82 deletions

diff --git a/build/src/build.bqn b/build/src/build.bqn
index 7f3d4c3c..cd2471c4 100755
--- a/build/src/build.bqn
+++ b/build/src/build.bqn
@@ -311,6 +311,7 @@ po ← { # parsed options
       ⟨"-Wno-parentheses"⟩ # gcc
       ⟨ # clang
         "-Wno-microsoft-anon-tag", "-Wno-bitwise-instead-of-logical", "-Wno-unknown-warning-option"
+        "-Wno-pass-failed" # vfor
       ⟩ ∾ (¬wasm)/⟨"-mllvm", "--x86-cmov-converter=0"⟩ # thing that converts `cmov`s to branching sometimes (?), but we don't want that
     ⟩
     args∾↩ 0⊑GetLibs@
diff --git a/src/builtins/arithd.c b/src/builtins/arithd.c
index d2a756ae..83a11a19 100644
--- a/src/builtins/arithd.c
+++ b/src/builtins/arithd.c
@@ -29,7 +29,7 @@ B leading_axis_arith(FC2 fc2, B w, B x, usz* wsh, usz* xsh, ur mr);
 #else
   static void base_andBytes(u8* r, u8* x, u64 repeatedMask, u64 numBytes) {
     u64* x64 = (u64*)x; usz i;
-    for (i = 0; i < numBytes/8; i++) ((u64*)r)[i] = x64[i] & repeatedMask;
+    vfor (i = 0; i < numBytes/8; i++) ((u64*)r)[i] = x64[i] & repeatedMask;
     if (i*8 != numBytes) {
       u64 v = x64[i]&repeatedMask;
       for (usz j = 0; j < (numBytes&7); j++) r[i*8 + j] = v>>(j*8);
@@ -47,7 +47,7 @@ B shape_c2(B t, B w, B x);
 // floordiv will return float result only on ¯2147483648÷¯1 or n÷0, but may not otherwise squeeze integer types; integer argument requirement may be relaxed in the future
 // divint will return float result if there's a fractional result, or in overflow cases same as floordiv
 // TODO overflow-checked Singeli code for exact integer divint, and maybe floordiv_AA
-#define DIVLOOP(RE, WE, EXPR) RE* rp; B r=m_##RE##arrc(&rp, w); usz ia=IA(w); WE* wp=WE##any_ptr(w); for(ux i=0; i<ia; i++) rp[i] = (EXPR);
+#define DIVLOOP(RE, WE, EXPR) RE* rp; B r=m_##RE##arrc(&rp, w); usz ia=IA(w); WE* wp=WE##any_ptr(w); vfor(ux i=0; i<ia; i++) rp[i] = (EXPR);
 static B divint_AA(B w, B x) { // consumes both
   w = toI32Any(w);
   x = toI32Any(x); i32* xp = tyany_ptr(x);
@@ -133,7 +133,7 @@ static B modint_AS(B w,   B xv) { return modint_AA(w, C2(shape, C1(fne, incG(w))
   #define Ri16(A) i16* rp; r=m_i16arrc(&rp, A);
   #define Ri32(A) i32* rp; r=m_i32arrc(&rp, A);
   #define Rf64(A) f64* rp; r=m_f64arrc(&rp, A);
-  #define DOF(EXPR,A,W,X) { for (usz i = 0; i < ia; i++) { f64 wv=W; f64 xv=X; rp[i]=EXPR; } }
+  #define DOF(EXPR,A,W,X) { vfor (usz i = 0; i < ia; i++) { f64 wv=W; f64 xv=X; rp[i]=EXPR; } }
   #define DOI8(EXPR,A,W,X,BASE)  { Ri8(A)  for (usz i=0; i<ia; i++) { i16 wv=W; i16 xv=X; i16 rv=EXPR; if (RARE(rv!=( i8)rv)) { decG(r); goto BASE; } rp[i]=rv; } goto dec_ret; }
   #define DOI16(EXPR,A,W,X,BASE) { Ri16(A) for (usz i=0; i<ia; i++) { i32 wv=W; i32 xv=X; i32 rv=EXPR; if (RARE(rv!=(i16)rv)) { decG(r); goto BASE; } rp[i]=rv; } goto dec_ret; }
   #define DOI32(EXPR,A,W,X,BASE) { Ri32(A) for (usz i=0; i<ia; i++) { i64 wv=W; i64 xv=X; i64 rv=EXPR; if (RARE(rv!=(i32)rv)) { decG(r); goto BASE; } rp[i]=rv; } goto dec_ret; }
@@ -151,20 +151,20 @@ static B modint_AS(B w,   B xv) { return modint_AA(w, C2(shape, C1(fne, incG(w))
           if (xe<el_i32) { x=taga(cpyI32Arr(x)); xe=el_i32; } void* xp=tyany_ptr(x); \
           Rf64(x);                                                \
           if (we==el_i32) { B w,x /*shadow*/;                     \
-            if (xe==el_i32) { DECOR for (usz i = 0; i < ia; i++) { w.f=((i32*)wp)[i]; x.f=((i32*)xp)[i]; rp[i]=EXPR; } } \
-            else            { DECOR for (usz i = 0; i < ia; i++) { w.f=((i32*)wp)[i]; x.f=((f64*)xp)[i]; rp[i]=EXPR; } } \
+            if (xe==el_i32) { DECOR vfor (usz i = 0; i < ia; i++) { w.f=((i32*)wp)[i]; x.f=((i32*)xp)[i]; rp[i]=EXPR; } } \
+            else            { DECOR vfor (usz i = 0; i < ia; i++) { w.f=((i32*)wp)[i]; x.f=((f64*)xp)[i]; rp[i]=EXPR; } } \
           } else {          B w,x /*shadow*/;                     \
-            if (xe==el_i32) { DECOR for (usz i = 0; i < ia; i++) { w.f=((f64*)wp)[i]; x.f=((i32*)xp)[i]; rp[i]=EXPR; } } \
