[llvm] 76e0e1a - [ARM] VCVTT instruction selection
David Green via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 26 00:59:10 PDT 2020
Author: David Green
Date: 2020-06-26T08:58:55+01:00
New Revision: 76e0e1a55d1ccecebbd98fb1fa5c7ab5a4f013cc
URL: https://github.com/llvm/llvm-project/commit/76e0e1a55d1ccecebbd98fb1fa5c7ab5a4f013cc
DIFF: https://github.com/llvm/llvm-project/commit/76e0e1a55d1ccecebbd98fb1fa5c7ab5a4f013cc.diff
LOG: [ARM] VCVTT instruction selection
We current extract and convert from a top lane of a f16 vector using a
VMOVX;VCVTB pair. We can simplify that to use a single VCVTT. The
pattern is mostly copied from a vector extract pattern, but produces a
VCVTTHS f32 directly.
This had to move some code around so that ARMInstrVFP had access to the
required pattern frags that were previously part of ARMInstrNEON.
Differential Revision: https://reviews.llvm.org/D81556
Added:
Modified:
llvm/lib/Target/ARM/ARMInstrInfo.td
llvm/lib/Target/ARM/ARMInstrNEON.td
llvm/lib/Target/ARM/ARMInstrVFP.td
llvm/test/CodeGen/ARM/fp16-insert-extract.ll
llvm/test/CodeGen/Thumb2/LowOverheadLoops/fast-fp-loops.ll
llvm/test/CodeGen/Thumb2/LowOverheadLoops/mve-float-loops.ll
llvm/test/CodeGen/Thumb2/mve-div-expand.ll
llvm/test/CodeGen/Thumb2/mve-fmath.ll
llvm/test/CodeGen/Thumb2/mve-masked-ldst.ll
llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/ARM/ARMInstrInfo.td b/llvm/lib/Target/ARM/ARMInstrInfo.td
index 3fed1cd4b586..da0a836c8f95 100644
--- a/llvm/lib/Target/ARM/ARMInstrInfo.td
+++ b/llvm/lib/Target/ARM/ARMInstrInfo.td
@@ -443,6 +443,62 @@ def fsub_mlx : PatFrag<(ops node:$lhs, node:$rhs),(fsub node:$lhs, node:$rhs),[{
return hasNoVMLxHazardUse(N);
}]>;
+def imm_even : ImmLeaf<i32, [{ return (Imm & 1) == 0; }]>;
+def imm_odd : ImmLeaf<i32, [{ return (Imm & 1) == 1; }]>;
+
+//===----------------------------------------------------------------------===//
+// NEON/MVE pattern fragments
+//
+
+// Extract D sub-registers of Q registers.
+def DSubReg_i8_reg : SDNodeXForm<imm, [{
+ assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, SDLoc(N),
+ MVT::i32);
+}]>;
+def DSubReg_i16_reg : SDNodeXForm<imm, [{
+ assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, SDLoc(N),
+ MVT::i32);
+}]>;
+def DSubReg_i32_reg : SDNodeXForm<imm, [{
+ assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, SDLoc(N),
+ MVT::i32);
+}]>;
+def DSubReg_f64_reg : SDNodeXForm<imm, [{
+ assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), SDLoc(N),
+ MVT::i32);
+}]>;
+
+// Extract S sub-registers of Q/D registers.
+def SSubReg_f32_reg : SDNodeXForm<imm, [{
+ assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), SDLoc(N),
+ MVT::i32);
+}]>;
+
+// Extract S sub-registers of Q/D registers containing a given f16/bf16 lane.
+def SSubReg_f16_reg : SDNodeXForm<imm, [{
+ assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
+ return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue()/2, SDLoc(N),
+ MVT::i32);
+}]>;
+
+// Translate lane numbers from Q registers to D subregs.
