Inline hot path of slot_getsomeattrs().

This yields a minor speedup, which roughly balances the loss from the
upcoming introduction of callbacks to do some operations on slots.

Author: Andres Freund
Discussion: https://postgr.es/m/20181105210039.hh4vvi4vwoq5ba2q@alap3.anarazel.de

Author: anarazel

src/backend/executor/execExprInterp.c

@@ -428,7 +428,6 @@ ExecInterpExpr(ExprState *state, ExprContext *econtext, bool *isnull)
 		{
 			CheckOpSlotCompatibility(op, innerslot);
 
-			/* XXX: worthwhile to check tts_nvalid inline first? */
 			slot_getsomeattrs(innerslot, op->d.fetch.last_var);
 
 			EEO_NEXT();

src/backend/executor/execTuples.c

@@ -1183,22 +1183,19 @@ slot_getattr(TupleTableSlot *slot, int attnum, bool *isnull)
 }
 
 /*
- * slot_getsomeattrs
- *		This function forces the entries of the slot's Datum/isnull
- *		arrays to be valid at least up through the attnum'th entry.
+ * slot_getsomeattrs_int - workhorse for slot_getsomeattrs()
  */
 void
-slot_getsomeattrs(TupleTableSlot *slot, int attnum)
+slot_getsomeattrs_int(TupleTableSlot *slot, int attnum)
 {
 	HeapTuple	tuple;
 	int			attno;
 
-	/* Quick out if we have 'em all already */
-	if (slot->tts_nvalid >= attnum)
-		return;
+	/* Check for caller errors */
+	Assert(slot->tts_nvalid < attnum); /* slot_getsomeattr checked */
+	Assert(attnum > 0);
 
-	/* Check for caller error */
-	if (attnum <= 0 || attnum > slot->tts_tupleDescriptor->natts)
+	if (unlikely(attnum > slot->tts_tupleDescriptor->natts))
 		elog(ERROR, "invalid attribute number %d", attnum);
 
 	/*
@@ -1209,9 +1206,7 @@ slot_getsomeattrs(TupleTableSlot *slot, int attnum)
 	if (tuple == NULL)			/* internal error */
 		elog(ERROR, "cannot extract attribute from empty tuple slot");
 
-	/*
-	 * load up any slots available from physical tuple
-	 */
+	/* Fetch as many attributes as possible from the underlying tuple. */
 	attno = HeapTupleHeaderGetNatts(tuple->t_data);
 	attno = Min(attno, attnum);
 
@@ -1220,13 +1215,14 @@ slot_getsomeattrs(TupleTableSlot *slot, int attnum)
 	attno = slot->tts_nvalid;
 
 	/*
-	 * If tuple doesn't have all the atts indicated by attnum, read the rest
-	 * as NULLs or missing values
+	 * If the underlying tuple doesn't have enough attributes, tuple descriptor
+	 * must have the missing attributes.
 	 */
-	if (attno < attnum)
-		slot_getmissingattrs(slot, attno, attnum);
-
-	slot->tts_nvalid = attnum;
+	if (unlikely(slot->tts_nvalid < attnum))
+	{
+		slot_getmissingattrs(slot, slot->tts_nvalid, attnum);
+		slot->tts_nvalid = attnum;
+	}
 }
 