-            else            { DECOR for (usz i = 0; i < ia; i++) { w.f=((f64*)wp)[i]; x.f=((f64*)xp)[i]; rp[i]=EXPR; } } \
+            if (xe==el_i32) { DECOR vfor (usz i = 0; i < ia; i++) { w.f=((f64*)wp)[i]; x.f=((i32*)xp)[i]; rp[i]=EXPR; } } \
+            else            { DECOR vfor (usz i = 0; i < ia; i++) { w.f=((f64*)wp)[i]; x.f=((f64*)xp)[i]; rp[i]=EXPR; } } \
           }                                                       \
           decG(w); decG(x); return num_squeeze(r);                \
         }                                                         \
       } else if (isF64(w)&isArr(x)) { usz ia=IA(x); u8 xe=TI(x,elType); \
-        if (elInt(xe)) {INT_SA Rf64(x); x=toI32Any(x); PI32(x) DECOR for (usz i=0; i<ia; i++) {B x/*shadow*/;x.f=xp[i];rp[i]=EXPR;} decG(x); return num_squeeze(r); } \
-        if (xe==el_f64){       Rf64(x);         PF(x)  FLT_SAI DECOR for (usz i=0; i<ia; i++) {B x/*shadow*/;x.f=xp[i];rp[i]=EXPR;} decG(x); return num_squeeze(r); } \
+        if (elInt(xe)) {INT_SA Rf64(x); x=toI32Any(x); PI32(x) DECOR vfor (usz i=0; i<ia; i++) {B x/*shadow*/;x.f=xp[i];rp[i]=EXPR;} decG(x); return num_squeeze(r); } \
+        if (xe==el_f64){       Rf64(x);         PF(x)  FLT_SAI DECOR vfor (usz i=0; i<ia; i++) {B x/*shadow*/;x.f=xp[i];rp[i]=EXPR;} decG(x); return num_squeeze(r); } \
       } else if (isF64(x)&isArr(w)) { usz ia=IA(w); u8 we=TI(w,elType); ANY_AS \
-        if (elInt(we)) {INT_AS Rf64(w); w=toI32Any(w); PI32(w) DECOR for (usz i=0; i<ia; i++) {B w/*shadow*/;w.f=wp[i];rp[i]=EXPR;} decG(w); return num_squeeze(r); } \
-        if (we==el_f64){       Rf64(w);         PF(w)          DECOR for (usz i=0; i<ia; i++) {B w/*shadow*/;w.f=wp[i];rp[i]=EXPR;} decG(w); return num_squeeze(r); } \
+        if (elInt(we)) {INT_AS Rf64(w); w=toI32Any(w); PI32(w) DECOR vfor (usz i=0; i<ia; i++) {B w/*shadow*/;w.f=wp[i];rp[i]=EXPR;} decG(w); return num_squeeze(r); } \
+        if (we==el_f64){       Rf64(w);         PF(w)          DECOR vfor (usz i=0; i<ia; i++) {B w/*shadow*/;w.f=wp[i];rp[i]=EXPR;} decG(w); return num_squeeze(r); } \
       }                                                           \
       P2(NAME)                                                    \
     }                                                             \
@@ -214,7 +214,7 @@ static B modint_AS(B w,   B xv) { return modint_AA(w, C2(shape, C1(fne, incG(w))
     }
     , /*INT_AS*/ if (q_i32(x)) return modint_AS(w, x);
     , /*INT_AA*/ return modint_AA(w, x);
-    , /*FLT_SAI*/ if (o2fG(w)==1) { for (usz i=0; i<ia; i++) rp[i] = xp[i]-floor(xp[i]); } else
+    , /*FLT_SAI*/ if (o2fG(w)==1) { vfor (usz i=0; i<ia; i++) rp[i] = xp[i]-floor(xp[i]); } else
     , /*ANY_AS*/
   )
   #undef GC2f
@@ -271,13 +271,13 @@ static B modint_AS(B w,   B xv) { return modint_AA(w, C2(shape, C1(fne, incG(w))
     static NOINLINE B bitAA1(B w, B x, usz ia) {
       u64* rp; B r = m_bitarrc(&rp, x);
       u64* wp=bitarr_ptr(w); u64* xp=bitarr_ptr(x);
-      for (usz i=0; i<BIT_N(ia); i++) rp[i] = wp[i]|xp[i];
+      vfor (usz i=0; i<BIT_N(ia); i++) rp[i] = wp[i]|xp[i];
       decG(w); decG(x); return r;
     }
     static NOINLINE B bitAA2(B w, B x, usz ia) {
       u64* rp; B r = m_bitarrc(&rp, x);
       u64* wp=bitarr_ptr(w); u64* xp=bitarr_ptr(x);
-      for (usz i=0; i<BIT_N(ia); i++) rp[i] = wp[i]&xp[i];
+      vfor (usz i=0; i<BIT_N(ia); i++) rp[i] = wp[i]&xp[i];
       decG(w); decG(x); return r;
     }
     
@@ -383,7 +383,7 @@ static B modint_AS(B w,   B xv) { return modint_AA(w, C2(shape, C1(fne, incG(w))
         if (we!=el_c32) w = taga(cpyC32Arr(w));
         u32* wp = c32any_ptr(w); usz wia = IA(w);
         i32* rp; r = m_i32arrc(&rp, w);
-        for (usz i = 0; i < wia; i++) rp[i] = (i32)wp[i] - xv;
+        vfor (usz i = 0; i < wia; i++) rp[i] = (i32)wp[i] - xv;
         goto dec_ret;
       }
     })
