Fix low reg issue in Thumb-2 movsi patterns

The Thumb-2 movsi patterns try to prefer low registers for loads and stores.
However this is done incorrectly by using 2 separate variants with 'l' and 'h'
register classes.  The register allocator will only use low registers, and
as a result we end up with significantly more spills and moves to high
registers.  Fix this by merging the alternatives and use 'l*r' to indicate
preference for low registers.  This saves ~400 instructions from the pr77308
testcase.

    gcc/
	* config/arm/thumb2.md (thumb2_movsi_insn): Fix load/store low reg.
	* config/arm/vfp.md (thumb2_movsi_vfp): Likewise.

From-SVN: r273802
This commit is contained in:
Wilco Dijkstra 2019-07-25 14:24:41 +00:00 committed by Wilco Dijkstra
parent 2fc233b731
commit 7508869693
3 changed files with 26 additions and 25 deletions

View File

@ -1,3 +1,8 @@
2019-07-25 Wilco Dijkstra <wdijkstr@arm.com>
* config/arm/thumb2.md (thumb2_movsi_insn): Fix load/store low reg.
* config/arm/vfp.md (thumb2_movsi_vfp): Likewise.
2019-07-23 Jan Hubicka <hubicka@ucw.cz>
* ipa-devirt.c (add_type_duplicate): Fix return value.

View File

@ -247,8 +247,8 @@
;; regs. The high register alternatives are not taken into account when
;; choosing register preferences in order to reflect their expense.
(define_insn "*thumb2_movsi_insn"
[(set (match_operand:SI 0 "nonimmediate_operand" "=rk,r,l,r,r,l ,*hk,m,*m")
(match_operand:SI 1 "general_operand" "rk,I,Py,K,j,mi,*mi,l,*hk"))]
[(set (match_operand:SI 0 "nonimmediate_operand" "=rk,r,l,r,r,lk*r,m")
(match_operand:SI 1 "general_operand" "rk,I,Py,K,j,mi,lk*r"))]
"TARGET_THUMB2 && !TARGET_IWMMXT && !TARGET_HARD_FLOAT
&& ( register_operand (operands[0], SImode)
|| register_operand (operands[1], SImode))"
@ -262,22 +262,20 @@
case 3: return \"mvn%?\\t%0, #%B1\";
case 4: return \"movw%?\\t%0, %1\";
case 5:
case 6:
/* Cannot load it directly, split to load it via MOV / MOVT. */
if (!MEM_P (operands[1]) && arm_disable_literal_pool)
return \"#\";
return \"ldr%?\\t%0, %1\";
case 7:
case 8: return \"str%?\\t%1, %0\";
case 6: return \"str%?\\t%1, %0\";
default: gcc_unreachable ();
}
}
[(set_attr "type" "mov_reg,mov_imm,mov_imm,mvn_imm,mov_imm,load_4,load_4,store_4,store_4")
(set_attr "length" "2,4,2,4,4,4,4,4,4")
[(set_attr "type" "mov_reg,mov_imm,mov_imm,mvn_imm,mov_imm,load_4,store_4")
(set_attr "length" "2,4,2,4,4,4,4")
(set_attr "predicable" "yes")
(set_attr "predicable_short_it" "yes,no,yes,no,no,no,no,no,no")
(set_attr "pool_range" "*,*,*,*,*,1018,4094,*,*")
(set_attr "neg_pool_range" "*,*,*,*,*,0,0,*,*")]
(set_attr "predicable_short_it" "yes,no,yes,no,no,no,no")
(set_attr "pool_range" "*,*,*,*,*,4094,*")
(set_attr "neg_pool_range" "*,*,*,*,*,0,*")]
)
(define_insn "tls_load_dot_plus_four"

View File

@ -258,8 +258,8 @@
;; is chosen with length 2 when the instruction is predicated for
;; arm_restrict_it.
(define_insn "*thumb2_movsi_vfp"
[(set (match_operand:SI 0 "nonimmediate_operand" "=rk,r,l,r,r, l,*hk,m, *m,*t, r,*t,*t, *Uv")
(match_operand:SI 1 "general_operand" "rk,I,Py,K,j,mi,*mi,l,*hk, r,*t,*t,*UvTu,*t"))]
[(set (match_operand:SI 0 "nonimmediate_operand" "=rk,r,l,r,r,lk*r,m,*t, r,*t,*t, *Uv")
(match_operand:SI 1 "general_operand" "rk,I,Py,K,j,mi,lk*r, r,*t,*t,*UvTu,*t"))]
"TARGET_THUMB2 && TARGET_HARD_FLOAT
&& ( s_register_operand (operands[0], SImode)
|| s_register_operand (operands[1], SImode))"
@ -275,32 +275,30 @@
case 4:
return \"movw%?\\t%0, %1\";
case 5:
case 6:
/* Cannot load it directly, split to load it via MOV / MOVT. */
if (!MEM_P (operands[1]) && arm_disable_literal_pool)
return \"#\";
return \"ldr%?\\t%0, %1\";
case 7:
case 8:
case 6:
return \"str%?\\t%1, %0\";
case 7:
return \"vmov%?\\t%0, %1\\t%@ int\";
case 8:
return \"vmov%?\\t%0, %1\\t%@ int\";
case 9:
return \"vmov%?\\t%0, %1\\t%@ int\";
case 10:
return \"vmov%?\\t%0, %1\\t%@ int\";
case 11:
return \"vmov%?.f32\\t%0, %1\\t%@ int\";
case 12: case 13:
case 10: case 11:
return output_move_vfp (operands);
default:
gcc_unreachable ();
}
"
[(set_attr "predicable" "yes")
(set_attr "predicable_short_it" "yes,no,yes,no,no,no,no,no,no,no,no,no,no,no")
(set_attr "type" "mov_reg,mov_reg,mov_reg,mvn_reg,mov_imm,load_4,load_4,store_4,store_4,f_mcr,f_mrc,fmov,f_loads,f_stores")
(set_attr "length" "2,4,2,4,4,4,4,4,4,4,4,4,4,4")
(set_attr "pool_range" "*,*,*,*,*,1018,4094,*,*,*,*,*,1018,*")
(set_attr "neg_pool_range" "*,*,*,*,*, 0, 0,*,*,*,*,*,1008,*")]
(set_attr "predicable_short_it" "yes,no,yes,no,no,no,no,no,no,no,no,no")
(set_attr "type" "mov_reg,mov_reg,mov_reg,mvn_reg,mov_imm,load_4,store_4,f_mcr,f_mrc,fmov,f_loads,f_stores")
(set_attr "length" "2,4,2,4,4,4,4,4,4,4,4,4")
(set_attr "pool_range" "*,*,*,*,*,4094,*,*,*,*,1018,*")
(set_attr "neg_pool_range" "*,*,*,*,*, 0,*,*,*,*,1008,*")]
)