[PATCH] D120738: [AArch64] Improve access to fixed-width object when stack has SVE.

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen created this revision.
sdesmalen added reviewers: bsmith, paulwalker-arm, david-arm, john.brawn.
Herald added subscribers: ctetreau, hiraditya, kristof.beyls, tschuett.
sdesmalen requested review of this revision.
Herald added subscribers: llvm-commits, alextsao1999.
Herald added a project: LLVM.

When the stack has SVE objects, fixed-width objects are often better accessed
from the SP, instead of the FP, because part/all of the fixed-width offset
can be folded into the (non-scalable) addressing mode, where otherwise an
ADDVL would be required.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D120738

Files:
  llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
  llvm/test/CodeGen/AArch64/framelayout-sve-fixed-width-access.mir
  llvm/test/CodeGen/AArch64/stack-guard-sve.ll


Index: llvm/test/CodeGen/AArch64/stack-guard-sve.ll
===================================================================
--- llvm/test/CodeGen/AArch64/stack-guard-sve.ll
+++ llvm/test/CodeGen/AArch64/stack-guard-sve.ll
@@ -152,15 +152,15 @@
 ; CHECK: sub sp, sp, #16, lsl #12
 ; CHECK: sub sp, sp, #16
 
-; Stack guard is placed below the SVE stack area
+; Stack guard is placed below the SVE stack area (and above all fixed-width objects)
+; CHECK-DAG: add [[STACK_GUARD_SPILL_PART_LOC:x[0-9]+]], sp, #8, lsl #12
+; CHECK-DAG: add [[STACK_GUARD_SPILL_PART_LOC]], [[STACK_GUARD_SPILL_PART_LOC]], #16
 ; CHECK-DAG: ldr [[STACK_GUARD:x[0-9]+]], [{{x[0-9]+}}, :lo12:__stack_chk_guard]
-; CHECK-DAG: addvl [[STACK_GUARD_POS:x[0-9]+]], x29, #-2
-; CHECK-DAG: stur [[STACK_GUARD]], [[[STACK_GUARD_POS]], #-8]
+; CHECK-DAG: str [[STACK_GUARD]], [[[STACK_GUARD_SPILL_PART_LOC]], #32760]
 
 ; char_arr is below the stack guard
-; CHECK-DAG: sub [[CHAR_ARR_1:x[0-9]+]], x29, #16
-; CHECK-DAG: addvl [[CHAR_ARR_2:x[0-9]+]], [[CHAR_ARR_1]], #-2
-; CHECK-DAG: strb wzr, [[[CHAR_ARR_2]]]
+; CHECK-DAG: add [[CHAR_ARR_LOC:x[0-9]+]], sp, #16, lsl #12
+; CHECK-DAG: strb wzr, [[[CHAR_ARR_LOC]]]
 
 ; large1 is accessed via a virtual base register
 ; CHECK-DAG: add [[LARGE1:x[0-9]+]], sp, #8, lsl #12
@@ -208,8 +208,9 @@
 ; CHECK-DAG: str [[STACK_GUARD]], [[[STACK_GUARD_POS]]]
 
 ; char_arr is below the SVE stack area
-; CHECK-DAG: addvl [[CHAR_ARR:x[0-9]+]], x29, #-3
-; CHECK-DAG: sturb wzr, [[[CHAR_ARR]], #-8]
+; CHECK-DAG: add [[CHAR_ARR:x[0-9]+]], sp, #15, lsl #12            // =61440
+; CHECK-DAG: add [[CHAR_ARR]], [[CHAR_ARR]], #9
+; CHECK-DAG: strb wzr, [[[CHAR_ARR]], #4095]
 
 ; large1 is accessed via a virtual base register
 ; CHECK-DAG: add [[LARGE1:x[0-9]+]], sp, #8, lsl #12
Index: llvm/test/CodeGen/AArch64/framelayout-sve-fixed-width-access.mir
===================================================================
--- llvm/test/CodeGen/AArch64/framelayout-sve-fixed-width-access.mir
+++ llvm/test/CodeGen/AArch64/framelayout-sve-fixed-width-access.mir
@@ -10,9 +10,7 @@
   ; CHECK-NEXT:    addvl sp, sp, #-32
   ; CHECK-NEXT:    addvl sp, sp, #-28
   ; CHECK-NEXT:    sub sp, sp, #2064
-  ; CHECK-NEXT:    addvl x8, x29, #-32
-  ; CHECK-NEXT:    addvl x8, x8, #-28
-  ; CHECK-NEXT:    ldur x8, [x8, #-16]
+  ; CHECK-NEXT:    ldr x8, [sp, #2048]
   ; CHECK-NEXT:    addvl sp, sp, #31
   ; CHECK-NEXT:    addvl sp, sp, #29
   ; CHECK-NEXT:    add sp, sp, #2064
Index: llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
+++ llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -2012,8 +2012,9 @@
   // right thing for the emergency spill slot.
   bool UseFP = false;
   if (AFI->hasStackFrame() && !isSVE) {
-    // We shouldn't prefer using the FP when there is an SVE area
-    // in between the FP and the non-SVE locals/spills.
+    // We shouldn't prefer using the FP to access fixed-sized stack objects when
+    // there are scalable (SVE) objects in between the FP and the fixed-sized
+    // objects.
     PreferFP &= !SVEStackSize;
 
     // Note: Keeping the following as multiple 'if' statements rather than
@@ -2034,7 +2035,7 @@
       // offsets is smaller than for positive ones. If an offset is available
       // via the FP and the SP, use whichever is closest.
       bool FPOffsetFits = !ForSimm || FPOffset >= -256;
-      PreferFP |= Offset > -FPOffset;
+      PreferFP |= Offset > -FPOffset && !SVEStackSize;
 
       if (MFI.hasVarSizedObjects()) {
         // If we have variable sized objects, we can use either FP or BP, as the


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D120738.412135.patch
Type: text/x-patch
Size: 3673 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220301/8a0ce967/attachment.bin>