caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
f3a47b9e25e7aca0dc2cd2dec30cedd9aeffaecf
clang-p2996/llvm/test/CodeGen/PowerPC/ctrloop-sh.ll
Roman Lebedev cc39c3b17f [Codegen][LegalizeIntegerTypes] New legalization strategy for scalar shifts: shift through stack 
https://reviews.llvm.org/D140493 is going to teach SROA how to promote allocas
that have variably-indexed loads. That does bring up questions of cost model,
since that requires creating wide shifts.

Indeed, our legalization for them is not optimal.
We either split it into parts, or lower it into a libcall.
But if the shift amount is by a multiple of CHAR_BIT,
we can also legalize it throught stack.

The basic idea is very simple:
1. Get a stack slot 2x the width of the shift type
2. store the value we are shifting into one half of the slot
3. pad the other half of the slot. for logical shifts, with zero, for arithmetic shift with signbit
4. index into the slot (starting from the base half into which we spilled, either upwards or downwards)
5. load
6. split loaded integer

This works for both little-endian and big-endian machines:
https://alive2.llvm.org/ce/z/YNVwd5

And better yet, if the original shift amount was not a multiple of CHAR_BIT,
we can just shift by that remainder afterwards: https://alive2.llvm.org/ce/z/pz5G-K

I think, if we are going perform shift->shift-by-parts expansion more than once,
we should instead go through stack, which is what this patch does.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D140638
2023-01-14 19:12:18 +03:00