diff --git a/src/builtins/arithm.c b/src/builtins/arithm.c
index fdce468e..d2e9e162 100644
--- a/src/builtins/arithm.c
+++ b/src/builtins/arithm.c
@@ -11,7 +11,7 @@ static inline B arith_recm(FC1 f, B x) {
 }
 
 void bit_negatePtr(u64* rp, u64* xp, usz count) {
-  for (usz i = 0; i < count; i++) rp[i] = ~xp[i];
+  vfor (usz i = 0; i < count; i++) rp[i] = ~xp[i];
 }
 B bit_negate(B x) { // consumes
   u64* xp = bitarr_ptr(x);
@@ -48,10 +48,10 @@ B add_c1(B t, B x) {
   i64 ia = IA(x); INIT;               \
   void* xp = tyany_ptr(x);            \
   switch(xe) { default: UD;           \
-    case el_i8:  for(usz i=0; i<ia; i++) { i8  c = ((i8* )xp)[i]; EXPR(i8,  c==I8_MIN)  } break; \
-    case el_i16: for(usz i=0; i<ia; i++) { i16 c = ((i16*)xp)[i]; EXPR(i16, c==I16_MIN) } break; \
-    case el_i32: for(usz i=0; i<ia; i++) { i32 c = ((i32*)xp)[i]; EXPR(i32, c==I32_MIN) } break; \
-    case el_f64: for(usz i=0; i<ia; i++) { f64 c = ((f64*)xp)[i]; EXPR(f64, 0)          } break; \
+    case el_i8:   for(usz i=0; i<ia; i++) { i8  c = ((i8* )xp)[i]; EXPR(i8,  c==I8_MIN)  } break; \
+    case el_i16:  for(usz i=0; i<ia; i++) { i16 c = ((i16*)xp)[i]; EXPR(i16, c==I16_MIN) } break; \
+    case el_i32:  for(usz i=0; i<ia; i++) { i32 c = ((i32*)xp)[i]; EXPR(i32, c==I32_MIN) } break; \
+    case el_f64: vfor(usz i=0; i<ia; i++) { f64 c = ((f64*)xp)[i]; EXPR(f64, 0)          } break; \
   } \
   decG(x); return r; POST \
 }
@@ -75,11 +75,11 @@ B add_c1(B t, B x) {
   #define STILE_BODY(FEXPR) LOOP_BODY(B r; void* rp = m_tyarrlc(&r, elWidth(xe), x, el2t(xe));, STILE_EXPR, bad: tyarr_freeF(v(r));)
 #endif
 
-#define FLOAT_BODY(FEXPR) { i64 ia = IA(x);                  \
-  assert(xe==el_f64); f64* xp = f64any_ptr(x);               \
-  f64* rp; B r = m_f64arrc(&rp, x);                          \
-  for (usz i = 0; i < ia; i++) { f64 v=xp[i]; rp[i]=FEXPR; } \
-  decG(x); return num_squeeze(r);                            \
+#define FLOAT_BODY(FEXPR) { i64 ia = IA(x);                   \
+  assert(xe==el_f64); f64* xp = f64any_ptr(x);                \
+  f64* rp; B r = m_f64arrc(&rp, x);                           \
+  vfor (usz i = 0; i < ia; i++) { f64 v=xp[i]; rp[i]=FEXPR; } \
+  decG(x); return num_squeeze(r);                             \
 }
 B sub_c2(B,B,B);
 #define SUB_BODY(FEXPR) return sub_c2(t, m_f64(0), x);
@@ -101,7 +101,7 @@ GC1i("¬", not,    1-v,             el_bit, bit_negate(x), NOT_BODY)
       u64 ia = IA(x);                       \
       f64* xp = f64any_ptr(x);              \
       f64* rp; B r = m_f64arrc(&rp, x);     \
-      for (i64 i = 0; i < ia; i++) {        \
+      vfor (i64 i = 0; i < ia; i++) {       \
         f64 xv=xp[i]; rp[i] = (F);          \
       }                                     \
       decG(x); return r;                    \
diff --git a/src/builtins/cmp.c b/src/builtins/cmp.c
index 3fa9c370..bd8aa0dd 100644
--- a/src/builtins/cmp.c
+++ b/src/builtins/cmp.c
@@ -69,7 +69,7 @@ CMP_REC(ne, ne, swapped=0;)
   #define CMP_AA0(N, T, BODY) void base_##N##AA##_##T(u64* r, void* w, void* x, u64 l) { BODY }
   #define CMP_AA1(N, T, OP) CMP_AA0(N, T, BASE_CMP_LOOP(OP, ((T*)w)[i], ((T*)x)[i]))
   #define CMP_AA_F(N, OP, BX) \
-    CMP_AA0(N, u1, ({usz bia = BIT_N(l); for (usz i=0; i<bia; i++) { u64 wv=((u64*)w)[i], xv=((u64*)x)[i]; ((u64*)r)[i] = BX; }});) \
+    CMP_AA0(N, u1, ({usz bia = BIT_N(l); vfor (usz i=0; i<bia; i++) { u64 wv=((u64*)w)[i], xv=((u64*)x)[i]; ((u64*)r)[i] = BX; }});) \
     CMP_AA1(N, i8, OP) CMP_AA1(N, i16, OP) CMP_AA1(N, i32, OP) CMP_AA1(N, f64, OP) \
     const CmpAAFn base_##N##AA##_u32 = base_##N##AA##_i32;
   
@@ -104,7 +104,7 @@ CMP_REC(ne, ne, swapped=0;)
   
   #define CMP_SA1(N, T, Q, C, SLOW, OP) CMP_SA0(N, T, Q, SLOW, ({ T xv = C(x); BASE_CMP_LOOP(OP, ((T*)w)[i], xv) }))
   #define CMP_SA_F(N, OP, SLOW, BX) \
-    CMP_SA0(N, u1, bit, SLOW, ({usz bia = BIT_N(l); u64 xv=bitx(x); for (usz i=0; i<bia; i++) { u64 wv=((u64*)w)[i]; ((u64*)r)[i] = BX; }})) \