+def SubReg_i8_lane : SDNodeXForm<imm, [{
+ return CurDAG->getTargetConstant(N->getZExtValue() & 7, SDLoc(N), MVT::i32);
+}]>;
+def SubReg_i16_lane : SDNodeXForm<imm, [{
+ return CurDAG->getTargetConstant(N->getZExtValue() & 3, SDLoc(N), MVT::i32);
+}]>;
+def SubReg_i32_lane : SDNodeXForm<imm, [{
+ return CurDAG->getTargetConstant(N->getZExtValue() & 1, SDLoc(N), MVT::i32);
+}]>;
+
+
+
//===----------------------------------------------------------------------===//
// Operand Definitions.
//
diff --git a/llvm/lib/Target/ARM/ARMInstrNEON.td b/llvm/lib/Target/ARM/ARMInstrNEON.td
index 17ad098f1f32..1b3f6075c0e9 100644
--- a/llvm/lib/Target/ARM/ARMInstrNEON.td
+++ b/llvm/lib/Target/ARM/ARMInstrNEON.td
@@ -2454,57 +2454,6 @@ def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
(VST1q8 addrmode6:$addr, QPR:$value)>;
}
-//===----------------------------------------------------------------------===//
-// NEON pattern fragments
-//===----------------------------------------------------------------------===//
-
-// Extract D sub-registers of Q registers.
-def DSubReg_i8_reg : SDNodeXForm<imm, [{
- assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, SDLoc(N),
- MVT::i32);
-}]>;
-def DSubReg_i16_reg : SDNodeXForm<imm, [{
- assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, SDLoc(N),
- MVT::i32);
-}]>;
-def DSubReg_i32_reg : SDNodeXForm<imm, [{
- assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, SDLoc(N),
- MVT::i32);
-}]>;
-def DSubReg_f64_reg : SDNodeXForm<imm, [{
- assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), SDLoc(N),
- MVT::i32);
-}]>;
-
-// Extract S sub-registers of Q/D registers.
-def SSubReg_f32_reg : SDNodeXForm<imm, [{
- assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), SDLoc(N),
- MVT::i32);
-}]>;
-
-// Extract S sub-registers of Q/D registers containing a given f16/bf16 lane.
-def SSubReg_f16_reg : SDNodeXForm<imm, [{
- assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
- return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue()/2, SDLoc(N),
- MVT::i32);
-}]>;
-
-// Translate lane numbers from Q registers to D subregs.
-def SubReg_i8_lane : SDNodeXForm<imm, [{
- return CurDAG->getTargetConstant(N->getZExtValue() & 7, SDLoc(N), MVT::i32);
-}]>;
-def SubReg_i16_lane : SDNodeXForm<imm, [{
- return CurDAG->getTargetConstant(N->getZExtValue() & 3, SDLoc(N), MVT::i32);
-}]>;
-def SubReg_i32_lane : SDNodeXForm<imm, [{
- return CurDAG->getTargetConstant(N->getZExtValue() & 1, SDLoc(N), MVT::i32);
