RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
Code Issues Projects Releases Wiki Activity
d382d6f3fc
llvm-mirror/test/CodeGen/Thumb2/float-ops.ll

292 lines
6.2 KiB
LLVM
Raw Normal View History

[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; RUN: llc < %s -mtriple=thumbv7-none-eabi -mcpu=cortex-m3 | FileCheck %s -check-prefix=CHECK -check-prefix=NONE
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218747
2014-10-01 11:02:17 +02:00
; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m4 | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=SP -check-prefix=VFP4-ALL
; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m7 | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP -check-prefix=FP-ARMv8
; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-a8 | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP -check-prefix=VFP4-ALL -check-prefix=VFP4-DP
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
define float @add_f(float %a, float %b) {
entry:
; CHECK-LABEL: add_f:
; NONE: bl __aeabi_fadd
; HARD: vadd.f32 s0, s0, s1
%0 = fadd float %a, %b
ret float %0
}
define double @add_d(double %a, double %b) {
entry:
; CHECK-LABEL: add_d:
; NONE: bl __aeabi_dadd
; SP: bl __aeabi_dadd
; DP: vadd.f64 d0, d0, d1
%0 = fadd double %a, %b
ret double %0
}
define float @sub_f(float %a, float %b) {
entry:
; CHECK-LABEL: sub_f:
; NONE: bl __aeabi_fsub
; HARD: vsub.f32 s
%0 = fsub float %a, %b
ret float %0
}
define double @sub_d(double %a, double %b) {
entry:
; CHECK-LABEL: sub_d:
; NONE: bl __aeabi_dsub
; SP: bl __aeabi_dsub
; DP: vsub.f64 d0, d0, d1
%0 = fsub double %a, %b
ret double %0
}
define float @mul_f(float %a, float %b) {
entry:
; CHECK-LABEL: mul_f:
; NONE: bl __aeabi_fmul
; HARD: vmul.f32 s
%0 = fmul float %a, %b
ret float %0
}
define double @mul_d(double %a, double %b) {
entry:
; CHECK-LABEL: mul_d:
; NONE: bl __aeabi_dmul
; SP: bl __aeabi_dmul
; DP: vmul.f64 d0, d0, d1
%0 = fmul double %a, %b
ret double %0
}
define float @div_f(float %a, float %b) {
entry:
; CHECK-LABEL: div_f:
; NONE: bl __aeabi_fdiv
; HARD: vdiv.f32 s
%0 = fdiv float %a, %b
ret float %0
}
define double @div_d(double %a, double %b) {
entry:
; CHECK-LABEL: div_d:
; NONE: bl __aeabi_ddiv
; SP: bl __aeabi_ddiv
; DP: vdiv.f64 d0, d0, d1
%0 = fdiv double %a, %b
ret double %0
}
define float @rem_f(float %a, float %b) {
entry:
; CHECK-LABEL: rem_f:
; NONE: bl fmodf
Revert "CodeGen: ensure that libcalls are always AAPCS CC" This reverts SVN r280683. Revert until I figure out why this is breaking lli tests. llvm-svn: 280778
2016-09-07 05:17:19 +02:00
; HARD: b fmodf
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
%0 = frem float %a, %b
ret float %0
}
define double @rem_d(double %a, double %b) {
entry:
; CHECK-LABEL: rem_d:
; NONE: bl fmod
Revert "CodeGen: ensure that libcalls are always AAPCS CC" This reverts SVN r280683. Revert until I figure out why this is breaking lli tests. llvm-svn: 280778
2016-09-07 05:17:19 +02:00
; HARD: b fmod
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
%0 = frem double %a, %b
ret double %0
}
define float @load_f(float* %a) {
entry:
; CHECK-LABEL: load_f:
; NONE: ldr r0, [r0]
; HARD: vldr s0, [r0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%0 = load float, float* %a, align 4
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
ret float %0
}
define double @load_d(double* %a) {
entry:
; CHECK-LABEL: load_d:
ARMLoadStoreOptimizer: Fix errata 602117 handling and make testcase actually test for it This fixes PR23912 Differential Revision: http://reviews.llvm.org/D10620 llvm-svn: 240582
2015-06-24 22:03:27 +02:00
; NONE: ldm r0, {r0, r1}
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; HARD: vldr d0, [r0]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-27 22:17:42 +01:00
%0 = load double, double* %a, align 8
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
ret double %0
}
define void @store_f(float* %a, float %b) {
entry:
; CHECK-LABEL: store_f:
; NONE: str r1, [r0]
; HARD: vstr s0, [r0]
store float %b, float* %a, align 4
ret void
}
define void @store_d(double* %a, double %b) {
entry:
; CHECK-LABEL: store_d:
ARM: Thumb2 LDRD/STRD supports independent input/output regs The existing code would unnecessarily break LDRD/STRD apart with non-adjacent registers, on thumb2 this is not necessary. Ideally on thumb2 we shouldn't match for ldrd/strd pre-regalloc anymore as there is not reason to set register hints anymore, changing that is something for a future patch however. Differential Revision: http://reviews.llvm.org/D9694 Recommiting after the revert in r238821, the buildbot still failed with the patch removed so there seems to be another reason for the breakage. llvm-svn: 238935