+    CMP_SA0(N, u1, bit, SLOW, ({usz bia = BIT_N(l); u64 xv=bitx(x); vfor (usz i=0; i<bia; i++) { u64 wv=((u64*)w)[i]; ((u64*)r)[i] = BX; }})) \
     CMP_SA1(N,i8,i8,o2iG,SLOW,OP) CMP_SA1(N,i16,i16,o2iG,SLOW,OP) CMP_SA1(N,i32,i32,o2iG,SLOW,OP) CMP_SA1(N,f64,f64,o2fG,SLOW,OP) \
     CMP_SA1(N,u8,c8,o2cG,SLOW,OP) CMP_SA1(N,u16,c16,o2cG,SLOW,OP) CMP_SA1(N,u32,c32,o2cG,SLOW,OP)
   
diff --git a/src/builtins/grade.h b/src/builtins/grade.h
index 80d3555d..7890d80d 100644
--- a/src/builtins/grade.h
+++ b/src/builtins/grade.h
@@ -319,12 +319,12 @@ B GRADE_CAT(c1)(B t, B x) {
     }
     
     TALLOC(I32I32p, tmp, ia);
-    for (usz i = 0; i < ia; i++) {
+    vfor (usz i = 0; i < ia; i++) {
       tmp[i].v = i;
       tmp[i].k = xp[i];
     }
     CAT(GRADE_CAT(IP),tim_sort)(tmp, ia);
-    for (usz i = 0; i < ia; i++) rp[i] = tmp[i].v;
+    vfor (usz i = 0; i < ia; i++) rp[i] = tmp[i].v;
     TFREE(tmp);
     goto decG_sq;
   }
diff --git a/src/builtins/search.c b/src/builtins/search.c
index 8ce7d8d1..49221225 100644
--- a/src/builtins/search.c
+++ b/src/builtins/search.c
@@ -574,7 +574,7 @@ B asNormalized(B x, usz n, bool nanBad) {
       }
     #endif
   } else {
-    for (; i < n; i++) rp[i] = normalizeFloat(fp[i]);
+    vfor (; i < n; i++) rp[i] = normalizeFloat(fp[i]);
   }
   
   if (r.u!=x.u) decG(x);
diff --git a/src/builtins/sfns.c b/src/builtins/sfns.c
index 1fa22c73..2d6e7e07 100644
--- a/src/builtins/sfns.c
+++ b/src/builtins/sfns.c
@@ -1066,18 +1066,18 @@ B reverse_c1(B t, B x) {
       case 0: {
         u64* rp; r = m_bitarrc(&rp, x);
         u64* xp=xv; usz g = BIT_N(n); usz e = g-1;
-        for (usz i = 0; i < g; i++) rp[i] = bit_reverse(xp[e-i]);
+        vfor (usz i = 0; i < g; i++) rp[i] = bit_reverse(xp[e-i]);
         if (n&63) {
           u64 sh=(-n)&63;
-          for (usz i=0; i<e; i++) rp[i]=rp[i]>>sh|rp[i+1]<<(64-sh);
+          vfor (usz i=0; i<e; i++) rp[i]=rp[i]>>sh|rp[i+1]<<(64-sh);
           rp[e]>>=sh;
         }
         break;
       }
-      case 3:                         { u8*  xp=xv; u8*  rp = m_tyarrc(&r, 1, x, xt); for (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
-      case 4:                         { u16* xp=xv; u16* rp = m_tyarrc(&r, 2, x, xt); for (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
-      case 5:                         { u32* xp=xv; u32* rp = m_tyarrc(&r, 4, x, xt); for (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
-      case 6: if (TI(x,elType)!=el_B) { u64* xp=xv; u64* rp = m_tyarrc(&r, 8, x, xt); for (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
+      case 3:                         { u8*  xp=xv; u8*  rp = m_tyarrc(&r, 1, x, xt); vfor (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
+      case 4:                         { u16* xp=xv; u16* rp = m_tyarrc(&r, 2, x, xt); vfor (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
+      case 5:                         { u32* xp=xv; u32* rp = m_tyarrc(&r, 4, x, xt); vfor (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
+      case 6: if (TI(x,elType)!=el_B) { u64* xp=xv; u64* rp = m_tyarrc(&r, 8, x, xt); vfor (ux i=0; i<n; i++) rp[i]=xp[n-i-1]; break; }
       else {
         HArr_p rp = m_harrUc(x);
         B* xp = arr_bptr(x);
diff --git a/src/builtins/slash.c b/src/builtins/slash.c
index 310b6f51..b2555a37 100644
--- a/src/builtins/slash.c
+++ b/src/builtins/slash.c
@@ -299,7 +299,7 @@ static B where(B x, usz xia, u64 s) {
       usz bs;
       if (b>xia-i) { b=xia-i; bs=s-(rp-rp0); } else { bs=bit_sum(xp,b); }
       where_block_u16(xp, buf, b, bs);
-      for (usz j=0; j<bs; j++) rp[j] = i+buf[j];
+      vfor (usz j=0; j<bs; j++) rp[j] = i+buf[j];
       rp+= bs;
       xp+= b/64;
     }
@@ -380,7 +380,7 @@ B grade_bool(B x, usz xia, bool up) {
     u64 xp0[4]; // 4 ≡ b/64
     u64* xp1 = xp;
     for (usz i=0; i<xia; i+=b) {
-      for (usz j=0; j<BIT_N(b); j++) xp0[j] = ~xp1[j];
