RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-30 23:42:52 +01:00
Code Issues Projects Releases Wiki Activity
54f125fb4b
llvm-mirror/test/CodeGen/ARM/reg_sequence.ll

349 lines
17 KiB
LLVM
Raw Normal View History

Try to make these tests more portable. llvm-svn: 164320
2012-09-20 23:35:21 +02:00
; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 | FileCheck %s
; RUN: llc < %s -mtriple=arm-apple-ios -mcpu=cortex-a8 -regalloc=basic | FileCheck %s
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; Implementing vld / vst as REG_SEQUENCE eliminates the extra vmov's.
%struct.int16x8_t = type { <8 x i16> }
%struct.int32x4_t = type { <4 x i32> }
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
%struct.__neon_int8x8x2_t = type { <8 x i8>, <8 x i8> }
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%struct.__neon_int8x8x3_t = type { <8 x i8>, <8 x i8>, <8 x i8> }
%struct.__neon_int16x8x2_t = type { <8 x i16>, <8 x i16> }
%struct.__neon_int32x4x2_t = type { <4 x i32>, <4 x i32> }
Remove arm_apcscc from the test files. It is the default and doing this matches what llvm-gcc and clang now produce. llvm-svn: 106221
2010-06-17 17:18:27 +02:00
define void @t1(i16* %i_ptr, i16* %o_ptr, %struct.int32x4_t* nocapture %vT0ptr, %struct.int32x4_t* nocapture %vT1ptr) nounwind {
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
entry:
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t1:
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vld1.16
; CHECK-NOT: vmov d
; CHECK: vmovl.s16
; CHECK: vshrn.i32
; CHECK: vshrn.i32
; CHECK-NOT: vmov d
; CHECK-NEXT: vst1.16
%0 = getelementptr inbounds %struct.int32x4_t* %vT0ptr, i32 0, i32 0 ; <<4 x i32>*> [#uses=1]
%1 = load <4 x i32>* %0, align 16 ; <<4 x i32>> [#uses=1]
%2 = getelementptr inbounds %struct.int32x4_t* %vT1ptr, i32 0, i32 0 ; <<4 x i32>*> [#uses=1]
%3 = load <4 x i32>* %2, align 16 ; <<4 x i32>> [#uses=1]
%4 = bitcast i16* %i_ptr to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%5 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %4, i32 1) ; <<8 x i16>> [#uses=1]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%6 = bitcast <8 x i16> %5 to <2 x double> ; <<2 x double>> [#uses=2]
%7 = extractelement <2 x double> %6, i32 0 ; <double> [#uses=1]
%8 = bitcast double %7 to <4 x i16> ; <<4 x i16>> [#uses=1]
Replace some NEON vmovl intrinsic that I missed earlier. llvm-svn: 111696
2010-08-21 01:22:43 +02:00
%9 = sext <4 x i16> %8 to <4 x i32> ; <<4 x i32>> [#uses=1]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%10 = extractelement <2 x double> %6, i32 1 ; <double> [#uses=1]
%11 = bitcast double %10 to <4 x i16> ; <<4 x i16>> [#uses=1]
Replace some NEON vmovl intrinsic that I missed earlier. llvm-svn: 111696
2010-08-21 01:22:43 +02:00
%12 = sext <4 x i16> %11 to <4 x i32> ; <<4 x i32>> [#uses=1]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%13 = mul <4 x i32> %1, %9 ; <<4 x i32>> [#uses=1]
%14 = mul <4 x i32> %3, %12 ; <<4 x i32>> [#uses=1]
%15 = tail call <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32> %13, <4 x i32> <i32 -12, i32 -12, i32 -12, i32 -12>) ; <<4 x i16>> [#uses=1]
%16 = tail call <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32> %14, <4 x i32> <i32 -12, i32 -12, i32 -12, i32 -12>) ; <<4 x i16>> [#uses=1]
%17 = shufflevector <4 x i16> %15, <4 x i16> %16, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> ; <<8 x i16>> [#uses=1]
%18 = bitcast i16* %o_ptr to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst1.v8i16(i8* %18, <8 x i16> %17, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
ret void
