maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/ARM/subtarget-no-movt.ll

98 lines
3.2 KiB
LLVM
Raw Normal View History

[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - | \
; RUN: FileCheck -check-prefixes=CHECK,NO-OPTION,NO-OPTION-COMMON %s
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - -mattr=-no-movt | \
; RUN: FileCheck -check-prefixes=CHECK,USE-MOVT,USE-MOVT-COMMON %s
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - -mattr=+no-movt | \
; RUN: FileCheck -check-prefixes=CHECK,NO-USE-MOVT,NO-USE-MOVT-COMMON %s
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - -O0 | \
; RUN: FileCheck -check-prefixes=CHECK,NO-OPTION-O0,NO-OPTION-COMMON %s
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - -O0 -mattr=-no-movt | \
; RUN: FileCheck -check-prefixes=CHECK,USE-MOVT-O0,USE-MOVT-COMMON %s
; RUN: llc -mcpu=cortex-a8 -relocation-model=static %s -o - -O0 -mattr=+no-movt | \
; RUN: FileCheck -check-prefixes=CHECK,NO-USE-MOVT-O0,NO-USE-MOVT-COMMON %s
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
target triple = "thumb-apple-darwin"
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-OPTION-COMMON-LABEL: {{_?}}foo0
; NO-OPTION-COMMON: ldr [[R0:r[0-9]+]], [[L0:.*]]
; NO-OPTION-COMMON: [[L0]]:
; NO-OPTION-COMMON: .long 2296237089
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; USE-MOVT-COMMON-LABEL: {{_?}}foo0
; USE-MOVT-COMMON: movw [[R0:r[0-9]+]], #52257
; USE-MOVT-COMMON: movt [[R0]], #35037
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-USE-MOVT-COMMON-LABEL: {{_?}}foo0
; NO-USE-MOVT-COMMON: ldr [[R0:r[0-9]+]], [[L0:.*]]
; NO-USE-MOVT-COMMON: [[L0]]:
; NO-USE-MOVT-COMMON: .long 2296237089
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
define i32 @foo0(i32 %a) #0 {
%1 = xor i32 -1998730207, %a
ret i32 %1
}
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-OPTION-COMMON-LABEL: {{_?}}foo1
; NO-OPTION-COMMON: movw [[R0:r[0-9]+]], #52257
; NO-OPTION-COMMON: movt [[R0]], #35037
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; USE-MOVT-COMMON-LABEL: {{_?}}foo1
; USE-MOVT-COMMON: movw [[R0:r[0-9]+]], #52257
; USE-MOVT-COMMON: movt [[R0]], #35037
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-USE-MOVT-COMMON-LABEL: {{_?}}foo1
; NO-USE-MOVT-COMMON: ldr [[R0:r[0-9]+]], [[L0:.*]]
; NO-USE-MOVT-COMMON: [[L0]]:
; NO-USE-MOVT-COMMON: .long 2296237089
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
define i32 @foo1(i32 %a) {
%1 = xor i32 -1998730207, %a
ret i32 %1
}
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-OPTION-COMMON-LABEL: {{_?}}foo2
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
; NO-OPTION: mov.w [[R0:r[0-9]+]], #-536813568
; NO-OPTION-O0: movw [[R0:r[0-9]+]], #57344
; NO-OPTION-O0: movt [[R0]], #57344
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; USE-MOVT-COMMON-LABEL: {{_?}}foo2
; USE-MOVT: mov.w [[R0:r[0-9]+]], #-536813568
; USE-MOVT-O0: movw [[R0:r[0-9]+]], #57344
; USE-MOVT-O0: movt [[R0]], #57344
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
; NO-USE-MOVT-COMMON-LABEL: {{_?}}foo2
; NO-USE-MOVT: mov.w [[R0:r[0-9]+]], #-536813568
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
; NO-USE-MOVT-O0: ldr [[R0:r[0-9]+]], [[L0:.*]]
; NO-USE-MOVT-O0: [[L0]]:
; NO-USE-MOVT-O0: .long 3758153728 @ 0xe000e000
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
[ARM] Constant Materialize: imms with specific value can be encoded into mov.w Summary: Thumb2 supports encoding immediates with specific patterns into mov.w by splatting the low 8 bits into other bytes. I'm resubmitting this patch. The test case in the original commit r277610 does not specify triple, so builds with differnt default triple will have different output. This patch fixed trile as thumb-darwin-apple. Reviewers: john.brawn, jmolloy, bruno Subscribers: jmolloy, aemerson, rengolin, samparker, llvm-commits Differential Revision: https://reviews.llvm.org/D23090 llvm-svn: 277865
2016-08-06 04:58:29 +08:00
define i32 @foo2() {
%1 = load i32, i32* inttoptr (i32 -536813568 to i32*) ; load from 0xe000e000
ret i32 %1
}
[ARM] Define a subtarget feature that is used to avoid using movt/movw pairs for 32-bit immediates. This change is needed to avoid emitting movt/movw pairs when doing LTO and do so on a per-function basis. Out-of-tree projects currently using cl::opt option -arm-use-movt=0 or false to avoid emitting movt/movw pairs should make changes to add subtarget feature "+no-movt" (see the changes made to clang in r242368). rdar://problem/21529937 Differential Revision: http://reviews.llvm.org/D11026 llvm-svn: 242369
2015-07-16 08:58:23 +08:00
attributes #0 = { "target-features"="+no-movt" }
[ARM] Thumb2: ConstantMaterializationCost Constants can also be materialised using the negated value and a MVN, and this case seem to have been missed for Thumb2. To check the constant materialisation costs, we now call getT2SOImmVal twice, once for the original constant and then also for its negated value, and this function checks if the constant can both be splatted or rotated. This was revealed by a test that optimises for minsize: instead of a LDR literal pool load and having a literal pool entry, just a MVN with an immediate is smaller (and also faster). Differential Revision: https://reviews.llvm.org/D57327 llvm-svn: 352737
2019-01-31 16:38:06 +08:00
define hidden i32 @no_litpool() minsize optsize {
; CHECK-LABEL: no_litpool:
; CHECK: mov.w r{{.}}, #65536
; CHECK: mov.w r{{.}}, #-134217728
; CHECK: mvn r{{.}}, #-134217728
entry:
%call0 = tail call i32 @eat_const(i32 65536)
%call1 = tail call i32 @eat_const(i32 -134217728)
%call2 = tail call i32 @eat_const(i32 134217727)
ret i32 %call2
}
define hidden i32 @litpool() minsize optsize {
; CHECK-LABEL: litpool:
; CHECK: ldr r0, {{.*}}LCPI{{.*}}
; CHECK-NEXT: b.w {{.*}}eat_const
entry:
%call1 = tail call i32 @eat_const(i32 8388601)
ret i32 %call1
}
declare dso_local i32 @eat_const(i32)