6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-21 00:32:23 +00:00
Code Issues Projects Releases Wiki Activity
bc051faae1
llvm-6502/test/CodeGen/PowerPC/bperm.ll

280 lines
6.6 KiB
LLVM
Raw Normal View History

[PowerPC] Improve instruction selection bit-permuting operations (32-bit) The PowerPC backend, somewhat embarrassingly, did not generate an optimal-length sequence of instructions for a 32-bit bswap. While adding a pattern for the bswap intrinsic to fix this would not have been terribly difficult, doing so would not have addressed the real problem: we had been generating poor code for many bit-permuting operations (by which I mean things like byte-swap that permute the bits of one or more inputs around in various ways). Here are some initial steps toward solving this deficiency. Bit-permuting operations are represented, at the SDAG level, using ISD::ROTL, SHL, SRL, AND and OR (mostly with constant second operands). Looking back through these operations, we can build up a description of the bits in the resulting value in terms of bits of one or more input values (and constant zeros). For each bit, we compute the rotation amount from the original value, and then group consecutive (value, rotation factor) bits into groups. Groups sharing these attributes are then collected and sorted, and we can then instruction select the entire permutation using a combination of masked rotations (rlwinm), imm ands (andi/andis), and masked rotation inserts (rlwimi). The result is that instead of lowering an i32 bswap as: rlwinm 5, 3, 24, 16, 23 rlwinm 4, 3, 24, 0, 7 rlwimi 4, 3, 8, 8, 15 rlwimi 5, 3, 8, 24, 31 rlwimi 4, 5, 0, 16, 31 we now produce: rlwinm 4, 3, 8, 0, 31 rlwimi 4, 3, 24, 16, 23 rlwimi 4, 3, 24, 0, 7 and for the 'test6' example in the PowerPC/README.txt file: unsigned test6(unsigned x) { return ((x & 0x00FF0000) >> 16) | ((x & 0x000000FF) << 16); } we used to produce: lis 4, 255 rlwinm 3, 3, 16, 0, 31 ori 4, 4, 255 and 3, 3, 4 and now we produce: rlwinm 4, 3, 16, 24, 31 rlwimi 4, 3, 16, 8, 15 and, as a nice bonus, this fixes the FIXME in test/CodeGen/PowerPC/rlwimi-and.ll. This commit does not include instruction-selection for i64 operations, those will come later. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224318 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-16 05:51:41 +00:00
; RUN: llc -mcpu=pwr7 < %s | FileCheck %s
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
; Function Attrs: nounwind readnone
define zeroext i32 @bs4(i32 zeroext %a) #0 {
entry:
%0 = tail call i32 @llvm.bswap.i32(i32 %a)
ret i32 %0
; CHECK-LABEL: @bs4
; CHECK: rlwinm [[REG1:[0-9]+]], 3, 8, 0, 31
; CHECK: rlwimi [[REG1]], 3, 24, 16, 23
; CHECK: rlwimi [[REG1]], 3, 24, 0, 7
; CHECK: mr 3, [[REG1]]
; CHECK: blr
}