}
Remove arm_apcscc from the test files. It is the default and doing this matches what llvm-gcc and clang now produce. llvm-svn: 106221
2010-06-17 17:18:27 +02:00
define void @t2(i16* %i_ptr, i16* %o_ptr, %struct.int16x8_t* nocapture %vT0ptr, %struct.int16x8_t* nocapture %vT1ptr) nounwind {
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
entry:
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t2:
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vld1.16
Fix some latency computation bugs: if the use is not a machine opcode do not just return zero. llvm-svn: 105061
2010-05-29 01:26:21 +02:00
; CHECK-NOT: vmov
Change ARM scheduling default to list-hybrid if the target supports floating point instructions (and is not using soft float). llvm-svn: 104307
2010-05-21 02:43:17 +02:00
; CHECK: vmul.i16
Correct some load / store instruction itinerary mistakes: 1. Cortex-A8 load / store multiplies can only issue on ALU0. 2. Eliminate A8_Issue, A8_LSPipe will correctly limit the load / store issues. 3. Correctly model all vld1 and vld2 variants. llvm-svn: 116134
2010-10-09 03:03:04 +02:00
; CHECK: vld1.16
Convert VLD1 and VLD2 instructions to use pseudo-instructions until after regalloc. llvm-svn: 112825
2010-09-02 18:00:54 +02:00
; CHECK: vmul.i16
Change ARM scheduling default to list-hybrid if the target supports floating point instructions (and is not using soft float). llvm-svn: 104307
2010-05-21 02:43:17 +02:00
; CHECK-NOT: vmov
; CHECK: vst1.16
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vst1.16
%0 = getelementptr inbounds %struct.int16x8_t* %vT0ptr, i32 0, i32 0 ; <<8 x i16>*> [#uses=1]
%1 = load <8 x i16>* %0, align 16 ; <<8 x i16>> [#uses=1]
%2 = getelementptr inbounds %struct.int16x8_t* %vT1ptr, i32 0, i32 0 ; <<8 x i16>*> [#uses=1]
%3 = load <8 x i16>* %2, align 16 ; <<8 x i16>> [#uses=1]
%4 = bitcast i16* %i_ptr to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%5 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %4, i32 1) ; <<8 x i16>> [#uses=1]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%6 = getelementptr inbounds i16* %i_ptr, i32 8 ; <i16*> [#uses=1]
%7 = bitcast i16* %6 to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%8 = tail call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %7, i32 1) ; <<8 x i16>> [#uses=1]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%9 = mul <8 x i16> %1, %5 ; <<8 x i16>> [#uses=1]
%10 = mul <8 x i16> %3, %8 ; <<8 x i16>> [#uses=1]
%11 = bitcast i16* %o_ptr to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst1.v8i16(i8* %11, <8 x i16> %9, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%12 = getelementptr inbounds i16* %o_ptr, i32 8 ; <i16*> [#uses=1]
%13 = bitcast i16* %12 to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst1.v8i16(i8* %13, <8 x i16> %10, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
ret void
}
define <8 x i8> @t3(i8* %A, i8* %B) nounwind {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t3:
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vld3.8
Revert "Tests: Be less dependent on a specific schedule/regalloc" This reverts r192454 Apparently FileCheck isn't as smart as I though and does not enforce a topological order between variable defs+uses. llvm-svn: 192472
2013-10-11 20:09:19 +02:00
; CHECK: vmul.i8
; CHECK: vmov r
putback r116983 and fix simple-fp-encoding.ll tests llvm-svn: 116992
2010-10-21 05:40:16 +02:00
; CHECK-NOT: vmov d
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vst3.8