2015-06-03 18:30:24 +02:00
; NONE: strd r2, r3, [r0]
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; HARD: vstr d0, [r0]
store double %b, double* %a, align 8
ret void
}
define double @f_to_d(float %a) {
; CHECK-LABEL: f_to_d:
; NONE: bl __aeabi_f2d
; SP: bl __aeabi_f2d
; DP: vcvt.f64.f32 d0, s0
%1 = fpext float %a to double
ret double %1
}
define float @d_to_f(double %a) {
; CHECK-LABEL: d_to_f:
; NONE: bl __aeabi_d2f
; SP: bl __aeabi_d2f
; DP: vcvt.f32.f64 s0, d0
%1 = fptrunc double %a to float
ret float %1
}
define i32 @f_to_si(float %a) {
; CHECK-LABEL: f_to_si:
; NONE: bl __aeabi_f2iz
; HARD: vcvt.s32.f32 s0, s0
; HARD: vmov r0, s0
%1 = fptosi float %a to i32
ret i32 %1
}
define i32 @d_to_si(double %a) {
; CHECK-LABEL: d_to_si:
; NONE: bl __aeabi_d2iz
; SP: vmov r0, r1, d0
; SP: bl __aeabi_d2iz
; DP: vcvt.s32.f64 s0, d0
; DP: vmov r0, s0
%1 = fptosi double %a to i32
ret i32 %1
}
define i32 @f_to_ui(float %a) {
; CHECK-LABEL: f_to_ui:
; NONE: bl __aeabi_f2uiz
; HARD: vcvt.u32.f32 s0, s0
; HARD: vmov r0, s0
%1 = fptoui float %a to i32
ret i32 %1
}
define i32 @d_to_ui(double %a) {
; CHECK-LABEL: d_to_ui:
; NONE: bl __aeabi_d2uiz
; SP: vmov r0, r1, d0
; SP: bl __aeabi_d2uiz
; DP: vcvt.u32.f64 s0, d0
; DP: vmov r0, s0
%1 = fptoui double %a to i32
ret i32 %1
}
define float @si_to_f(i32 %a) {
; CHECK-LABEL: si_to_f:
; NONE: bl __aeabi_i2f
; HARD: vcvt.f32.s32 s0, s0
%1 = sitofp i32 %a to float
ret float %1
}
define double @si_to_d(i32 %a) {
; CHECK-LABEL: si_to_d:
; NONE: bl __aeabi_i2d
; SP: bl __aeabi_i2d
; DP: vcvt.f64.s32 d0, s0
%1 = sitofp i32 %a to double
ret double %1
}
define float @ui_to_f(i32 %a) {
; CHECK-LABEL: ui_to_f:
; NONE: bl __aeabi_ui2f
; HARD: vcvt.f32.u32 s0, s0
%1 = uitofp i32 %a to float
ret float %1
}
define double @ui_to_d(i32 %a) {
; CHECK-LABEL: ui_to_d:
; NONE: bl __aeabi_ui2d
; SP: bl __aeabi_ui2d
; DP: vcvt.f64.u32 d0, s0
%1 = uitofp i32 %a to double
ret double %1
}
define float @bitcast_i_to_f(i32 %a) {
; CHECK-LABEL: bitcast_i_to_f:
; NONE-NOT: mov
; HARD: vmov s0, r0
%1 = bitcast i32 %a to float
ret float %1
}
define double @bitcast_i_to_d(i64 %a) {
; CHECK-LABEL: bitcast_i_to_d:
; NONE-NOT: mov
; HARD: vmov d0, r0, r1
%1 = bitcast i64 %a to double
ret double %1
}
define i32 @bitcast_f_to_i(float %a) {
; CHECK-LABEL: bitcast_f_to_i:
; NONE-NOT: mov
; HARD: vmov r0, s0
%1 = bitcast float %a to i32
ret i32 %1
}
define i64 @bitcast_d_to_i(double %a) {
; CHECK-LABEL: bitcast_d_to_i:
; NONE-NOT: mov
; HARD: vmov r0, r1, d0
%1 = bitcast double %a to i64
ret i64 %1
}
define float @select_f(float %a, float %b, i1 %c) {
; CHECK-LABEL: select_f:
[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently This is essentially a recommit of r285893, but with a correctness fix. The problem of the original commit was that this: bic r5, r7, #31 cbz r5, .LBB2_10 got rewritten into: lsrs r5, r7, #5 beq .LBB2_10 The result in destination register r5 is not the same and this is incorrect when r5 is not dead. So this fix includes checking the uses of the AND destination register. And also, compared to the original commit, some regression tests didn't need changing anymore because of this extra check. For completeness, this was the original commit message: For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)). 1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS. 2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS. 3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS). 4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask. 1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win. Differential Revision: https://reviews.llvm.org/D27761 llvm-svn: 289794
2016-12-15 10:38:59 +01:00
; NONE: lsls r2, r2, #31
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; NONE: moveq r0, r1
[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently This is essentially a recommit of r285893, but with a correctness fix. The problem of the original commit was that this: bic r5, r7, #31 cbz r5, .LBB2_10 got rewritten into: lsrs r5, r7, #5 beq .LBB2_10 The result in destination register r5 is not the same and this is incorrect when r5 is not dead. So this fix includes checking the uses of the AND destination register. And also, compared to the original commit, some regression tests didn't need changing anymore because of this extra check. For completeness, this was the original commit message: For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)). 1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS. 2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS. 3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS). 4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask. 1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win. Differential Revision: https://reviews.llvm.org/D27761 llvm-svn: 289794