[PowerPC] Improve instruction selection bit-permuting operations (64-bit) This is the second installment of improvements to instruction selection for "bit permutation" instruction sequences. r224318 added logic for instruction selection for 32-bit bit permutation sequences, and this adds lowering for 64-bit sequences. The 64-bit sequences are more complicated than the 32-bit ones because: a) the 64-bit versions of the 32-bit rotate-and-mask instructions work by replicating the lower 32-bits of the value-to-be-rotated into the upper 32 bits -- and integrating this into the cost modeling for the various bit group operations is non-trivial b) unlike the 32-bit instructions in 32-bit mode, the rotate-and-mask instructions cannot, in one instruction, specify the mask starting index, the mask ending index, and the rotation factor. Also, forming arbitrary 64-bit constants is more complicated than in 32-bit mode because the number of instructions necessary is value dependent. Plus, support for 'late masking' was added: it is sometimes more efficient to treat the overall value as if it had no mandatory zero bits when planning the bit-group insertions, and then mask them in at the very end. Unfortunately, as the structure of the bit groups is different in the two cases, the more feasible implementation technique was to generate both instruction sequences, and then pick the shorter one. And finally, we now generate reasonable code for i64 bswap: rldicl 5, 3, 16, 0 rldicl 4, 3, 8, 0 rldicl 6, 3, 24, 0 rldimi 4, 5, 8, 48 rldicl 5, 3, 32, 0 rldimi 4, 6, 16, 40 rldicl 6, 3, 48, 0 rldimi 4, 5, 24, 32 rldicl 5, 3, 56, 0 rldimi 4, 6, 40, 16 rldimi 4, 5, 48, 8 rldimi 4, 3, 56, 0 vs. what we used to produce: li 4, 255 rldicl 5, 3, 24, 40 rldicl 6, 3, 40, 24 rldicl 7, 3, 56, 8 sldi 8, 3, 8 sldi 10, 3, 24 sldi 12, 3, 40 rldicl 0, 3, 8, 56 sldi 9, 4, 32 sldi 11, 4, 40 sldi 4, 4, 48 andi. 5, 5, 65280 andis. 6, 6, 255 andis. 7, 7, 65280 sldi 3, 3, 56 and 8, 8, 9 and 4, 12, 4 and 9, 10, 11 or 6, 7, 6 or 5, 5, 0 or 3, 3, 4 or 7, 9, 8 or 4, 6, 5 or 3, 3, 7 or 3, 3, 4 which is 12 instructions, instead of 25, and seems optimal (at least in terms of code size). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225056 91177308-0d34-0410-b5e6-96231b3b80d8
2015-01-01 02:53:29 +00:00
define i64 @bs8(i64 %x) #0 {
entry:
%0 = tail call i64 @llvm.bswap.i64(i64 %x)
ret i64 %0
; CHECK-LABEL: @bs8
; CHECK-DAG: rldicl [[REG1:[0-9]+]], 3, 16, 0
; CHECK-DAG: rldicl [[REG2:[0-9]+]], 3, 8, 0
; CHECK-DAG: rldicl [[REG3:[0-9]+]], 3, 24, 0
; CHECK-DAG: rldimi [[REG2]], [[REG1]], 8, 48
; CHECK-DAG: rldicl [[REG4:[0-9]+]], 3, 32, 0
; CHECK-DAG: rldimi [[REG2]], [[REG3]], 16, 40
; CHECK-DAG: rldicl [[REG5:[0-9]+]], 3, 48, 0
; CHECK-DAG: rldimi [[REG2]], [[REG4]], 24, 32
; CHECK-DAG: rldicl [[REG6:[0-9]+]], 3, 56, 0
; CHECK-DAG: rldimi [[REG2]], [[REG5]], 40, 16
; CHECK-DAG: rldimi [[REG2]], [[REG6]], 48, 8
; CHECK-DAG: rldimi [[REG2]], 3, 56, 0
; CHECK: mr 3, [[REG2]]
; CHECK: blr
}
define i64 @test1(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i1, 8
%and = and i64 %0, 5963776000
ret i64 %and
; CHECK-LABEL: @test1
; CHECK-DAG: li [[REG1:[0-9]+]], 11375
; CHECK-DAG: rldicl [[REG3:[0-9]+]], 4, 56, 0
; CHECK-DAG: sldi [[REG2:[0-9]+]], [[REG1]], 19
; CHECK: and 3, [[REG3]], [[REG2]]
; CHECK: blr
}
define i64 @test2(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i1, 6
%and = and i64 %0, 133434808670355456
ret i64 %and
; CHECK-LABEL: @test2
; CHECK-DAG: lis [[REG1:[0-9]+]], 474
; CHECK-DAG: rldicl [[REG5:[0-9]+]], 4, 58, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 3648