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp1 = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=2]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%tmp2 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 0 ; <<8 x i8>> [#uses=1]
%tmp3 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 2 ; <<8 x i8>> [#uses=1]
%tmp4 = extractvalue %struct.__neon_int8x8x3_t %tmp1, 1 ; <<8 x i8>> [#uses=1]
%tmp5 = sub <8 x i8> %tmp3, %tmp4
%tmp6 = add <8 x i8> %tmp2, %tmp3 ; <<8 x i8>> [#uses=1]
%tmp7 = mul <8 x i8> %tmp4, %tmp2
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst3.v8i8(i8* %B, <8 x i8> %tmp5, <8 x i8> %tmp6, <8 x i8> %tmp7, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
ret <8 x i8> %tmp4
}
Remove arm_apcscc from the test files. It is the default and doing this matches what llvm-gcc and clang now produce. llvm-svn: 106221
2010-06-17 17:18:27 +02:00
define void @t4(i32* %in, i32* %out) nounwind {
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
entry:
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t4:
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK: vld2.32
; CHECK-NOT: vmov
; CHECK: vld2.32
; CHECK-NOT: vmov
; CHECK: bne
%tmp1 = bitcast i32* %in to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp2 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8* %tmp1, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%tmp3 = getelementptr inbounds i32* %in, i32 8 ; <i32*> [#uses=1]
%tmp4 = bitcast i32* %tmp3 to i8* ; <i8*> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp5 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8* %tmp4, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%tmp8 = bitcast i32* %out to i8* ; <i8*> [#uses=1]
br i1 undef, label %return1, label %return2
return1:
; CHECK: %return1
; CHECK-NOT: vmov
; CHECK-NEXT: vadd.i32
; CHECK-NEXT: vadd.i32
; CHECK-NEXT: vst2.32
%tmp52 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
%tmp57 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 1 ; <<4 x i32>> [#uses=1]
%tmp = extractvalue %struct.__neon_int32x4x2_t %tmp5, 0 ; <<4 x i32>> [#uses=1]
%tmp39 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
%tmp6 = add <4 x i32> %tmp52, %tmp ; <<4 x i32>> [#uses=1]
%tmp7 = add <4 x i32> %tmp57, %tmp39 ; <<4 x i32>> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst2.v4i32(i8* %tmp8, <4 x i32> %tmp6, <4 x i32> %tmp7, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
ret void
return2:
; CHECK: %return2
; CHECK: vadd.i32
; CHECK-NOT: vmov
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; CHECK: vst2.32 {d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}, d{{[0-9]+}}}
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%tmp100 = extractvalue %struct.__neon_int32x4x2_t %tmp2, 0 ; <<4 x i32>> [#uses=1]
%tmp101 = extractvalue %struct.__neon_int32x4x2_t %tmp5, 1 ; <<4 x i32>> [#uses=1]
%tmp102 = add <4 x i32> %tmp100, %tmp101 ; <<4 x i32>> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst2.v4i32(i8* %tmp8, <4 x i32> %tmp102, <4 x i32> %tmp101, i32 1)
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
call void @llvm.trap()
unreachable
}
define <8 x i16> @t5(i16* %A, <8 x i16>* %B) nounwind {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t5:
Use vld1 / vst2 for unaligned v2f64 load / store. e.g. Use vld1.16 for 2-byte aligned address. Based on patch by David Peixotto. Also use vld1.64 / vst1.64 with 128-bit alignment to take advantage of alignment hints. rdar://12090772, rdar://12238782 llvm-svn: 164089
2012-09-18 03:42:45 +02:00
; CHECK: vld1.32
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; How can FileCheck match Q and D registers? We need a lisp interpreter.