-}]>;
-
//===----------------------------------------------------------------------===//
// Instruction Classes
//===----------------------------------------------------------------------===//
@@ -6410,9 +6359,6 @@ def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
(EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
}
-def imm_even : ImmLeaf<i32, [{ return (Imm & 1) == 0; }]>;
-def imm_odd : ImmLeaf<i32, [{ return (Imm & 1) == 1; }]>;
-
multiclass ExtractEltEvenF16<ValueType VT4, ValueType VT8> {
def : Pat<(extractelt (VT4 DPR:$src), imm_even:$lane),
(EXTRACT_SUBREG
diff --git a/llvm/lib/Target/ARM/ARMInstrVFP.td b/llvm/lib/Target/ARM/ARMInstrVFP.td
index 06955476c68e..0742ce548a56 100644
--- a/llvm/lib/Target/ARM/ARMInstrVFP.td
+++ b/llvm/lib/Target/ARM/ARMInstrVFP.td
@@ -759,10 +759,17 @@ def : FP16Pat<(fp_to_f16 SPR:$a),
def VCVTTHS: ASuI<0b11101, 0b11, 0b0010, 0b11, 0, (outs SPR:$Sd), (ins SPR:$Sm),
/* FIXME */ IIC_fpCVTSH, "vcvtt", ".f32.f16\t$Sd, $Sm",
- [/* For disassembly only; pattern left blank */]>,
+ [/* Intentionally left blank, see patterns below */]>,
Requires<[HasFP16]>,
Sched<[WriteFPCVT]>;
+def : FP16Pat<(f32 (fpextend (extractelt (v8f16 MQPR:$src), imm_odd:$lane))),
+ (VCVTTHS (EXTRACT_SUBREG MQPR:$src, (SSubReg_f16_reg imm_odd:$lane)))>;
+def : FP16Pat<(f32 (fpextend (extractelt (v4f16 DPR:$src), imm_odd:$lane))),
+ (VCVTTHS (EXTRACT_SUBREG
+ (v2f32 (COPY_TO_REGCLASS (v4f16 DPR:$src), DPR_VFP2)),
+ (SSubReg_f16_reg imm_odd:$lane)))>;
+
def VCVTTSH: ASuI<0b11101, 0b11, 0b0011, 0b11, 0, (outs SPR:$Sd), (ins SPR:$Sm),
/* FIXME */ IIC_fpCVTHS, "vcvtt", ".f16.f32\t$Sd, $Sm",
[/* For disassembly only; pattern left blank */]>,
diff --git a/llvm/test/CodeGen/ARM/fp16-insert-extract.ll b/llvm/test/CodeGen/ARM/fp16-insert-extract.ll
index 0079159722cb..f9f5b89c119c 100644
--- a/llvm/test/CodeGen/ARM/fp16-insert-extract.ll
+++ b/llvm/test/CodeGen/ARM/fp16-insert-extract.ll
@@ -5,15 +5,13 @@
define float @test_vget_lane_f16_1(<4 x half> %a) nounwind {
; CHECKHARD-LABEL: test_vget_lane_f16_1:
; CHECKHARD: @ %bb.0: @ %entry
-; CHECKHARD-NEXT: vmovx.f16 s0, s0
-; CHECKHARD-NEXT: vcvtb.f32.f16 s0, s0
+; CHECKHARD-NEXT: vcvtt.f32.f16 s0, s0
; CHECKHARD-NEXT: bx lr
;
; CHECKSOFT-LABEL: test_vget_lane_f16_1:
; CHECKSOFT: @ %bb.0: @ %entry
; CHECKSOFT-NEXT: vmov d0, r0, r1
-; CHECKSOFT-NEXT: vmovx.f16 s0, s0
-; CHECKSOFT-NEXT: vcvtb.f32.f16 s0, s0
+; CHECKSOFT-NEXT: vcvtt.f32.f16 s0, s0
; CHECKSOFT-NEXT: vmov r0, s0