+      vfor (usz j=0; j<BIT_N(b); j++) xp0[j] = ~xp1[j];
       usz b2 = b>xia-i? xia-i : b;
       if (b2<b) { u64 q=b2%64; usz e=b2/64; u64 m=((u64)1<<q)-1; xp0[e]&=m; xp1[e]&=m; }
       usz s0=bit_sum(xp0,b2); si_1slash32(xp0, i, rp0, b2, s0); rp0+=s0;
@@ -888,7 +888,7 @@ B slash_im(B t, B x) {
         for (usz i=0; i<xia; i++) t[(u##N)xp[i]]++;                              \
         t[m/2]=xia; usz ria=0; for (u64 s=0; s<xia; ria++) s+=t[ria];            \
         if (ria>m/2) thrM("/⁼: Argument cannot contain negative numbers");       \
-        i32* rp; r = m_i32arrv(&rp, ria); for (usz i=0; i<ria; i++) rp[i]=t[i];  \
+        i32* rp; r = m_i32arrv(&rp, ria); vfor (usz i=0; i<ria; i++) rp[i]=t[i]; \
         TFREE(t);                                                                \
         r = num_squeeze(r);                                                      \
       }                                                                          \
@@ -901,7 +901,7 @@ B slash_im(B t, B x) {
       i8 max = avx2_count_i8(t, (i8*)xp, xia, 0);                              \
       if (max < 0) thrM("/⁼: Argument cannot contain negative numbers");       \
       usz ria=max+1;                                                           \
-      i32* rp; r = m_i32arrv(&rp, ria); for (usz i=0; i<ria; i++) rp[i]=t[i];  \
+      i32* rp; r = m_i32arrv(&rp, ria); vfor (usz i=0; i<ria; i++) rp[i]=t[i]; \
       TFREE(t);                                                                \
       r = num_squeeze(r);                                                      \
     } else
diff --git a/src/builtins/sort.c b/src/builtins/sort.c
index e803ba21..8492aac7 100644
--- a/src/builtins/sort.c
+++ b/src/builtins/sort.c
@@ -21,7 +21,7 @@ static NOINLINE void generic_grade(B x, usz ia, B r, i32* rp, void (*fn)(BI32p*,
     tmp[i].k = GetU(x,i);
   }
   fn(tmp, ia);
-  for (usz i = 0; i < ia; i++) rp[i] = tmp[i].v;
+  vfor (usz i = 0; i < ia; i++) rp[i] = tmp[i].v;
   TFREE(tmp);
 }
 
diff --git a/src/builtins/sysfn.c b/src/builtins/sysfn.c
index 81e9675e..d585773a 100644
--- a/src/builtins/sysfn.c
+++ b/src/builtins/sysfn.c
@@ -1566,7 +1566,7 @@ B bitop1(B f, B x, enum BitOp1 op, char* name) {
     case op_neg: switch (ow) {
       default: thrF("•bit._%U: unhandled width %s", name, ow);
       #define CASE(W) case W: \
-        NOUNROLL for (usz i=0; i<n/W; i++) ((u##W*)rp)[i] = -((u##W*)xp)[i]; \
+        NOUNROLL vfor (usz i=0; i<n/W; i++) ((u##W*)rp)[i] = -((u##W*)xp)[i]; \
         break;
       CASE(8) CASE(16) CASE(32) CASE(64)
       #undef CASE
@@ -1643,11 +1643,11 @@ B bitop2(B f, B w, B x, enum BitOp2 op, char* name) {
     }
   if (noextend) {
     #define BINOP(O,P) case op_##O: { \
-      usz l = n/64; NOUNROLL for (usz i=0; i<l; i++) rp[i] = wp[i] P xp[i];      \
+      usz l = n/64; NOUNROLL vfor (usz i=0; i<l; i++) rp[i] = wp[i] P xp[i];     \
       usz q = (-n)%64; if (q) rp[l] ^= (~(u64)0 >> q) & (rp[l]^(wp[l] P xp[l])); \
       } break;
     #define CASE(W, Q, P) case W: \
-      NOUNROLL for (usz i=0; i<n/W; i++)                  \
+      NOUNROLL vfor (usz i=0; i<n/W; i++)                 \
         ((Q##W*)rp)[i] = ((Q##W*)wp)[i] P ((Q##W*)xp)[i]; \
       break;
     SWITCH
@@ -1659,12 +1659,12 @@ B bitop2(B f, B w, B x, enum BitOp2 op, char* name) {
       if (ow>64) thrF("•bit._%U: scalar extension with width over 64 unhandled", name); \
       u64 wv = *wp & (~(u64)0>>(64-ow));                                      \
       for (usz tw=ow; tw<64; tw*=2) wv|=wv<<tw;                               \
-      usz l = n/64; NOUNROLL for (usz i=0; i<l; i++) rp[i] = wv P xp[i];      \
+      usz l = n/64; NOUNROLL vfor (usz i=0; i<l; i++) rp[i] = wv P xp[i];     \
       usz q = (-n)%64; if (q) rp[l] ^= (~(u64)0 >> q) & (rp[l]^(wv P xp[l])); \
       } break;
     #define CASE(W, Q, P) case W: { \
       Q##W wv = *(Q##W*)wp;                   \
