RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-23 13:02:52 +02:00
Code Issues Projects Releases Wiki Activity
6333fd2b12
llvm-mirror/test/CodeGen/PowerPC/code-align.ll

111 lines
3.1 KiB
LLVM
Raw Normal View History

[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; RUN: llc -mcpu=ppc64 < %s | FileCheck %s -check-prefix=GENERIC
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; RUN: llc -mcpu=970 < %s | FileCheck %s -check-prefix=PWR
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; RUN: llc -mcpu=a2 < %s | FileCheck %s -check-prefix=BASIC
; RUN: llc -mcpu=e500mc < %s | FileCheck %s -check-prefix=BASIC
; RUN: llc -mcpu=e5500 < %s | FileCheck %s -check-prefix=BASIC
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; RUN: llc -mcpu=pwr4 < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr5 < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr5x < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr6 < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr6x < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr7 < %s | FileCheck %s -check-prefix=PWR
; RUN: llc -mcpu=pwr8 < %s | FileCheck %s -check-prefix=PWR
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
; Function Attrs: nounwind readnone
define signext i32 @foo(i32 signext %x) #0 {
entry:
%mul = shl nsw i32 %x, 1
ret i32 %mul
; GENERIC-LABEL: .globl foo
; BASIC-LABEL: .globl foo
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR-LABEL: .globl foo
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; GENERIC: .align 2
; BASIC: .align 4
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR: .align 4
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; GENERIC: @foo
; BASIC: @foo
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR: @foo
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
}
; Function Attrs: nounwind
define void @loop(i32 signext %x, i32* nocapture %a) #1 {
entry:
br label %vector.body
; GENERIC-LABEL: @loop
; BASIC-LABEL: @loop
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR-LABEL: @loop
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; GENERIC: mtctr
; BASIC: mtctr
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR: mtctr
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; GENERIC-NOT: .align
; BASIC: .align 4
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR: .align 4
[PowerPC] Prepare loops for pre-increment loads/stores PowerPC supports pre-increment load/store instructions (except for Altivec/VSX vector load/stores). Using these on embedded cores can be very important, but most loops are not naturally set up to use them. We can often change that, however, by placing loops into a non-canonical form. Generically, this means transforming loops like this: for (int i = 0; i < n; ++i) array[i] = c; to look like this: T *p = array[-1]; for (int i = 0; i < n; ++i) *++p = c; the key point is that addresses accessed are pulled into dedicated PHIs and "pre-decremented" in the loop preheader. This allows the use of pre-increment load/store instructions without loop peeling. A target-specific late IR-level pass (running post-LSR), PPCLoopPreIncPrep, is introduced to perform this transformation. I've used this code out-of-tree for generating code for the PPC A2 for over a year. Somewhat to my surprise, running the test suite + externals on a P7 with this transformation enabled showed no performance regressions, and one speedup: External/SPEC/CINT2006/483.xalancbmk/483.xalancbmk -2.32514% +/- 1.03736% So I'm going to enable it on everything for now. I was surprised by this because, on the POWER cores, these pre-increment load/store instructions are cracked (and, thus, harder to schedule effectively). But seeing no regressions, and feeling that it is generally easier to split instructions apart late than it is to combine them late, this might be the better approach regardless. In the future, we might want to integrate this functionality into LSR (but currently LSR does not create new PHI nodes, so (for that and other reasons) significant work would need to be done). llvm-svn: 228328
2015-02-05 19:43:00 +01:00
; GENERIC: lwzu
; BASIC: lwzu
; PWR: lwzu
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
; GENERIC: bdnz
; BASIC: bdnz
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; PWR: bdnz
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
vector.body: ; preds = %vector.body, %entry
%index = phi i64 [ 0, %entry ], [ %index.next, %vector.body ]
%induction45 = or i64 %index, 1
%0 = getelementptr inbounds i32* %a, i64 %index
%1 = getelementptr inbounds i32* %a, i64 %induction45
%2 = load i32* %0, align 4
%3 = load i32* %1, align 4
%4 = add nsw i32 %2, 4
%5 = add nsw i32 %3, 4
[PowerPC] Prepare loops for pre-increment loads/stores PowerPC supports pre-increment load/store instructions (except for Altivec/VSX vector load/stores). Using these on embedded cores can be very important, but most loops are not naturally set up to use them. We can often change that, however, by placing loops into a non-canonical form. Generically, this means transforming loops like this: for (int i = 0; i < n; ++i) array[i] = c; to look like this: T *p = array[-1]; for (int i = 0; i < n; ++i) *++p = c; the key point is that addresses accessed are pulled into dedicated PHIs and "pre-decremented" in the loop preheader. This allows the use of pre-increment load/store instructions without loop peeling. A target-specific late IR-level pass (running post-LSR), PPCLoopPreIncPrep, is introduced to perform this transformation. I've used this code out-of-tree for generating code for the PPC A2 for over a year. Somewhat to my surprise, running the test suite + externals on a P7 with this transformation enabled showed no performance regressions, and one speedup: External/SPEC/CINT2006/483.xalancbmk/483.xalancbmk -2.32514% +/- 1.03736% So I'm going to enable it on everything for now. I was surprised by this because, on the POWER cores, these pre-increment load/store instructions are cracked (and, thus, harder to schedule effectively). But seeing no regressions, and feeling that it is generally easier to split instructions apart late than it is to combine them late, this might be the better approach regardless. In the future, we might want to integrate this functionality into LSR (but currently LSR does not create new PHI nodes, so (for that and other reasons) significant work would need to be done). llvm-svn: 228328