236 lines
7.4 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -verify-machineinstrs < %s | FileCheck %s
target datalayout = "E-m:e-p:32:32-i128:64-n32"
target triple = "powerpc-ellcc-linux"
; Function Attrs: nounwind
define void @foo1(ptr %a, ptr readonly %b, ptr readonly %c) #0 {
; CHECK-LABEL: foo1:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: stwu 1, -64(1)
; CHECK-NEXT: stw 28, 48(1) # 4-byte Folded Spill
; CHECK-NEXT: li 8, 2048
; CHECK-NEXT: stw 29, 52(1) # 4-byte Folded Spill
; CHECK-NEXT: li 6, 0
; CHECK-NEXT: stw 30, 56(1) # 4-byte Folded Spill
; CHECK-NEXT: li 7, 7
; CHECK-NEXT: mtctr 8
; CHECK-NEXT: addi 8, 1, 16
; CHECK-NEXT: .LBB0_1: # %for.body
; CHECK-NEXT: #
; CHECK-NEXT: lwz 9, 0(4)
; CHECK-NEXT: lwz 10, 4(4)
; CHECK-NEXT: lwz 11, 8(4)
; CHECK-NEXT: lwz 12, 12(4)
; CHECK-NEXT: lwz 0, 12(5)
; CHECK-NEXT: stw 6, 44(1)
; CHECK-NEXT: stw 6, 40(1)
; CHECK-NEXT: stw 6, 36(1)
; CHECK-NEXT: stw 6, 32(1)
; CHECK-NEXT: stw 12, 28(1)
; CHECK-NEXT: clrlwi 12, 0, 29
; CHECK-NEXT: stw 11, 24(1)
; CHECK-NEXT: nand 11, 0, 7
; CHECK-NEXT: stw 10, 20(1)
; CHECK-NEXT: subfic 29, 12, 32
; CHECK-NEXT: stw 9, 16(1)
; CHECK-NEXT: rlwinm 9, 0, 29, 28, 31
; CHECK-NEXT: lwzux 10, 9, 8
; CHECK-NEXT: clrlwi 11, 11, 27
; CHECK-NEXT: lwz 0, 8(9)
; CHECK-NEXT: slw 10, 10, 12
; CHECK-NEXT: lwz 30, 4(9)
; CHECK-NEXT: lwz 9, 12(9)
; CHECK-NEXT: slw 28, 30, 12
; CHECK-NEXT: srw 30, 30, 29
; CHECK-NEXT: srw 29, 9, 29
; CHECK-NEXT: slw 9, 9, 12
; CHECK-NEXT: slw 12, 0, 12
; CHECK-NEXT: srwi 0, 0, 1
; CHECK-NEXT: stw 9, 12(3)
; CHECK-NEXT: or 9, 12, 29
; CHECK-NEXT: srw 11, 0, 11
; CHECK-NEXT: stw 9, 8(3)
; CHECK-NEXT: or 9, 10, 30
; CHECK-NEXT: stw 9, 0(3)
; CHECK-NEXT: or 9, 28, 11
; CHECK-NEXT: stw 9, 4(3)
; CHECK-NEXT: bdnz .LBB0_1
; CHECK-NEXT: # %bb.2: # %for.end
; CHECK-NEXT: lwz 30, 56(1) # 4-byte Folded Reload
; CHECK-NEXT: lwz 29, 52(1) # 4-byte Folded Reload
; CHECK-NEXT: lwz 28, 48(1) # 4-byte Folded Reload
; CHECK-NEXT: addi 1, 1, 64
; CHECK-NEXT: blr
entry:
br label %for.body
for.body: ; preds = %for.body, %entry
%i.02 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%0 = load i128, ptr %b, align 16
%1 = load i128, ptr %c, align 16
%shl = shl i128 %0, %1
store i128 %shl, ptr %a, align 16
%inc = add nsw i32 %i.02, 1
%exitcond = icmp eq i32 %inc, 2048
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
; Function Attrs: nounwind
define void @foo2(ptr %a, ptr readonly %b, ptr readonly %c) #0 {
; CHECK-LABEL: foo2:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: stwu 1, -64(1)
; CHECK-NEXT: stw 29, 52(1) # 4-byte Folded Spill
; CHECK-NEXT: li 7, 2048
; CHECK-NEXT: stw 30, 56(1) # 4-byte Folded Spill
; CHECK-NEXT: li 6, 7
; CHECK-NEXT: mtctr 7
; CHECK-NEXT: addi 7, 1, 36
; CHECK-NEXT: .LBB1_1: # %for.body
; CHECK-NEXT: #
; CHECK-NEXT: lwz 8, 0(4)
; CHECK-NEXT: lwz 10, 8(4)
; CHECK-NEXT: lwz 12, 12(5)
; CHECK-NEXT: lwz 9, 4(4)
; CHECK-NEXT: lwz 11, 12(4)
; CHECK-NEXT: stw 10, 44(1)
; CHECK-NEXT: rlwinm 10, 12, 29, 28, 31
; CHECK-NEXT: stw 8, 36(1)
; CHECK-NEXT: srawi 8, 8, 31
; CHECK-NEXT: stw 11, 48(1)
; CHECK-NEXT: clrlwi 11, 12, 29
; CHECK-NEXT: stw 9, 40(1)
; CHECK-NEXT: nand 9, 12, 6
; CHECK-NEXT: stw 8, 32(1)
; CHECK-NEXT: subfic 30, 11, 32
; CHECK-NEXT: stw 8, 28(1)
; CHECK-NEXT: clrlwi 9, 9, 27
; CHECK-NEXT: stw 8, 24(1)
; CHECK-NEXT: stw 8, 20(1)
; CHECK-NEXT: sub 8, 7, 10