Remove VMOVDneon and VMOVQ, which are just aliases for VORR. This continues to simplify the path towards an auto-generated disassembler. llvm-svn: 135290
2011-07-15 20:46:47 +02:00
; CHECK: vorr {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK-NOT: vmov
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; CHECK: vld2.16 {d{{[0-9]+}}[1], d{{[0-9]+}}[1]}, [r0]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
; CHECK-NOT: vmov
; CHECK: vadd.i16
%tmp0 = bitcast i16* %A to i8* ; <i8*> [#uses=1]
%tmp1 = load <8 x i16>* %B ; <<8 x i16>> [#uses=2]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp2 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16(i8* %tmp0, <8 x i16> %tmp1, <8 x i16> %tmp1, i32 1, i32 1) ; <%struct.__neon_int16x8x2_t> [#uses=2]
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
%tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 0 ; <<8 x i16>> [#uses=1]
%tmp4 = extractvalue %struct.__neon_int16x8x2_t %tmp2, 1 ; <<8 x i16>> [#uses=1]
%tmp5 = add <8 x i16> %tmp3, %tmp4 ; <<8 x i16>> [#uses=1]
ret <8 x i16> %tmp5
}
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
define <8 x i8> @t6(i8* %A, <8 x i8>* %B) nounwind {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t6:
ARM VLDR/VSTR instructions don't need a size suffix. Canonicallize on the non-suffixed form, but continue to accept assembly that has any correctly sized type suffix. llvm-svn: 144583
2011-11-15 00:03:21 +01:00
; CHECK: vldr
Remove VMOVDneon and VMOVQ, which are just aliases for VORR. This continues to simplify the path towards an auto-generated disassembler. llvm-svn: 135290
2011-07-15 20:46:47 +02:00
; CHECK: vorr d[[D0:[0-9]+]], d[[D1:[0-9]+]]
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; CHECK-NEXT: vld2.8 {d[[D1]][1], d[[D0]][1]}
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
%tmp1 = load <8 x i8>* %B ; <<8 x i8>> [#uses=2]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp2 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8(i8* %A, <8 x i8> %tmp1, <8 x i8> %tmp1, i32 1, i32 1) ; <%struct.__neon_int8x8x2_t> [#uses=2]
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
%tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 0 ; <<8 x i8>> [#uses=1]
%tmp4 = extractvalue %struct.__neon_int8x8x2_t %tmp2, 1 ; <<8 x i8>> [#uses=1]
%tmp5 = add <8 x i8> %tmp3, %tmp4 ; <<8 x i8>> [#uses=1]
ret <8 x i8> %tmp5
}
Remove arm_apcscc from the test files. It is the default and doing this matches what llvm-gcc and clang now produce. llvm-svn: 106221
2010-06-17 17:18:27 +02:00
define void @t7(i32* %iptr, i32* %optr) nounwind {
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
entry:
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t7:
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
; CHECK: vld2.32
; CHECK: vst2.32
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; CHECK: vld1.32 {d{{[0-9]+}}, d{{[0-9]+}}},
Remove VMOVDneon and VMOVQ, which are just aliases for VORR. This continues to simplify the path towards an auto-generated disassembler. llvm-svn: 135290
2011-07-15 20:46:47 +02:00
; CHECK: vorr q[[Q0:[0-9]+]], q[[Q1:[0-9]+]], q[[Q1:[0-9]+]]
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
; CHECK-NOT: vmov
Fix ARM tests to be register allocator independent. llvm-svn: 128680
2011-04-01 00:14:03 +02:00
; CHECK: vuzp.32 q[[Q1]], q[[Q0]]
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
; CHECK: vst1.32
%0 = bitcast i32* %iptr to i8* ; <i8*> [#uses=2]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%1 = tail call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8* %0, i32 1) ; <%struct.__neon_int32x4x2_t> [#uses=2]
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
%tmp57 = extractvalue %struct.__neon_int32x4x2_t %1, 0 ; <<4 x i32>> [#uses=1]
%tmp60 = extractvalue %struct.__neon_int32x4x2_t %1, 1 ; <<4 x i32>> [#uses=1]