; CHECK-DAG: sldi [[REG3:[0-9]+]], [[REG2]], 32
; CHECK-DAG: oris [[REG4:[0-9]+]], [[REG3]], 25464
; CHECK: and 3, [[REG5]], [[REG4]]
; CHECK: blr
}
define i64 @test3(i64 %i0, i64 %i1) #0 {
entry:
%0 = shl i64 %i0, 34
%and = and i64 %0, 191795733152661504
ret i64 %and
; CHECK-LABEL: @test3
; CHECK-DAG: lis [[REG1:[0-9]+]], 170
; CHECK-DAG: rldicl [[REG4:[0-9]+]], 3, 34, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 22861
; CHECK-DAG: sldi [[REG3:[0-9]+]], [[REG2]], 34
; CHECK: and 3, [[REG4]], [[REG3]]
; CHECK: blr
}
define i64 @test4(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i1, 15
%and = and i64 %0, 58195968
ret i64 %and
; CHECK-LABEL: @test4
; CHECK: rldicl [[REG1:[0-9]+]], 4, 49, 0
; CHECK: andis. 3, [[REG1]], 888
; CHECK: blr
}
define i64 @test5(i64 %i0, i64 %i1) #0 {
entry:
%0 = shl i64 %i1, 12
%and = and i64 %0, 127252959854592
ret i64 %and
; CHECK-LABEL: @test5
; CHECK-DAG: lis [[REG1:[0-9]+]], 3703
; CHECK-DAG: rldicl [[REG4:[0-9]+]], 4, 12, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 35951
; CHECK-DAG: sldi [[REG3:[0-9]+]], [[REG2]], 19
; CHECK: and 3, [[REG4]], [[REG3]]
; CHECK: blr
}
[PowerPC] Improve instruction selection bit-permuting operations (32-bit) The PowerPC backend, somewhat embarrassingly, did not generate an optimal-length sequence of instructions for a 32-bit bswap. While adding a pattern for the bswap intrinsic to fix this would not have been terribly difficult, doing so would not have addressed the real problem: we had been generating poor code for many bit-permuting operations (by which I mean things like byte-swap that permute the bits of one or more inputs around in various ways). Here are some initial steps toward solving this deficiency. Bit-permuting operations are represented, at the SDAG level, using ISD::ROTL, SHL, SRL, AND and OR (mostly with constant second operands). Looking back through these operations, we can build up a description of the bits in the resulting value in terms of bits of one or more input values (and constant zeros). For each bit, we compute the rotation amount from the original value, and then group consecutive (value, rotation factor) bits into groups. Groups sharing these attributes are then collected and sorted, and we can then instruction select the entire permutation using a combination of masked rotations (rlwinm), imm ands (andi/andis), and masked rotation inserts (rlwimi). The result is that instead of lowering an i32 bswap as: rlwinm 5, 3, 24, 16, 23 rlwinm 4, 3, 24, 0, 7 rlwimi 4, 3, 8, 8, 15 rlwimi 5, 3, 8, 24, 31 rlwimi 4, 5, 0, 16, 31 we now produce: rlwinm 4, 3, 8, 0, 31 rlwimi 4, 3, 24, 16, 23 rlwimi 4, 3, 24, 0, 7 and for the 'test6' example in the PowerPC/README.txt file: unsigned test6(unsigned x) { return ((x & 0x00FF0000) >> 16) | ((x & 0x000000FF) << 16); } we used to produce: lis 4, 255 rlwinm 3, 3, 16, 0, 31 ori 4, 4, 255 and 3, 3, 4 and now we produce: rlwinm 4, 3, 16, 24, 31 rlwimi 4, 3, 16, 8, 15 and, as a nice bonus, this fixes the FIXME in test/CodeGen/PowerPC/rlwimi-and.ll. This commit does not include instruction-selection for i64 operations, those will come later. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224318 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-16 05:51:41 +00:00