 /* ----------------------------------------------------------------

src/backend/jit/llvm/llvmjit.c

@@ -79,7 +79,7 @@ LLVMTypeRef StructAggStatePerTransData;
 LLVMValueRef AttributeTemplate;
 LLVMValueRef FuncStrlen;
 LLVMValueRef FuncVarsizeAny;
-LLVMValueRef FuncSlotGetsomeattrs;
+LLVMValueRef FuncSlotGetsomeattrsInt;
 LLVMValueRef FuncSlotGetmissingattrs;
 LLVMValueRef FuncMakeExpandedObjectReadOnlyInternal;
 LLVMValueRef FuncExecEvalArrayRefSubscript;
@@ -820,7 +820,7 @@ llvm_create_types(void)
 	AttributeTemplate = LLVMGetNamedFunction(mod, "AttributeTemplate");
 	FuncStrlen = LLVMGetNamedFunction(mod, "strlen");
 	FuncVarsizeAny = LLVMGetNamedFunction(mod, "varsize_any");
-	FuncSlotGetsomeattrs = LLVMGetNamedFunction(mod, "slot_getsomeattrs");
+	FuncSlotGetsomeattrsInt = LLVMGetNamedFunction(mod, "slot_getsomeattrs_int");
 	FuncSlotGetmissingattrs = LLVMGetNamedFunction(mod, "slot_getmissingattrs");
 	FuncMakeExpandedObjectReadOnlyInternal = LLVMGetNamedFunction(mod, "MakeExpandedObjectReadOnlyInternal");
 	FuncExecEvalArrayRefSubscript = LLVMGetNamedFunction(mod, "ExecEvalArrayRefSubscript");

src/backend/jit/llvm/llvmjit_expr.c

@@ -345,7 +345,7 @@ llvm_compile_expr(ExprState *state)
 						params[1] = l_int32_const(op->d.fetch.last_var);
 
 						LLVMBuildCall(b,
-									  llvm_get_decl(mod, FuncSlotGetsomeattrs),
+									  llvm_get_decl(mod, FuncSlotGetsomeattrsInt),
 									  params, lengthof(params), "");
 					}
 

src/backend/jit/llvm/llvmjit_types.c

@@ -97,7 +97,7 @@ void	   *referenced_functions[] =
 {
 	strlen,
 	varsize_any,
-	slot_getsomeattrs,
+	slot_getsomeattrs_int,
 	slot_getmissingattrs,
 	MakeExpandedObjectReadOnlyInternal,
 	ExecEvalArrayRefSubscript,

src/include/executor/tuptable.h

@@ -219,17 +219,28 @@ extern void slot_getmissingattrs(TupleTableSlot *slot, int startAttNum,
 					 int lastAttNum);
 extern Datum slot_getattr(TupleTableSlot *slot, int attnum,
 			 bool *isnull);
-extern void slot_getsomeattrs(TupleTableSlot *slot, int attnum);
 
 /* in access/common/heaptuple.c */
 extern bool slot_attisnull(TupleTableSlot *slot, int attnum);
 extern bool slot_getsysattr(TupleTableSlot *slot, int attnum,
 				Datum *value, bool *isnull);
 extern Datum getmissingattr(TupleDesc tupleDesc,
 			   int attnum, bool *isnull);
+extern void slot_getsomeattrs_int(TupleTableSlot *slot, int attnum);
 
 #ifndef FRONTEND
 
+/*
+ * This function forces the entries of the slot's Datum/isnull arrays to be
+ * valid at least up through the attnum'th entry.
+ */
+static inline void
+slot_getsomeattrs(TupleTableSlot *slot, int attnum)
+{
+	if (slot->tts_nvalid < attnum)
+		slot_getsomeattrs_int(slot, attnum);
+}
+
 /*
  * slot_getallattrs
  *		This function forces all the entries of the slot's Datum/isnull

src/include/jit/llvmjit.h

@@ -77,8 +77,8 @@ extern LLVMTypeRef StructAggStatePerGroupData;
 extern LLVMValueRef AttributeTemplate;
 extern LLVMValueRef FuncStrlen;
 extern LLVMValueRef FuncVarsizeAny;
-extern LLVMValueRef FuncSlotGetsomeattrs;
 extern LLVMValueRef FuncSlotGetmissingattrs;
+extern LLVMValueRef FuncSlotGetsomeattrsInt;
 extern LLVMValueRef FuncMakeExpandedObjectReadOnlyInternal;
 extern LLVMValueRef FuncExecEvalArrayRefSubscript;
 extern LLVMValueRef FuncExecEvalSysVar;