%2 = bitcast i32* %optr to i8* ; <i8*> [#uses=2]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst2.v4i32(i8* %2, <4 x i32> %tmp57, <4 x i32> %tmp60, i32 1)
%3 = tail call <4 x i32> @llvm.arm.neon.vld1.v4i32(i8* %0, i32 1) ; <<4 x i32>> [#uses=1]
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
%4 = shufflevector <4 x i32> %3, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2> ; <<4 x i32>> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst1.v4i32(i8* %2, <4 x i32> %4, i32 1)
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
ret void
}
Fix PR7156. If the sources of a REG_SEQUENCE are all IMPLICIT_DEF's. Replace it with an IMPLICIT_DEF rather than deleting it or else it would be left without a def. llvm-svn: 103984
2010-05-18 00:09:49 +02:00
; PR7156
define arm_aapcs_vfpcc i32 @t8() nounwind {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t8:
Change register allocation order for ARM VFP and NEON registers to put the callee-saved registers at the end of the lists. Also prefer to avoid using the low registers that are in register subclasses required by certain instructions, so that those registers will more likely be available when needed. This change makes a huge improvement in spilling in some cases. Thanks to Jakob for helping me realize the problem. Most of this patch is fixing the testsuite. There are quite a few places where we're checking for specific registers. I changed those to wildcards in places where that doesn't weaken the tests. The spill-q.ll and thumb2-spill-q.ll tests stopped spilling with this change, so I added a bunch of live values to force spills on those tests. llvm-svn: 116055
2010-10-08 08:15:13 +02:00
; CHECK: vrsqrte.f32 q8, q8
Fix PR7156. If the sources of a REG_SEQUENCE are all IMPLICIT_DEF's. Replace it with an IMPLICIT_DEF rather than deleting it or else it would be left without a def. llvm-svn: 103984
2010-05-18 00:09:49 +02:00
bb.nph55.bb.nph55.split_crit_edge:
br label %bb3
bb3: ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
br i1 undef, label %bb5, label %bb3
bb5: ; preds = %bb3
br label %bb.i25
bb.i25: ; preds = %bb.i25, %bb5
%0 = shufflevector <2 x float> undef, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
%1 = call <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float> %0) nounwind ; <<4 x float>> [#uses=1]
%2 = fmul <4 x float> %1, undef ; <<4 x float>> [#uses=1]
%3 = fmul <4 x float> undef, %2 ; <<4 x float>> [#uses=1]
%tmp26.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
%4 = extractelement <2 x double> %tmp26.i, i32 0 ; <double> [#uses=1]
%5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
%6 = extractelement <2 x float> %5, i32 1 ; <float> [#uses=1]
store float %6, float* undef, align 4
br i1 undef, label %bb6, label %bb.i25
bb6: ; preds = %bb.i25
br i1 undef, label %bb7, label %bb14
bb7: ; preds = %bb6
br label %bb.i49
bb.i49: ; preds = %bb.i49, %bb7
br i1 undef, label %bb.i19, label %bb.i49
bb.i19: ; preds = %bb.i19, %bb.i49
br i1 undef, label %exit, label %bb.i19
exit: ; preds = %bb.i19
unreachable
bb14: ; preds = %bb6
ret i32 0
}
Fix PR7175. Insert copies of a REG_SEQUENCE source if it is used by other REG_SEQUENCE instructions. llvm-svn: 103994
2010-05-18 01:24:12 +02:00
%0 = type { %1, %1, %1, %1 }
%1 = type { %2 }
%2 = type { <4 x float> }
%3 = type { %0, %1 }
; PR7157
define arm_aapcs_vfpcc float @t9(%0* nocapture, %3* nocapture) nounwind {
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t9:
ARM: implement some simple f64 materializations. Previously we used a const-pool load for virtually all 64-bit floating values. Actually, we can get quite a few common values (including 0.0, 1.0) via "vmov" instructions of one stripe or another. llvm-svn: 188773
2013-08-20 10:57:11 +02:00
; CHECK: vmov.i32 d16, #0x0
Make ARMAsmPrinter generate the correct alignment specifier syntax in instructions. The Printer will now print instructions with the correct alignment specifier syntax, like vld1.8 {d16}, [r0:64] llvm-svn: 175884