; Function Attrs: nounwind readnone
define zeroext i32 @test6(i32 zeroext %x) #0 {
entry:
%and = lshr i32 %x, 16
%shr = and i32 %and, 255
%and1 = shl i32 %x, 16
%shl = and i32 %and1, 16711680
%or = or i32 %shr, %shl
ret i32 %or
; CHECK-LABEL: @test6
; CHECK: rlwinm [[REG1:[0-9]+]], 3, 16, 24, 31
; CHECK: rlwimi [[REG1]], 3, 16, 8, 15
; CHECK: mr 3, [[REG1]]
; CHECK: blr
}
[PowerPC] Improve instruction selection bit-permuting operations (64-bit) This is the second installment of improvements to instruction selection for "bit permutation" instruction sequences. r224318 added logic for instruction selection for 32-bit bit permutation sequences, and this adds lowering for 64-bit sequences. The 64-bit sequences are more complicated than the 32-bit ones because: a) the 64-bit versions of the 32-bit rotate-and-mask instructions work by replicating the lower 32-bits of the value-to-be-rotated into the upper 32 bits -- and integrating this into the cost modeling for the various bit group operations is non-trivial b) unlike the 32-bit instructions in 32-bit mode, the rotate-and-mask instructions cannot, in one instruction, specify the mask starting index, the mask ending index, and the rotation factor. Also, forming arbitrary 64-bit constants is more complicated than in 32-bit mode because the number of instructions necessary is value dependent. Plus, support for 'late masking' was added: it is sometimes more efficient to treat the overall value as if it had no mandatory zero bits when planning the bit-group insertions, and then mask them in at the very end. Unfortunately, as the structure of the bit groups is different in the two cases, the more feasible implementation technique was to generate both instruction sequences, and then pick the shorter one. And finally, we now generate reasonable code for i64 bswap: rldicl 5, 3, 16, 0 rldicl 4, 3, 8, 0 rldicl 6, 3, 24, 0 rldimi 4, 5, 8, 48 rldicl 5, 3, 32, 0 rldimi 4, 6, 16, 40 rldicl 6, 3, 48, 0 rldimi 4, 5, 24, 32 rldicl 5, 3, 56, 0 rldimi 4, 6, 40, 16 rldimi 4, 5, 48, 8 rldimi 4, 3, 56, 0 vs. what we used to produce: li 4, 255 rldicl 5, 3, 24, 40 rldicl 6, 3, 40, 24 rldicl 7, 3, 56, 8 sldi 8, 3, 8 sldi 10, 3, 24 sldi 12, 3, 40 rldicl 0, 3, 8, 56 sldi 9, 4, 32 sldi 11, 4, 40 sldi 4, 4, 48 andi. 5, 5, 65280 andis. 6, 6, 255 andis. 7, 7, 65280 sldi 3, 3, 56 and 8, 8, 9 and 4, 12, 4 and 9, 10, 11 or 6, 7, 6 or 5, 5, 0 or 3, 3, 4 or 7, 9, 8 or 4, 6, 5 or 3, 3, 7 or 3, 3, 4 which is 12 instructions, instead of 25, and seems optimal (at least in terms of code size). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225056 91177308-0d34-0410-b5e6-96231b3b80d8
2015-01-01 02:53:29 +00:00
define i64 @test7(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i0, 5
%and = and i64 %0, 58195968
ret i64 %and
; CHECK-LABEL: @test7
; CHECK: rlwinm [[REG1:[0-9]+]], 3, 27, 9, 12
; CHECK: rlwimi [[REG1]], 3, 27, 6, 7
; CHECK: mr 3, [[REG1]]
; CHECK: blr
}
define i64 @test8(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i0, 1
%and = and i64 %0, 169172533248
ret i64 %and
; CHECK-LABEL: @test8
; CHECK-DAG: lis [[REG1:[0-9]+]], 4