2015-02-05 19:43:00 +01:00
%6 = mul nsw i32 %4, 3
%7 = mul nsw i32 %5, 3
store i32 %6, i32* %0, align 4
store i32 %7, i32* %1, align 4
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
%index.next = add i64 %index, 2
[PowerPC] Prepare loops for pre-increment loads/stores PowerPC supports pre-increment load/store instructions (except for Altivec/VSX vector load/stores). Using these on embedded cores can be very important, but most loops are not naturally set up to use them. We can often change that, however, by placing loops into a non-canonical form. Generically, this means transforming loops like this: for (int i = 0; i < n; ++i) array[i] = c; to look like this: T *p = array[-1]; for (int i = 0; i < n; ++i) *++p = c; the key point is that addresses accessed are pulled into dedicated PHIs and "pre-decremented" in the loop preheader. This allows the use of pre-increment load/store instructions without loop peeling. A target-specific late IR-level pass (running post-LSR), PPCLoopPreIncPrep, is introduced to perform this transformation. I've used this code out-of-tree for generating code for the PPC A2 for over a year. Somewhat to my surprise, running the test suite + externals on a P7 with this transformation enabled showed no performance regressions, and one speedup: External/SPEC/CINT2006/483.xalancbmk/483.xalancbmk -2.32514% +/- 1.03736% So I'm going to enable it on everything for now. I was surprised by this because, on the POWER cores, these pre-increment load/store instructions are cracked (and, thus, harder to schedule effectively). But seeing no regressions, and feeling that it is generally easier to split instructions apart late than it is to combine them late, this might be the better approach regardless. In the future, we might want to integrate this functionality into LSR (but currently LSR does not create new PHI nodes, so (for that and other reasons) significant work would need to be done). llvm-svn: 228328
2015-02-05 19:43:00 +01:00
%8 = icmp eq i64 %index.next, 2048
br i1 %8, label %for.end, label %vector.body
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
for.end: ; preds = %vector.body
ret void
}
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
; Function Attrs: nounwind
define void @sloop(i32 signext %x, i32* nocapture %a) #1 {
entry:
br label %for.body
; GENERIC-LABEL: @sloop
; BASIC-LABEL: @sloop
; PWR-LABEL: @sloop
; GENERIC: mtctr
; BASIC: mtctr
; PWR: mtctr
; GENERIC-NOT: .align
; BASIC: .align 4
; PWR: .align 5
; GENERIC: bdnz
; BASIC: bdnz
; PWR: bdnz
for.body: ; preds = %for.body, %entry
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%arrayidx = getelementptr inbounds i32* %a, i64 %indvars.iv
%0 = load i32* %arrayidx, align 4
%add = add nsw i32 %0, 4
[PowerPC] Prepare loops for pre-increment loads/stores PowerPC supports pre-increment load/store instructions (except for Altivec/VSX vector load/stores). Using these on embedded cores can be very important, but most loops are not naturally set up to use them. We can often change that, however, by placing loops into a non-canonical form. Generically, this means transforming loops like this: for (int i = 0; i < n; ++i) array[i] = c; to look like this: T *p = array[-1]; for (int i = 0; i < n; ++i) *++p = c; the key point is that addresses accessed are pulled into dedicated PHIs and "pre-decremented" in the loop preheader. This allows the use of pre-increment load/store instructions without loop peeling. A target-specific late IR-level pass (running post-LSR), PPCLoopPreIncPrep, is introduced to perform this transformation. I've used this code out-of-tree for generating code for the PPC A2 for over a year. Somewhat to my surprise, running the test suite + externals on a P7 with this transformation enabled showed no performance regressions, and one speedup: External/SPEC/CINT2006/483.xalancbmk/483.xalancbmk -2.32514% +/- 1.03736% So I'm going to enable it on everything for now. I was surprised by this because, on the POWER cores, these pre-increment load/store instructions are cracked (and, thus, harder to schedule effectively). But seeing no regressions, and feeling that it is generally easier to split instructions apart late than it is to combine them late, this might be the better approach regardless. In the future, we might want to integrate this functionality into LSR (but currently LSR does not create new PHI nodes, so (for that and other reasons) significant work would need to be done). llvm-svn: 228328
2015-02-05 19:43:00 +01:00
%mul = mul nsw i32 %add, 3
store i32 %mul, i32* %arrayidx, align 4
[PowerPC/BlockPlacement] Allow target to provide a per-loop alignment preference The existing code provided for specifying a global loop alignment preference. However, the preferred loop alignment might depend on the loop itself. For recent POWER cores, loops between 5 and 8 instructions should have 32-byte alignment (while the others are better with 16-byte alignment) so that the entire loop will fit in one i-cache line. To support this, getPrefLoopAlignment has been made virtual, and can be provided with an optional MachineLoop* so the target can inspect the loop before answering the query. The default behavior, as before, is to return the value set with setPrefLoopAlignment. MachineBlockPlacement now queries the target for each loop instead of only once per function. There should be no functional change for other targets. llvm-svn: 225117
2015-01-03 18:58:24 +01:00
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, 2048
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body
ret void
}
[PowerPC] Use 16-byte alignment for modern cores for functions/loops Most modern PowerPC cores prefer that functions and loops start on 16-byte-aligned boundaries (*), so instruct block placement, etc. to make this happen. The branch selector has also been adjusted so account for the extra nops that might now be inserted before loop headers. (*) Some cores actually prefer other alignments for small loops, but that will be addressed in a follow-up commit. llvm-svn: 225115
2015-01-03 15:58:25 +01:00
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind }
Reference in New Issue Copy Permalink