-      NOUNROLL for (usz i=0; i<n/W; i++)      \
+      NOUNROLL vfor (usz i=0; i<n/W; i++)     \
         ((Q##W*)rp)[i] = wv P ((Q##W*)xp)[i]; \
       } break;
     SWITCH
diff --git a/src/core/harr.c b/src/core/harr.c
index 9db881da..1b00a10a 100644
--- a/src/core/harr.c
+++ b/src/core/harr.c
@@ -4,7 +4,7 @@
 
 NOINLINE B m_caB(usz ia, B* a) {
   HArr_p r = m_harrUv(ia);
-  for (usz i = 0; i < ia; i++) r.a[i] = a[i];
+  vfor (usz i = 0; i < ia; i++) r.a[i] = a[i];
   NOGC_E;
   return r.b;
 }
diff --git a/src/ffi.c b/src/ffi.c
index 041d499b..7e6fcf2f 100644
--- a/src/ffi.c
+++ b/src/ffi.c
@@ -78,7 +78,7 @@ BQN_EXP size_t bqn_rank(BQNV a) { return RNK(getB(a)); }
 BQN_EXP void bqn_shape(BQNV a, size_t* buf) { B b = getB(a);
   ur r = RNK(b);
   usz* sh = SH(b);
-  for (usz i = 0; i < r; i++) buf[i] = sh[i];
+  vfor (usz i = 0; i < r; i++) buf[i] = sh[i];
 }
 BQN_EXP BQNV bqn_pick(BQNV a, size_t pos) {
   return makeX(IGet(getB(a),pos));
@@ -590,7 +590,7 @@ FORCE_INLINE u64 i64abs(i64 x) { return x<0?-x:x; }
   usz ia = IA(x);                                 \
   B t = WIDEN(x); WEL* tp = WEL##any_ptr(t);      \
   REL* rp; B r = m_##REL##arrv(&rp, ia);          \
-  for (usz i=0; i<ia; i++) ((UEL*)rp)[i] = tp[i]; \
+  vfor (usz i=0; i<ia; i++) ((UEL*)rp)[i] = tp[i];\
   decG(t); return r;
 
 // copy elements of x to array of unsigned integers (using a signed integer array type as a "container"); consumes argument
@@ -603,7 +603,7 @@ NOINLINE B cpyF32Bits(B x) { // copy x to a 32-bit float array (using an i32arr
   usz ia = IA(x);
   B t = toF64Any(x); f64* tp = f64any_ptr(t);
   i32* rp; B r = m_i32arrv(&rp, ia);
-  for (usz i=0; i<ia; i++) ((f32*)rp)[i]=tp[i];
+  vfor (usz i=0; i<ia; i++) ((f32*)rp)[i]=tp[i];
   dec(t); return r;
 }
 
@@ -613,10 +613,10 @@ static B toU16Bits(B x) { return TI(x,elType)==el_i16? x : cpyU16Bits(x); }
 static B toU8Bits(B x)  { return TI(x,elType)==el_i8?  x : cpyU8Bits(x); }
 
 // read x as the specified type (assuming a container of the respective width signed integer array); consumes x
-NOINLINE B readU8Bits(B x)  { usz ia=IA(x); u8*  xp=tyarr_ptr(x); i16* rp; B r=m_i16arrv(&rp, ia); for (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
-NOINLINE B readU16Bits(B x) { usz ia=IA(x); u16* xp=tyarr_ptr(x); i32* rp; B r=m_i32arrv(&rp, ia); for (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
-NOINLINE B readU32Bits(B x) { usz ia=IA(x); u32* xp=tyarr_ptr(x); f64* rp; B r=m_f64arrv(&rp, ia); for (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
-NOINLINE B readF32Bits(B x) { usz ia=IA(x); f32* xp=tyarr_ptr(x); f64* rp; B r=m_f64arrv(&rp, ia); for (usz i=0; i<ia; i++) rp[i]=xp[i]; return r; }
+NOINLINE B readU8Bits(B x)  { usz ia=IA(x); u8*  xp=tyarr_ptr(x); i16* rp; B r=m_i16arrv(&rp, ia); vfor (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
+NOINLINE B readU16Bits(B x) { usz ia=IA(x); u16* xp=tyarr_ptr(x); i32* rp; B r=m_i32arrv(&rp, ia); vfor (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
+NOINLINE B readU32Bits(B x) { usz ia=IA(x); u32* xp=tyarr_ptr(x); f64* rp; B r=m_f64arrv(&rp, ia); vfor (usz i=0; i<ia; i++) rp[i]=xp[i]; return num_squeeze(r); }
+NOINLINE B readF32Bits(B x) { usz ia=IA(x); f32* xp=tyarr_ptr(x); f64* rp; B r=m_f64arrv(&rp, ia); vfor (usz i=0; i<ia; i++) rp[i]=xp[i]; return r; }
 B m_ptrobj_s(void* ptr, B o); // consumes o, sets stride to size of o
 B m_ptrobj(void* ptr, B o, ux stride); // consumes o
 static NOINLINE B ptrobj_checkget(B x); // doesn't consume
diff --git a/src/h.h b/src/h.h
index 2158d346..c975e660 100644
--- a/src/h.h
+++ b/src/h.h
@@ -123,13 +123,16 @@ typedef size_t ux;
 #if __clang__
   #define NOUNROLL _Pragma("clang loop unroll(disable)")
   #define NOVECTORIZE _Pragma("clang loop vectorize(disable)")
+  #define vfor _Pragma("clang loop vectorize(assume_safety)") for
 #elif __GNUC__
   #define EXACTLY_GCC 1
   #define NOUNROLL _Pragma("GCC unroll 1")
+  #define vfor _Pragma("GCC ivdep") for