; CHECK-DAG: rldicl [[REG4:[0-9]+]], 3, 63, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 60527
; CHECK-DAG: sldi [[REG3:[0-9]+]], [[REG2]], 19
; CHECK: and 3, [[REG4]], [[REG3]]
; CHECK: blr
}
define i64 @test9(i64 %i0, i64 %i1) #0 {
entry:
%0 = lshr i64 %i1, 14
%and = and i64 %0, 18848677888
%1 = shl i64 %i1, 51
%and3 = and i64 %1, 405323966463344640
%or4 = or i64 %and, %and3
ret i64 %or4
; CHECK-LABEL: @test9
; CHECK-DAG: lis [[REG1:[0-9]+]], 1440
; CHECK-DAG: rldicl [[REG5:[0-9]+]], 4, 62, 0
; CHECK-DAG: rldicl [[REG6:[0-9]+]], 4, 50, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 4
; CHECK-DAG: rldimi [[REG6]], [[REG5]], 53, 0
; CHECK-DAG: sldi [[REG3:[0-9]+]], [[REG2]], 32
; CHECK-DAG: oris [[REG4:[0-9]+]], [[REG3]], 25464
; CHECK: and 3, [[REG6]], [[REG4]]
; CHECK: blr
}
define i64 @test10(i64 %i0, i64 %i1) #0 {
entry:
%0 = shl i64 %i0, 37
%and = and i64 %0, 15881483390550016
%1 = shl i64 %i0, 25
%and3 = and i64 %1, 2473599172608
%or4 = or i64 %and, %and3
ret i64 %or4
; CHECK-LABEL: @test10
; CHECK-DAG: lis [[REG1:[0-9]+]], 1
; CHECK-DAG: rldicl [[REG6:[0-9]+]], 3, 25, 0
; CHECK-DAG: rldicl [[REG7:[0-9]+]], 3, 37, 0
; CHECK-DAG: ori [[REG2:[0-9]+]], [[REG1]], 8183
; CHECK-DAG: ori [[REG3:[0-9]+]], [[REG1]], 50017
; CHECK-DAG: sldi [[REG4:[0-9]+]], [[REG2]], 25
; CHECK-DAG: sldi [[REG5:[0-9]+]], [[REG3]], 37
; CHECK-DAG: and [[REG8:[0-9]+]], [[REG6]], [[REG4]]
; CHECK-DAG: and [[REG9:[0-9]+]], [[REG7]], [[REG5]]
; CHECK: or 3, [[REG9]], [[REG8]]
; CHECK: blr
}
define i64 @test11(i64 %x) #0 {
entry:
%and = and i64 %x, 4294967295
%shl = shl i64 %x, 32
%or = or i64 %and, %shl
ret i64 %or
; CHECK-LABEL: @test11
; CHECK: rlwinm 3, 3, 0, 1, 0
; CHECK: blr
}
define i64 @test12(i64 %x) #0 {
entry:
%and = and i64 %x, 4294905855
%shl = shl i64 %x, 32
%or = or i64 %and, %shl
ret i64 %or
; CHECK-LABEL: @test12
; CHECK: rlwinm 3, 3, 0, 20, 15
; CHECK: blr
}
define i64 @test13(i64 %x) #0 {
entry:
%shl = shl i64 %x, 4
%and = and i64 %shl, 240
%shr = lshr i64 %x, 28
%and1 = and i64 %shr, 15
%or = or i64 %and, %and1
ret i64 %or
; CHECK-LABEL: @test13
; CHECK: rlwinm 3, 3, 4, 24, 31
; CHECK: blr
}
define i64 @test14(i64 %x) #0 {
entry:
%shl = shl i64 %x, 4
%and = and i64 %shl, 240
%shr = lshr i64 %x, 28
%and1 = and i64 %shr, 15
%and2 = and i64 %x, -4294967296
%or = or i64 %and1, %and2
%or3 = or i64 %or, %and
ret i64 %or3
; CHECK-LABEL: @test14
; CHECK: rldicr [[REG1:[0-9]+]], 3, 0, 31
; CHECK: rlwimi [[REG1]], 3, 4, 24, 31
; CHECK: mr 3, [[REG1]]
; CHECK: blr
}
define i64 @test15(i64 %x) #0 {
entry:
%shl = shl i64 %x, 4
%and = and i64 %shl, 240
%shr = lshr i64 %x, 28
%and1 = and i64 %shr, 15
%and2 = and i64 %x, -256
%or = or i64 %and1, %and2
%or3 = or i64 %or, %and
ret i64 %or3
; CHECK-LABEL: @test15
; CHECK: rlwimi 3, 3, 4, 24, 31
; CHECK: blr
}