2016-12-15 10:38:59 +01:00
; HARD: lsls r0, r0, #31
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218747
2014-10-01 11:02:17 +02:00
; VFP4-ALL: vmovne.f32 s1, s0
; VFP4-ALL: vmov.f32 s0, s1
; FP-ARMv8: vseleq.f32 s0, s1, s0
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
%1 = select i1 %c, float %a, float %b
ret float %1
}
define double @select_d(double %a, double %b, i1 %c) {
; CHECK-LABEL: select_d:
[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently This is essentially a recommit of r285893, but with a correctness fix. The problem of the original commit was that this: bic r5, r7, #31 cbz r5, .LBB2_10 got rewritten into: lsrs r5, r7, #5 beq .LBB2_10 The result in destination register r5 is not the same and this is incorrect when r5 is not dead. So this fix includes checking the uses of the AND destination register. And also, compared to the original commit, some regression tests didn't need changing anymore because of this extra check. For completeness, this was the original commit message: For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)). 1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS. 2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS. 3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS). 4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask. 1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win. Differential Revision: https://reviews.llvm.org/D27761 llvm-svn: 289794
2016-12-15 10:38:59 +01:00
; NONE: ldr{{(.w)?}} [[REG:r[0-9]+]], [sp]
; NONE ands [[REG]], [[REG]], #1
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; NONE: moveq r0, r2
; NONE: moveq r1, r3
Revert "[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently" This reverts commit r285893. It caused (probably) http://lab.llvm.org:8011/builders/clang-cmake-thumbv7-a15-full-sh/builds/83 . llvm-svn: 285912
2016-11-03 15:08:01 +01:00
; SP: ands r0, r0, #1
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
; SP-DAG: vmov [[ALO:r[0-9]+]], [[AHI:r[0-9]+]], d0
; SP-DAG: vmov [[BLO:r[0-9]+]], [[BHI:r[0-9]+]], d1
; SP: itt ne
; SP-DAG: movne [[BLO]], [[ALO]]
; SP-DAG: movne [[BHI]], [[AHI]]
; SP: vmov d0, [[BLO]], [[BHI]]
[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently This is essentially a recommit of r285893, but with a correctness fix. The problem of the original commit was that this: bic r5, r7, #31 cbz r5, .LBB2_10 got rewritten into: lsrs r5, r7, #5 beq .LBB2_10 The result in destination register r5 is not the same and this is incorrect when r5 is not dead. So this fix includes checking the uses of the AND destination register. And also, compared to the original commit, some regression tests didn't need changing anymore because of this extra check. For completeness, this was the original commit message: For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)). 1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS. 2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS. 3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS). 4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask. 1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win. Differential Revision: https://reviews.llvm.org/D27761 llvm-svn: 289794
2016-12-15 10:38:59 +01:00
; DP: lsls r0, r0, #31
[ARM] Add support for Cortex-M7, FPv5-SP and FPv5-DP (LLVM) The Cortex-M7 has 3 options for its FPU: none, FPv5-SP-D16 and FPv5-DP-D16. FPv5 has the same instructions as FP-ARMv8, so it can be modelled using the same target feature, and all double-precision operations are already disabled by the fp-only-sp target features. llvm-svn: 218747
2014-10-01 11:02:17 +02:00
; VFP4-DP: vmovne.f64 d1, d0
; VFP4-DP: vmov.f64 d0, d1
; FP-ARMV8: vseleq.f64 d0, d1, d0
[ARM] Enable DP copy, load and store instructions for FPv4-SP The FPv4-SP floating-point unit is generally referred to as single-precision only, but it does have double-precision registers and load, store and GPR<->DPR move instructions which operate on them. This patch enables the use of these registers, the main advantage of which is that we now comply with the AAPCS-VFP calling convention. This partially reverts r209650, which added some AAPCS-VFP support, but did not handle return values or alignment of double arguments in registers. This patch also adds tests for Thumb2 code generation for floating-point instructions and intrinsics, which previously only existed for ARM. llvm-svn: 216172
2014-08-21 14:50:31 +02:00
%1 = select i1 %c, double %a, double %b
ret double %1
}
Reference in New Issue Copy Permalink