; CHECK-NEXT: lwz 10, 4(8)
; CHECK-NEXT: lwz 12, 8(8)
; CHECK-NEXT: lwz 0, 0(8)
; CHECK-NEXT: lwz 8, 12(8)
; CHECK-NEXT: srw 29, 12, 11
; CHECK-NEXT: slw 12, 12, 30
; CHECK-NEXT: slw 30, 0, 30
; CHECK-NEXT: srw 8, 8, 11
; CHECK-NEXT: sraw 0, 0, 11
; CHECK-NEXT: srw 11, 10, 11
; CHECK-NEXT: slwi 10, 10, 1
; CHECK-NEXT: or 8, 12, 8
; CHECK-NEXT: slw 9, 10, 9
; CHECK-NEXT: stw 8, 12(3)
; CHECK-NEXT: or 8, 30, 11
; CHECK-NEXT: stw 8, 4(3)
; CHECK-NEXT: or 8, 29, 9
; CHECK-NEXT: stw 0, 0(3)
; CHECK-NEXT: stw 8, 8(3)
; CHECK-NEXT: bdnz .LBB1_1
; CHECK-NEXT: # %bb.2: # %for.end
; CHECK-NEXT: lwz 30, 56(1) # 4-byte Folded Reload
; CHECK-NEXT: lwz 29, 52(1) # 4-byte Folded Reload
; CHECK-NEXT: addi 1, 1, 64
; CHECK-NEXT: blr
entry:
br label %for.body
for.body: ; preds = %for.body, %entry
%i.02 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%0 = load i128, ptr %b, align 16
%1 = load i128, ptr %c, align 16
%shl = ashr i128 %0, %1
store i128 %shl, ptr %a, align 16
%inc = add nsw i32 %i.02, 1
%exitcond = icmp eq i32 %inc, 2048
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
; Function Attrs: nounwind
define void @foo3(ptr %a, ptr readonly %b, ptr readonly %c) #0 {
; CHECK-LABEL: foo3:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: stwu 1, -64(1)
; CHECK-NEXT: stw 28, 48(1) # 4-byte Folded Spill
; CHECK-NEXT: li 8, 2048
; CHECK-NEXT: stw 29, 52(1) # 4-byte Folded Spill
; CHECK-NEXT: li 6, 0
; CHECK-NEXT: stw 30, 56(1) # 4-byte Folded Spill
; CHECK-NEXT: li 7, 7
; CHECK-NEXT: mtctr 8
; CHECK-NEXT: addi 8, 1, 32
; CHECK-NEXT: .LBB2_1: # %for.body
; CHECK-NEXT: #
; CHECK-NEXT: lwz 10, 4(4)
; CHECK-NEXT: lwz 0, 12(5)
; CHECK-NEXT: lwz 9, 0(4)
; CHECK-NEXT: lwz 11, 8(4)
; CHECK-NEXT: lwz 12, 12(4)
; CHECK-NEXT: stw 10, 36(1)
; CHECK-NEXT: rlwinm 10, 0, 29, 28, 31
; CHECK-NEXT: stw 6, 28(1)
; CHECK-NEXT: sub 10, 8, 10
; CHECK-NEXT: stw 6, 24(1)
; CHECK-NEXT: stw 6, 20(1)
; CHECK-NEXT: stw 6, 16(1)
; CHECK-NEXT: stw 12, 44(1)
; CHECK-NEXT: clrlwi 12, 0, 29
; CHECK-NEXT: stw 11, 40(1)
; CHECK-NEXT: subfic 29, 12, 32
; CHECK-NEXT: stw 9, 32(1)
; CHECK-NEXT: nand 9, 0, 7
; CHECK-NEXT: lwz 11, 4(10)
; CHECK-NEXT: clrlwi 9, 9, 27
; CHECK-NEXT: lwz 0, 8(10)
; CHECK-NEXT: lwz 30, 0(10)
; CHECK-NEXT: lwz 10, 12(10)
; CHECK-NEXT: srw 28, 0, 12
; CHECK-NEXT: slw 0, 0, 29
; CHECK-NEXT: slw 29, 30, 29
; CHECK-NEXT: srw 10, 10, 12
; CHECK-NEXT: srw 30, 30, 12
; CHECK-NEXT: srw 12, 11, 12
; CHECK-NEXT: slwi 11, 11, 1
; CHECK-NEXT: slw 9, 11, 9
; CHECK-NEXT: or 10, 0, 10
; CHECK-NEXT: stw 10, 12(3)
; CHECK-NEXT: or 10, 29, 12
; CHECK-NEXT: or 9, 28, 9
; CHECK-NEXT: stw 30, 0(3)
; CHECK-NEXT: stw 10, 4(3)
; CHECK-NEXT: stw 9, 8(3)
; CHECK-NEXT: bdnz .LBB2_1
; CHECK-NEXT: # %bb.2: # %for.end
; CHECK-NEXT: lwz 30, 56(1) # 4-byte Folded Reload
; CHECK-NEXT: lwz 29, 52(1) # 4-byte Folded Reload
; CHECK-NEXT: lwz 28, 48(1) # 4-byte Folded Reload
; CHECK-NEXT: addi 1, 1, 64
; CHECK-NEXT: blr
entry:
br label %for.body
for.body: ; preds = %for.body, %entry
%i.02 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%0 = load i128, ptr %b, align 16
%1 = load i128, ptr %c, align 16
%shl = lshr i128 %0, %1
store i128 %shl, ptr %a, align 16
%inc = add nsw i32 %i.02, 1
%exitcond = icmp eq i32 %inc, 2048
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
attributes #0 = { nounwind }
Reference in New Issue View Git Blame Copy Permalink