This patch recognizes when we are doing the optimization of vector extract with
a constant element number when the vector is in memory and the vector's address
is PC-relative, to directly re-form the address using a PC-relative load,
instead of loading the address into a temporary register, and then doing an
indirect load.

I have bootstrapped a compiler on a little endian power8 machine and ran the
testsuite with no regressions.  Can I check this into the trunk?

2019-12-20  Michael Meissner  <meiss...@linux.ibm.com>

        * config/rs6000/rs6000.c (rs6000_reg_to_addr_mask): New helper
        function to identify the address mask of a hard register.
        (rs6000_adjust_vec_address): If we have a PC-relative address and
        a constant vector element number, fold the element number into the
        PC-relative address.

Index: gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/config/rs6000/rs6000.c  (revision 279597)
+++ gcc/config/rs6000/rs6000.c  (working copy)
@@ -6722,6 +6722,30 @@ rs6000_expand_vector_extract (rtx target
     }
 }
 
+/* Helper function to return an address mask based on a physical register.  */
+
+static addr_mask_type
+rs6000_reg_to_addr_mask (rtx reg, machine_mode mode)
+{
+  unsigned int r = reg_or_subregno (reg);
+  addr_mask_type addr_mask;
+
+  gcc_assert (HARD_REGISTER_NUM_P (r));
+  if (INT_REGNO_P (r))
+    addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_GPR];
+
+  else if (FP_REGNO_P (r))
+    addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_FPR];
+
+  else if (ALTIVEC_REGNO_P (r))
+    addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_VMX];
+
+  else
+    gcc_unreachable ();
+
+  return addr_mask;
+}
+
 /* Adjust a memory address (MEM) of a vector type to point to a scalar field
    within the vector (ELEMENT) with a mode (SCALAR_MODE).  Use a base register
    temporary (BASE_TMP) to fixup the address.  Return the new memory address
@@ -6854,6 +6878,51 @@ rs6000_adjust_vec_address (rtx scalar_re
        }
     }
 
+  /* For references to local static variables, try to fold a constant offset
+     into the address.  */
+  else if (pcrel_local_address (addr, Pmode) && CONST_INT_P (element_offset))
+    {
+      if (GET_CODE (addr) == CONST)
+       addr = XEXP (addr, 0);
+
+      if (GET_CODE (addr) == PLUS)
+       {
+         rtx op0 = XEXP (addr, 0);
+         rtx op1 = XEXP (addr, 1);
+         if (CONST_INT_P (op1))
+           {
+             HOST_WIDE_INT offset
+               = INTVAL (XEXP (addr, 1)) + INTVAL (element_offset);
+
+             if (offset == 0)
+               new_addr = op0;
+
+             else if (SIGNED_INTEGER_34BIT_P (offset))
+               {
+                 rtx plus = gen_rtx_PLUS (Pmode, op0, GEN_INT (offset));
+                 new_addr = gen_rtx_CONST (Pmode, plus);
+               }
+
+             else
+               {
+                 emit_move_insn (base_tmp, addr);
+                 new_addr = gen_rtx_PLUS (Pmode, base_tmp, element_offset);
+               }
+           }
+         else
+           {
+             emit_move_insn (base_tmp, addr);
+             new_addr = gen_rtx_PLUS (Pmode, base_tmp, element_offset);
+           }
+       }
+
+      else
+       {
+         rtx plus = gen_rtx_PLUS (Pmode, addr, element_offset);
+         new_addr = gen_rtx_CONST (Pmode, plus);
+       }
+    }
+
   else
     {
       /* Make sure base_tmp is not the same as element_offset.  This can happen
@@ -6869,21 +6938,8 @@ rs6000_adjust_vec_address (rtx scalar_re
   if (GET_CODE (new_addr) == PLUS)
     {
       rtx op1 = XEXP (new_addr, 1);
-      addr_mask_type addr_mask;
-      unsigned int scalar_regno = reg_or_subregno (scalar_reg);
-
-      gcc_assert (HARD_REGISTER_NUM_P (scalar_regno));
-      if (INT_REGNO_P (scalar_regno))
-       addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_GPR];
-
-      else if (FP_REGNO_P (scalar_regno))
-       addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_FPR];
-
-      else if (ALTIVEC_REGNO_P (scalar_regno))
-       addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_VMX];
-
-      else
-       gcc_unreachable ();
+      addr_mask_type addr_mask
+       = rs6000_reg_to_addr_mask (scalar_reg, scalar_mode);
 
       if (REG_P (op1) || SUBREG_P (op1))
        valid_addr_p = (addr_mask & RELOAD_REG_INDEXED) != 0;
@@ -6891,9 +6947,21 @@ rs6000_adjust_vec_address (rtx scalar_re
        valid_addr_p = (addr_mask & RELOAD_REG_OFFSET) != 0;
     }
 
+  /* An address that is a single register is always valid for either indexed or
+     offsettable loads.  */
   else if (REG_P (new_addr) || SUBREG_P (new_addr))
     valid_addr_p = true;
 
+  /* If we have a PC-relative address, check if offsetable loads are
+     allowed.  */
+  else if (pcrel_local_address (new_addr, Pmode))
+    {
+      addr_mask_type addr_mask
+       = rs6000_reg_to_addr_mask (scalar_reg, scalar_mode);
+
+      valid_addr_p = (addr_mask & RELOAD_REG_OFFSET) != 0;
+    }
+
   else
     valid_addr_p = false;
 

-- 
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meiss...@linux.ibm.com, phone: +1 (978) 899-4797

Reply via email to