; CHECKSOFT-NEXT: bx lr
entry:
@@ -61,15 +59,13 @@ entry:
define float @test_vget_laneq_f16_7(<8 x half> %a) nounwind {
; CHECKHARD-LABEL: test_vget_laneq_f16_7:
; CHECKHARD: @ %bb.0: @ %entry
-; CHECKHARD-NEXT: vmovx.f16 s0, s3
-; CHECKHARD-NEXT: vcvtb.f32.f16 s0, s0
+; CHECKHARD-NEXT: vcvtt.f32.f16 s0, s3
; CHECKHARD-NEXT: bx lr
;
; CHECKSOFT-LABEL: test_vget_laneq_f16_7:
; CHECKSOFT: @ %bb.0: @ %entry
; CHECKSOFT-NEXT: vmov d1, r2, r3
-; CHECKSOFT-NEXT: vmovx.f16 s0, s3
-; CHECKSOFT-NEXT: vcvtb.f32.f16 s0, s0
+; CHECKSOFT-NEXT: vcvtt.f32.f16 s0, s3
; CHECKSOFT-NEXT: vmov r0, s0
; CHECKSOFT-NEXT: bx lr
entry:
diff --git a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/fast-fp-loops.ll b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/fast-fp-loops.ll
index 0380d5f20bc7..8c18159c24c5 100644
--- a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/fast-fp-loops.ll
+++ b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/fast-fp-loops.ll
@@ -344,18 +344,16 @@ define arm_aapcs_vfpcc float @fast_float_half_mac(half* nocapture readonly %b, h
; CHECK-NEXT: vmov.16 q6[3], r2
; CHECK-NEXT: .LBB2_3: @ %else26
; CHECK-NEXT: @ in Loop: Header=BB2_4 Depth=1
-; CHECK-NEXT: vmul.f16 q5, q6, q5
+; CHECK-NEXT: vmul.f16 q0, q6, q5
; CHECK-NEXT: adds r0, #8
-; CHECK-NEXT: vmovx.f16 s2, s21
-; CHECK-NEXT: vmovx.f16 s0, s20
-; CHECK-NEXT: vcvtb.f32.f16 s27, s2
+; CHECK-NEXT: vcvtt.f32.f16 s23, s1
; CHECK-NEXT: adds r1, #8
-; CHECK-NEXT: vcvtb.f32.f16 s26, s21
+; CHECK-NEXT: vcvtb.f32.f16 s22, s1
; CHECK-NEXT: adds r3, #4
-; CHECK-NEXT: vcvtb.f32.f16 s25, s0
+; CHECK-NEXT: vcvtt.f32.f16 s21, s0
; CHECK-NEXT: subs.w lr, lr, #1
-; CHECK-NEXT: vcvtb.f32.f16 s24, s20
-; CHECK-NEXT: vadd.f32 q5, q3, q6
+; CHECK-NEXT: vcvtb.f32.f16 s20, s0
+; CHECK-NEXT: vadd.f32 q5, q3, q5
; CHECK-NEXT: bne .LBB2_4
; CHECK-NEXT: b .LBB2_21
; CHECK-NEXT: .LBB2_4: @ %vector.body
diff --git a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/mve-float-loops.ll b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/mve-float-loops.ll
index b01c98cedeb7..b64c23789490 100644
--- a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/mve-float-loops.ll
+++ b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/mve-float-loops.ll
@@ -1011,13 +1011,11 @@ define arm_aapcs_vfpcc void @half_half_mul(half* nocapture readonly %a, half* no
; CHECK-NEXT: vmov.32 q1[1], r10
; CHECK-NEXT: adds r5, #8
; CHECK-NEXT: vmul.f16 q0, q0, q1
-; CHECK-NEXT: vmovx.f16 s6, s1
-; CHECK-NEXT: vmovx.f16 s4, s0