   #define NOVECTORIZE
 #else
   #define NOUNROLL
   #define NOVECTORIZE
+  #define vfor for
 #endif
 #define PLAINLOOP NOUNROLL NOVECTORIZE
 #if EXACTLY_GCC
diff --git a/src/utils/mut.c b/src/utils/mut.c
index 4a538a08..1deea741 100644
--- a/src/utils/mut.c
+++ b/src/utils/mut.c
@@ -289,12 +289,12 @@ DEF_G(void, copy, B,             (void* a, usz ms, B x, usz xs, usz l), ms, x, x
   B* mpo = ms+(B*)a;
   switch(TY(x)) {
     case t_bitarr: { u64* xp = bitarr_ptr(x); for (usz i = 0; i < l; i++) mpo[i] = m_i32(bitp_get(xp, xs+i)); return; }
-    case t_i8arr:  case t_i8slice:  { i8*  xp = i8any_ptr (x); for (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
-    case t_i16arr: case t_i16slice: { i16* xp = i16any_ptr(x); for (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
-    case t_i32arr: case t_i32slice: { i32* xp = i32any_ptr(x); for (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
-    case t_c8arr:  case t_c8slice:  { u8*  xp = c8any_ptr (x); for (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
-    case t_c16arr: case t_c16slice: { u16* xp = c16any_ptr(x); for (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
-    case t_c32arr: case t_c32slice: { u32* xp = c32any_ptr(x); for (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
+    case t_i8arr:  case t_i8slice:  { i8*  xp = i8any_ptr (x); vfor (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
+    case t_i16arr: case t_i16slice: { i16* xp = i16any_ptr(x); vfor (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
+    case t_i32arr: case t_i32slice: { i32* xp = i32any_ptr(x); vfor (usz i = 0; i < l; i++) mpo[i] = m_i32(xp[i+xs]); return; }
+    case t_c8arr:  case t_c8slice:  { u8*  xp = c8any_ptr (x); vfor (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
+    case t_c16arr: case t_c16slice: { u16* xp = c16any_ptr(x); vfor (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
+    case t_c32arr: case t_c32slice: { u32* xp = c32any_ptr(x); vfor (usz i = 0; i < l; i++) mpo[i] = m_c32(xp[i+xs]); return; }
     case t_harr: case t_hslice: case t_fillarr: case t_fillslice:;
       B* xp = arr_bptr(x)+xs;
       for (usz i = 0; i < l; i++) inc(xp[i]);
@@ -393,15 +393,15 @@ DEF_G(void, copy, B,             (void* a, usz ms, B x, usz xs, usz l), ms, x, x
     E* rp; Arr* r = m_##E##arrp(&rp, ia); \
     arr_shCopy(r, x);      \
     u8 xe = TI(x,elType);  \
-    if      (xe==el_bit) { u64* xp = bitarr_ptr(x); for(usz i=0; i<ia; i++) rp[i]=bitp_get(xp,i); } \
-    else if (xe==el_i8 ) { i8*  xp = i8any_ptr (x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
-    else if (xe==el_i16) { i16* xp = i16any_ptr(x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
-    else if (xe==el_i32) { i32* xp = i32any_ptr(x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
-    else if (xe==el_f64) { f64* xp = f64any_ptr(x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    if      (xe==el_bit) { u64* xp = bitarr_ptr(x);  for(usz i=0; i<ia; i++) rp[i]=bitp_get(xp,i); } \
+    else if (xe==el_i8 ) { i8*  xp = i8any_ptr (x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    else if (xe==el_i16) { i16* xp = i16any_ptr(x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    else if (xe==el_i32) { i32* xp = i32any_ptr(x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    else if (xe==el_f64) { f64* xp = f64any_ptr(x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
     else {                 \
       B* xp = arr_bptr(x); \
-      if (xp!=NULL) { for (usz i=0; i<ia; i++) rp[i]=o2fG(xp[i]    ); } \
-      else { SGetU(x) for (usz i=0; i<ia; i++) rp[i]=o2fG(GetU(x,i)); } \
+      if (xp!=NULL) { vfor (usz i=0; i<ia; i++) rp[i]=o2fG(xp[i]    ); } \
+      else { SGetU(x)  for (usz i=0; i<ia; i++) rp[i]=o2fG(GetU(x,i)); } \
     }                      \
     ptr_decT(a(x));        \
     return r;              \
@@ -413,13 +413,13 @@ DEF_G(void, copy, B,             (void* a, usz ms, B x, usz xs, usz l), ms, x, x