[PowerPC] Improve instruction selection bit-permuting operations (32-bit) The PowerPC backend, somewhat embarrassingly, did not generate an optimal-length sequence of instructions for a 32-bit bswap. While adding a pattern for the bswap intrinsic to fix this would not have been terribly difficult, doing so would not have addressed the real problem: we had been generating poor code for many bit-permuting operations (by which I mean things like byte-swap that permute the bits of one or more inputs around in various ways). Here are some initial steps toward solving this deficiency. Bit-permuting operations are represented, at the SDAG level, using ISD::ROTL, SHL, SRL, AND and OR (mostly with constant second operands). Looking back through these operations, we can build up a description of the bits in the resulting value in terms of bits of one or more input values (and constant zeros). For each bit, we compute the rotation amount from the original value, and then group consecutive (value, rotation factor) bits into groups. Groups sharing these attributes are then collected and sorted, and we can then instruction select the entire permutation using a combination of masked rotations (rlwinm), imm ands (andi/andis), and masked rotation inserts (rlwimi). The result is that instead of lowering an i32 bswap as: rlwinm 5, 3, 24, 16, 23 rlwinm 4, 3, 24, 0, 7 rlwimi 4, 3, 8, 8, 15 rlwimi 5, 3, 8, 24, 31 rlwimi 4, 5, 0, 16, 31 we now produce: rlwinm 4, 3, 8, 0, 31 rlwimi 4, 3, 24, 16, 23 rlwimi 4, 3, 24, 0, 7 and for the 'test6' example in the PowerPC/README.txt file: unsigned test6(unsigned x) { return ((x & 0x00FF0000) >> 16) | ((x & 0x000000FF) << 16); } we used to produce: lis 4, 255 rlwinm 3, 3, 16, 0, 31 ori 4, 4, 255 and 3, 3, 4 and now we produce: rlwinm 4, 3, 16, 24, 31 rlwimi 4, 3, 16, 8, 15 and, as a nice bonus, this fixes the FIXME in test/CodeGen/PowerPC/rlwimi-and.ll. This commit does not include instruction-selection for i64 operations, those will come later. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224318 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-16 05:51:41 +00:00
; Function Attrs: nounwind readnone
declare i32 @llvm.bswap.i32(i32) #0
[PowerPC] Improve instruction selection bit-permuting operations (64-bit) This is the second installment of improvements to instruction selection for "bit permutation" instruction sequences. r224318 added logic for instruction selection for 32-bit bit permutation sequences, and this adds lowering for 64-bit sequences. The 64-bit sequences are more complicated than the 32-bit ones because: a) the 64-bit versions of the 32-bit rotate-and-mask instructions work by replicating the lower 32-bits of the value-to-be-rotated into the upper 32 bits -- and integrating this into the cost modeling for the various bit group operations is non-trivial b) unlike the 32-bit instructions in 32-bit mode, the rotate-and-mask instructions cannot, in one instruction, specify the mask starting index, the mask ending index, and the rotation factor. Also, forming arbitrary 64-bit constants is more complicated than in 32-bit mode because the number of instructions necessary is value dependent. Plus, support for 'late masking' was added: it is sometimes more efficient to treat the overall value as if it had no mandatory zero bits when planning the bit-group insertions, and then mask them in at the very end. Unfortunately, as the structure of the bit groups is different in the two cases, the more feasible implementation technique was to generate both