if a bitstype contains padding (or is really large), emit inline code to

test the fields rather than using memcmp over the whole thing this is the compile time system side of JuliaLang#11813 (closes JuliaLang#11813)
jiahao · Jun 23, 2015 · 9cd6ad4 · 9cd6ad4
1 parent 6ec7c21
commit 9cd6ad4
Show file tree

Hide file tree

Showing 2 changed files with 49 additions and 16 deletions.
diff --git a/src/codegen.cpp b/src/codegen.cpp
@@ -1960,20 +1960,45 @@ static Value *emit_f_is(jl_value_t *rt1, jl_value_t *rt2,
  if (at1->isPointerTy()) {
  Type *elty = julia_type_to_llvm(rt1);
  if (elty->isAggregateType()) {
+ assert(jl_is_datatype(rt1));
+ size_t sz = jl_datatype_size(rt1);
+ if (sz > 512 && !((jl_datatype_t*)rt1)->haspadding) {
 #ifdef LLVM37
- answer = builder.CreateCall(prepare_call(memcmp_func),
- {
- builder.CreatePointerCast(varg1, T_pint8),
- builder.CreatePointerCast(varg2, T_pint8),
- builder.CreateTrunc(ConstantExpr::getSizeOf(elty), T_size)
- });
+  answer = builder.CreateCall(prepare_call(memcmp_func),
+  {
+  builder.CreatePointerCast(varg1, T_pint8),
+  builder.CreatePointerCast(varg2, T_pint8),
+  ConstantInt::get(T_size, sz)
+  });
 #else
- answer = builder.CreateCall3(memcmp_func,
- builder.CreatePointerCast(varg1, T_pint8),
- builder.CreatePointerCast(varg2, T_pint8),
- builder.CreateTrunc(ConstantExpr::getSizeOf(elty), T_size));
+  answer = builder.CreateCall3(prepare_call(memcmp_func),
+  builder.CreatePointerCast(varg1, T_pint8),
+  builder.CreatePointerCast(varg2, T_pint8),
+  ConstantInt::get(T_size, sz));
 #endif
- answer = builder.CreateICmpEQ(answer, ConstantInt::get(T_int32, 0));
+ answer = builder.CreateICmpEQ(answer, ConstantInt::get(T_int32, 0));
+ }
+ else {
+ varg1 = builder.CreatePointerCast(varg1, elty->getPointerTo());
+ varg2 = builder.CreatePointerCast(varg2, elty->getPointerTo());
+ jl_svec_t *types = ((jl_datatype_t*)rt1)->types;
+ answer = ConstantInt::get(T_int1, 1);
+ size_t l = jl_svec_len(types);
+ for(unsigned i=0; i < l; i++) {
+ jl_value_t *fldty = jl_svecref(types, i);
+ Value *subAns;
+ Value *fld1 = builder.CreateConstGEP2_32(varg1, 0, i);
+ Value *fld2 = builder.CreateConstGEP2_32(varg2, 0, i);
+ if (type_is_ghost(fld1->getType()))
+ continue;
+ if (!fld1->getType()->getContainedType(0)->isAggregateType()) {
+ fld1 = builder.CreateLoad(fld1);
+ fld2 = builder.CreateLoad(fld2);
+ }
+ subAns = emit_f_is(fldty, fldty, NULL, NULL, fld1, fld2, ctx);
+ answer = builder.CreateAnd(answer, subAns);
+ }
+ }
  goto done;
  }
  }
@@ -1990,11 +2015,10 @@ static Value *emit_f_is(jl_value_t *rt1, jl_value_t *rt2,
  for(unsigned i=0; i < l; i++) {
  jl_value_t *fldty = jl_svecref(types,i);
  Value *subAns;
- subAns =
- emit_f_is(fldty, fldty, NULL, NULL,
- builder.CreateExtractElement(varg1, ConstantInt::get(T_int32,i)),
- builder.CreateExtractElement(varg2, ConstantInt::get(T_int32,i)),
- ctx);
+ subAns = emit_f_is(fldty, fldty, NULL, NULL,
+ builder.CreateExtractElement(varg1, ConstantInt::get(T_int32,i)),
+ builder.CreateExtractElement(varg2, ConstantInt::get(T_int32,i)),
+ ctx);
  answer = builder.CreateAnd(answer, subAns);
  }
  goto done;

diff --git a/test/core.jl b/test/core.jl
@@ -3004,4 +3004,13 @@ let a = UInt8[1, 107, 66, 88, 2, 99, 254, 13, 0, 0, 0, 0]
  a[9:end] = reinterpret(UInt8, u32)
  p = pointer(a)
  @test (Int8(1),(Int8(2),Int32(3))) === unsafe_load(convert(Ptr{Tuple{Int8,Tuple{Int8,Int32}}},p))
+ f11813(p) = (Int8(1),(Int8(2),Int32(3))) === unsafe_load(convert(Ptr{Tuple{Int8,Tuple{Int8,Int32}}},p))
+ @test f11813(p) === true # redundant comparison test seems to make this test more reliable, don't remove
+end
+let a = (1:1000...),
+ b = (1:1000...)
+ @test a == b
+ @test a === b
+ @test (a == b) === true
+ @test (a === b) === true
 end