     T##Atom* rp; Arr* r = m_##E##arrp(&rp, ia); \
     arr_shCopy(r, x);      \
     u8 xe = TI(x,elType);  \
-    if      (xe==el_c8 ) { u8*  xp = c8any_ptr (x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
-    else if (xe==el_c16) { u16* xp = c16any_ptr(x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
-    else if (xe==el_c32) { u32* xp = c32any_ptr(x); for(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    if      (xe==el_c8 ) { u8*  xp = c8any_ptr (x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    else if (xe==el_c16) { u16* xp = c16any_ptr(x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
+    else if (xe==el_c32) { u32* xp = c32any_ptr(x); vfor(usz i=0; i<ia; i++) rp[i]=xp[i]; } \
     else {                 \
       B* xp = arr_bptr(x); \
-      if (xp!=NULL) { for (usz i=0; i<ia; i++) rp[i]=o2cG(xp[i]    ); } \
-      else { SGetU(x) for (usz i=0; i<ia; i++) rp[i]=o2cG(GetU(x,i)); } \
+      if (xp!=NULL) { vfor (usz i=0; i<ia; i++) rp[i]=o2cG(xp[i]    ); } \
+      else { SGetU(x)  for (usz i=0; i<ia; i++) rp[i]=o2cG(GetU(x,i)); } \
     }                      \
     ptr_decT(a(x));        \
     return r;              \
@@ -429,14 +429,14 @@ DEF_G(void, copy, B,             (void* a, usz ms, B x, usz xs, usz l), ms, x, x
     usz ia = IA(x);
     HArr_p r = m_harrUc(x);
     u8 xe = TI(x,elType);
-    if      (xe==el_bit) { u64* xp = bitarr_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_f64(bitp_get(xp, i)); }
-    else if (xe==el_i8 ) { i8*  xp = i8any_ptr (x); for(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
-    else if (xe==el_i16) { i16* xp = i16any_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
-    else if (xe==el_i32) { i32* xp = i32any_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
-    else if (xe==el_f64) { f64* xp = f64any_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
-    else if (xe==el_c8 ) { u8*  xp = c8any_ptr (x); for(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
-    else if (xe==el_c16) { u16* xp = c16any_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
-    else if (xe==el_c32) { u32* xp = c32any_ptr(x); for(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
+    if      (xe==el_bit) { u64* xp = bitarr_ptr(x);  for(usz i=0; i<ia; i++) r.a[i]=m_f64(bitp_get(xp, i)); }
+    else if (xe==el_i8 ) { i8*  xp = i8any_ptr (x); vfor(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
+    else if (xe==el_i16) { i16* xp = i16any_ptr(x); vfor(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
+    else if (xe==el_i32) { i32* xp = i32any_ptr(x); vfor(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
+    else if (xe==el_f64) { f64* xp = f64any_ptr(x); vfor(usz i=0; i<ia; i++) r.a[i]=m_f64(xp[i]); }
+    else if (xe==el_c8 ) { u8*  xp = c8any_ptr (x); vfor(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
+    else if (xe==el_c16) { u16* xp = c16any_ptr(x); vfor(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
+    else if (xe==el_c32) { u32* xp = c32any_ptr(x); vfor(usz i=0; i<ia; i++) r.a[i]=m_c32(xp[i]); }
     else {
       B* xp = arr_bptr(x);
       if (xp!=NULL) { for (usz i=0; i<ia; i++) r.a[i] = inc(xp[i]); }
@@ -451,7 +451,7 @@ DEF_G(void, copy, B,             (void* a, usz ms, B x, usz xs, usz l), ms, x, x
     u64* rp; Arr* r = m_bitarrp(&rp, ia);
     arr_shCopy(r, x);
     u8 xe = TI(x,elType);
-    if      (xe==el_bit) { u64* xp = bitarr_ptr(x); for(usz i=0; i<BIT_N(ia); i++) rp[i] = xp[i]; }
+    if      (xe==el_bit) { u64* xp = bitarr_ptr(x); vfor(usz i=0; i<BIT_N(ia); i++) rp[i] = xp[i]; }
     else if (xe==el_i8 ) { i8*  xp = i8any_ptr (x); for(usz i=0; i<ia; i++) bitp_set(rp,i,xp[i]); }
     else if (xe==el_i16) { i16* xp = i16any_ptr(x); for(usz i=0; i<ia; i++) bitp_set(rp,i,xp[i]); }
     else if (xe==el_i32) { i32* xp = i32any_ptr(x); for(usz i=0; i<ia; i++) bitp_set(rp,i,xp[i]); }