2013-02-22 11:01:33 +01:00
; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
ARM: implement some simple f64 materializations. Previously we used a const-pool load for virtually all 64-bit floating values. Actually, we can get quite a few common values (including 0.0, 1.0) via "vmov" instructions of one stripe or another. llvm-svn: 188773
2013-08-20 10:57:11 +02:00
; CHECK-NEXT: vorr d17, d16, d16
Make ARMAsmPrinter generate the correct alignment specifier syntax in instructions. The Printer will now print instructions with the correct alignment specifier syntax, like vld1.8 {d16}, [r0:64] llvm-svn: 175884
2013-02-22 11:01:33 +01:00
; CHECK-NEXT: vst1.64 {d16, d17}, [r0:128]
Fix PR7175. Insert copies of a REG_SEQUENCE source if it is used by other REG_SEQUENCE instructions. llvm-svn: 103994
2010-05-18 01:24:12 +02:00
%3 = bitcast double 0.000000e+00 to <2 x float> ; <<2 x float>> [#uses=2]
%4 = shufflevector <2 x float> %3, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
store <4 x float> %4, <4 x float>* undef, align 16
%5 = shufflevector <2 x float> %3, <2 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; <<4 x float>> [#uses=1]
store <4 x float> %5, <4 x float>* undef, align 16
br label %8
; <label>:6 ; preds = %8
Eliminate the other half of the BRCOND optimization, and update as many tests as possible. llvm-svn: 106749
2010-06-24 17:24:03 +02:00
br label %7
Fix PR7175. Insert copies of a REG_SEQUENCE source if it is used by other REG_SEQUENCE instructions. llvm-svn: 103994
2010-05-18 01:24:12 +02:00
; <label>:7 ; preds = %6
br label %8
; <label>:8 ; preds = %7, %2
Eliminate the other half of the BRCOND optimization, and update as many tests as possible. llvm-svn: 106749
2010-06-24 17:24:03 +02:00
br label %6
Fix PR7175. Insert copies of a REG_SEQUENCE source if it is used by other REG_SEQUENCE instructions. llvm-svn: 103994
2010-05-18 01:24:12 +02:00
; <label>:9 ; preds = %8
ret float undef
; <label>:10 ; preds = %6
ret float 9.990000e+02
}
Fix PR7162: Use source register classes and sub-indices to determine the correct register class of the definitions of REG_SEQUENCE. llvm-svn: 104050
2010-05-18 22:03:28 +02:00
; PR7162
define arm_aapcs_vfpcc i32 @t10() nounwind {
entry:
Mass update to CodeGen tests to use CHECK-LABEL for labels corresponding to function definitions for more informative error messages. No functionality change and all updated tests passed locally. This update was done with the following bash script: find test/CodeGen -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc.*debug" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_-]*\):\( *\)$FUNC: *\$/;\1\2-LABEL:\3$FUNC:/g" $TEMP done sed -i '' "s/;\(.*\)-LABEL-LABEL:/;\1-LABEL:/" $TEMP sed -i '' "s/;\(.*\)-NEXT-LABEL:/;\1-NEXT:/" $TEMP sed -i '' "s/;\(.*\)-NOT-LABEL:/;\1-NOT:/" $TEMP sed -i '' "s/;\(.*\)-DAG-LABEL:/;\1-DAG:/" $TEMP mv $TEMP $NAME fi done llvm-svn: 186280
2013-07-14 08:24:09 +02:00
; CHECK-LABEL: t10:
Simplify some uses of utohexstr. As a side effect hex is printed lowercase instead of uppercase now. llvm-svn: 144013
2011-11-07 22:00:59 +01:00
; CHECK: vmov.i32 q[[Q0:[0-9]+]], #0x3f000000
This commit contains a few changes that had to go in together. 1. Simplify xor/and/or (bitcast(A), bitcast(B)) -> bitcast(op (A,B)) (and also scalar_to_vector). 2. Xor/and/or are indifferent to the swizzle operation (shuffle of one src). Simplify xor/and/or (shuff(A), shuff(B)) -> shuff(op (A, B)) 3. Optimize swizzles of shuffles: shuff(shuff(x, y), undef) -> shuff(x, y). 4. Fix an X86ISelLowering optimization which was very bitcast-sensitive. Code which was previously compiled to this: movd (%rsi), %xmm0 movdqa .LCPI0_0(%rip), %xmm2 pshufb %xmm2, %xmm0 movd (%rdi), %xmm1 pshufb %xmm2, %xmm1 pxor %xmm0, %xmm1 pshufb .LCPI0_1(%rip), %xmm1 movd %xmm1, (%rdi) ret Now compiles to this: movl (%rsi), %eax xorl %eax, (%rdi) ret llvm-svn: 153848