instruction sequences, and then pick the shorter one. And finally, we now generate reasonable code for i64 bswap: rldicl 5, 3, 16, 0 rldicl 4, 3, 8, 0 rldicl 6, 3, 24, 0 rldimi 4, 5, 8, 48 rldicl 5, 3, 32, 0 rldimi 4, 6, 16, 40 rldicl 6, 3, 48, 0 rldimi 4, 5, 24, 32 rldicl 5, 3, 56, 0 rldimi 4, 6, 40, 16 rldimi 4, 5, 48, 8 rldimi 4, 3, 56, 0 vs. what we used to produce: li 4, 255 rldicl 5, 3, 24, 40 rldicl 6, 3, 40, 24 rldicl 7, 3, 56, 8 sldi 8, 3, 8 sldi 10, 3, 24 sldi 12, 3, 40 rldicl 0, 3, 8, 56 sldi 9, 4, 32 sldi 11, 4, 40 sldi 4, 4, 48 andi. 5, 5, 65280 andis. 6, 6, 255 andis. 7, 7, 65280 sldi 3, 3, 56 and 8, 8, 9 and 4, 12, 4 and 9, 10, 11 or 6, 7, 6 or 5, 5, 0 or 3, 3, 4 or 7, 9, 8 or 4, 6, 5 or 3, 3, 7 or 3, 3, 4 which is 12 instructions, instead of 25, and seems optimal (at least in terms of code size). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225056 91177308-0d34-0410-b5e6-96231b3b80d8
2015-01-01 02:53:29 +00:00
declare i64 @llvm.bswap.i64(i64) #0
[PowerPC] Improve instruction selection bit-permuting operations (32-bit) The PowerPC backend, somewhat embarrassingly, did not generate an optimal-length sequence of instructions for a 32-bit bswap. While adding a pattern for the bswap intrinsic to fix this would not have been terribly difficult, doing so would not have addressed the real problem: we had been generating poor code for many bit-permuting operations (by which I mean things like byte-swap that permute the bits of one or more inputs around in various ways). Here are some initial steps toward solving this deficiency. Bit-permuting operations are represented, at the SDAG level, using ISD::ROTL, SHL, SRL, AND and OR (mostly with constant second operands). Looking back through these operations, we can build up a description of the bits in the resulting value in terms of bits of one or more input values (and constant zeros). For each bit, we compute the rotation amount from the original value, and then group consecutive (value, rotation factor) bits into groups. Groups sharing these attributes are then collected and sorted, and we can then instruction select the entire permutation using a combination of masked rotations (rlwinm), imm ands (andi/andis), and masked rotation inserts (rlwimi). The result is that instead of lowering an i32 bswap as: rlwinm 5, 3, 24, 16, 23 rlwinm 4, 3, 24, 0, 7 rlwimi 4, 3, 8, 8, 15 rlwimi 5, 3, 8, 24, 31 rlwimi 4, 5, 0, 16, 31 we now produce: rlwinm 4, 3, 8, 0, 31 rlwimi 4, 3, 24, 16, 23 rlwimi 4, 3, 24, 0, 7 and for the 'test6' example in the PowerPC/README.txt file: unsigned test6(unsigned x) { return ((x & 0x00FF0000) >> 16) | ((x & 0x000000FF) << 16); } we used to produce: lis 4, 255 rlwinm 3, 3, 16, 0, 31 ori 4, 4, 255 and 3, 3, 4 and now we produce: rlwinm 4, 3, 16, 24, 31 rlwimi 4, 3, 16, 8, 15 and, as a nice bonus, this fixes the FIXME in test/CodeGen/PowerPC/rlwimi-and.ll. This commit does not include instruction-selection for i64 operations, those will come later. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224318 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-16 05:51:41 +00:00
attributes #0 = { nounwind readnone }
Reference in New Issue Copy Permalink