2012-04-01 21:31:22 +02:00
; CHECK: vmul.f32 q8, q8, d[[DREG:[0-1]+]]
Making use of VFP / NEON floating point multiply-accumulate / subtraction is difficult on current ARM implementations for a few reasons. 1. Even though a single vmla has latency that is one cycle shorter than a pair of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause additional pipeline stall. So it's frequently better to single codegen vmul + vadd. 2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to stall for 4 cycles. We need to schedule them apart. 3. A vmla followed vmla is a special case. Obvious issuing back to back RAW vmla + vmla is very bad. But this isn't ideal either: vmul vadd vmla Instead, we want to expand the second vmla: vmla vmul vadd Even with the 4 cycle vmul stall, the second sequence is still 2 cycles faster. Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough but it isn't the optimial solution. This patch attempts to make it possible to use vmla / vmls in cases where it is profitable. A. Add missing isel predicates which cause vmla to be codegen'ed. B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to compute a fmul and a fmla. C. Add additional isel checks for vmla, avoid cases where vmla is feeding into fp instructions (except for the #3 exceptional case). D. Add ARM hazard recognizer to model the vmla / vmls hazards. E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the vmla / vmls will trigger one of the special hazards. Work in progress, only A+B are enabled. llvm-svn: 120960
2010-12-05 23:04:16 +01:00
; CHECK: vadd.f32 q8, q8, q8
Fix PR7162: Use source register classes and sub-indices to determine the correct register class of the definitions of REG_SEQUENCE. llvm-svn: 104050
2010-05-18 22:03:28 +02:00
%0 = shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
%1 = insertelement <4 x float> %0, float undef, i32 1 ; <<4 x float>> [#uses=1]
%2 = insertelement <4 x float> %1, float undef, i32 2 ; <<4 x float>> [#uses=1]
%3 = insertelement <4 x float> %2, float undef, i32 3 ; <<4 x float>> [#uses=1]
%tmp54.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
%4 = extractelement <2 x double> %tmp54.i, i32 1 ; <double> [#uses=1]
%5 = bitcast double %4 to <2 x float> ; <<2 x float>> [#uses=1]
%6 = shufflevector <2 x float> %5, <2 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
%7 = fmul <4 x float> undef, %6 ; <<4 x float>> [#uses=1]
%8 = fadd <4 x float> %7, undef ; <<4 x float>> [#uses=1]
%9 = fadd <4 x float> %8, undef ; <<4 x float>> [#uses=1]
%10 = shufflevector <4 x float> undef, <4 x float> %9, <4 x i32> <i32 0, i32 1, i32 2, i32 7> ; <<4 x float>> [#uses=1]
%11 = fmul <4 x float> %10, <float 5.000000e-01, float 5.000000e-01, float 5.000000e-01, float 5.000000e-01> ; <<4 x float>> [#uses=1]
%12 = shufflevector <4 x float> %11, <4 x float> undef, <4 x i32> <i32 3, i32 undef, i32 undef, i32 undef> ; <<4 x float>> [#uses=1]
%13 = shufflevector <4 x float> %12, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
%14 = fmul <4 x float> %13, undef ; <<4 x float>> [#uses=1]
%15 = fadd <4 x float> undef, %14 ; <<4 x float>> [#uses=1]
%16 = shufflevector <4 x float> undef, <4 x float> %15, <4 x i32> <i32 0, i32 1, i32 6, i32 3> ; <<4 x float>> [#uses=1]
%17 = fmul <4 x float> %16, undef ; <<4 x float>> [#uses=1]
%18 = extractelement <4 x float> %17, i32 2 ; <float> [#uses=1]
store float %18, float* undef, align 4
br i1 undef, label %exit, label %bb14
exit: ; preds = %bb.i19
unreachable
bb14: ; preds = %bb6
ret i32 0
}
TwoAddressInstructionPass doesn't really know how to merge live intervals when lowering REG_SEQUENCE instructions. Insert copies for REG_SEQUENCE sources not killed to avoid breaking later passes. llvm-svn: 104146
2010-05-19 22:08:00 +02:00
; This test crashes the coalescer because live variables were not updated properly.
define <8 x i8> @t11(i8* %A1, i8* %A2, i8* %A3, i8* %A4, i8* %A5, i8* %A6, i8* %A7, i8* %A8, i8* %B) nounwind {
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp1d = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A4, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
TwoAddressInstructionPass doesn't really know how to merge live intervals when lowering REG_SEQUENCE instructions. Insert copies for REG_SEQUENCE sources not killed to avoid breaking later passes. llvm-svn: 104146
2010-05-19 22:08:00 +02:00
%tmp2d = extractvalue %struct.__neon_int8x8x3_t %tmp1d, 0 ; <<8 x i8>> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
%tmp1f = call %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8* %A6, i32 1) ; <%struct.__neon_int8x8x3_t> [#uses=1]
TwoAddressInstructionPass doesn't really know how to merge live intervals when lowering REG_SEQUENCE instructions. Insert copies for REG_SEQUENCE sources not killed to avoid breaking later passes. llvm-svn: 104146
2010-05-19 22:08:00 +02:00
%tmp2f = extractvalue %struct.__neon_int8x8x3_t %tmp1f, 0 ; <<8 x i8>> [#uses=1]
%tmp2bd = add <8 x i8> zeroinitializer, %tmp2d ; <<8 x i8>> [#uses=1]
%tmp2abcd = mul <8 x i8> zeroinitializer, %tmp2bd ; <<8 x i8>> [#uses=1]
%tmp2ef = sub <8 x i8> zeroinitializer, %tmp2f ; <<8 x i8>> [#uses=1]
%tmp2efgh = mul <8 x i8> %tmp2ef, undef ; <<8 x i8>> [#uses=2]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
call void @llvm.arm.neon.vst3.v8i8(i8* %A2, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp2efgh, i32 1)
TwoAddressInstructionPass doesn't really know how to merge live intervals when lowering REG_SEQUENCE instructions. Insert copies for REG_SEQUENCE sources not killed to avoid breaking later passes. llvm-svn: 104146
2010-05-19 22:08:00 +02:00
%tmp2 = sub <8 x i8> %tmp2efgh, %tmp2abcd ; <<8 x i8>> [#uses=1]
%tmp7 = mul <8 x i8> undef, %tmp2 ; <<8 x i8>> [#uses=1]
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
tail call void @llvm.arm.neon.vst3.v8i8(i8* %B, <8 x i8> undef, <8 x i8> undef, <8 x i8> %tmp7, i32 1)
TwoAddressInstructionPass doesn't really know how to merge live intervals when lowering REG_SEQUENCE instructions. Insert copies for REG_SEQUENCE sources not killed to avoid breaking later passes. llvm-svn: 104146
2010-05-19 22:08:00 +02:00
ret <8 x i8> undef
}
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare <4 x i32> @llvm.arm.neon.vld1.v4i32(i8*, i32) nounwind readonly
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare <8 x i16> @llvm.arm.neon.vld1.v8i16(i8*, i32) nounwind readonly
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
declare <4 x i16> @llvm.arm.neon.vshiftn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare void @llvm.arm.neon.vst1.v4i32(i8*, <4 x i32>, i32) nounwind
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare void @llvm.arm.neon.vst1.v8i16(i8*, <8 x i16>, i32) nounwind
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare void @llvm.arm.neon.vst3.v8i8(i8*, <8 x i8>, <8 x i8>, <8 x i8>, i32)
nounwind
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare %struct.__neon_int8x8x3_t @llvm.arm.neon.vld3.v8i8(i8*, i32) nounwind readonly
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8*, i32) nounwind readonly
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2lane.v8i8(i8*, <8 x i8>, <8 x i8>, i32, i32) nounwind readonly
Careful with reg_sequence coalescing to not to overwrite sub-register indices. llvm-svn: 103971
2010-05-17 22:57:12 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2lane.v8i16(i8*, <8 x i16>, <8 x i16>, i32, i32) nounwind readonly
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Add alignment arguments to all the NEON load/store intrinsics. Update all the tests using those intrinsics and add support for auto-upgrading bitcode files with the old versions of the intrinsics. llvm-svn: 112271
2010-08-27 19:13:24 +02:00
declare void @llvm.arm.neon.vst2.v4i32(i8*, <4 x i32>, <4 x i32>, i32) nounwind
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
Fix PR7156. If the sources of a REG_SEQUENCE are all IMPLICIT_DEF's. Replace it with an IMPLICIT_DEF rather than deleting it or else it would be left without a def. llvm-svn: 103984
2010-05-18 00:09:49 +02:00
declare <4 x float> @llvm.arm.neon.vrsqrte.v4f32(<4 x float>) nounwind readnone
Turn on -neon-reg-sequence by default. Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers! llvm-svn: 103960
2010-05-17 21:51:20 +02:00
declare void @llvm.trap() nounwind
Reference in New Issue Copy Permalink