-; CHECK-NEXT: vcvtb.f32.f16 s11, s6
-; CHECK-NEXT: vcvtb.f32.f16 s10, s1
-; CHECK-NEXT: vcvtb.f32.f16 s9, s4
-; CHECK-NEXT: vcvtb.f32.f16 s8, s0
-; CHECK-NEXT: vstrb.8 q2, [r6], #16
+; CHECK-NEXT: vcvtt.f32.f16 s7, s1
+; CHECK-NEXT: vcvtb.f32.f16 s6, s1
+; CHECK-NEXT: vcvtt.f32.f16 s5, s0
+; CHECK-NEXT: vcvtb.f32.f16 s4, s0
+; CHECK-NEXT: vstrb.8 q1, [r6], #16
; CHECK-NEXT: le lr, .LBB5_4
; CHECK-NEXT: @ %bb.5: @ %middle.block
; CHECK-NEXT: cmp r12, r3
@@ -1130,13 +1128,11 @@ define arm_aapcs_vfpcc void @half_half_add(half* nocapture readonly %a, half* no
; CHECK-NEXT: vmov.32 q1[1], r10
; CHECK-NEXT: adds r5, #8
; CHECK-NEXT: vadd.f16 q0, q0, q1
-; CHECK-NEXT: vmovx.f16 s6, s1
-; CHECK-NEXT: vmovx.f16 s4, s0
-; CHECK-NEXT: vcvtb.f32.f16 s11, s6
-; CHECK-NEXT: vcvtb.f32.f16 s10, s1
-; CHECK-NEXT: vcvtb.f32.f16 s9, s4
-; CHECK-NEXT: vcvtb.f32.f16 s8, s0
-; CHECK-NEXT: vstrb.8 q2, [r6], #16
+; CHECK-NEXT: vcvtt.f32.f16 s7, s1
+; CHECK-NEXT: vcvtb.f32.f16 s6, s1
+; CHECK-NEXT: vcvtt.f32.f16 s5, s0
+; CHECK-NEXT: vcvtb.f32.f16 s4, s0
+; CHECK-NEXT: vstrb.8 q1, [r6], #16
; CHECK-NEXT: le lr, .LBB6_4
; CHECK-NEXT: @ %bb.5: @ %middle.block
; CHECK-NEXT: cmp r12, r3
@@ -1249,13 +1245,11 @@ define arm_aapcs_vfpcc void @half_half_sub(half* nocapture readonly %a, half* no
; CHECK-NEXT: vmov.32 q1[1], r10
; CHECK-NEXT: adds r5, #8
; CHECK-NEXT: vsub.f16 q0, q0, q1
-; CHECK-NEXT: vmovx.f16 s6, s1
-; CHECK-NEXT: vmovx.f16 s4, s0
-; CHECK-NEXT: vcvtb.f32.f16 s11, s6
-; CHECK-NEXT: vcvtb.f32.f16 s10, s1
-; CHECK-NEXT: vcvtb.f32.f16 s9, s4
-; CHECK-NEXT: vcvtb.f32.f16 s8, s0
-; CHECK-NEXT: vstrb.8 q2, [r6], #16
+; CHECK-NEXT: vcvtt.f32.f16 s7, s1
+; CHECK-NEXT: vcvtb.f32.f16 s6, s1
+; CHECK-NEXT: vcvtt.f32.f16 s5, s0
+; CHECK-NEXT: vcvtb.f32.f16 s4, s0
+; CHECK-NEXT: vstrb.8 q1, [r6], #16
; CHECK-NEXT: le lr, .LBB7_4
; CHECK-NEXT: @ %bb.5: @ %middle.block
; CHECK-NEXT: cmp r12, r3
@@ -1373,13 +1367,11 @@ define arm_aapcs_vfpcc void @half_short_mul(half* nocapture readonly %a, i16* no
; CHECK-NEXT: vmov.32 q1[0], r9
; CHECK-NEXT: vmov.32 q1[1], r8
; CHECK-NEXT: vmul.f16 q0, q1, q0
-; CHECK-NEXT: vmovx.f16 s6, s1
-; CHECK-NEXT: vmovx.f16 s4, s0
-; CHECK-NEXT: vcvtb.f32.f16 s11, s6
-; CHECK-NEXT: vcvtb.f32.f16 s10, s1
-; CHECK-NEXT: vcvtb.f32.f16 s9, s4
-; CHECK-NEXT: vcvtb.f32.f16 s8, s0
-; CHECK-NEXT: vstrb.8 q2, [r6], #16
+; CHECK-NEXT: vcvtt.f32.f16 s7, s1
+; CHECK-NEXT: vcvtb.f32.f16 s6, s1
+; CHECK-NEXT: vcvtt.f32.f16 s5, s0
+; CHECK-NEXT: vcvtb.f32.f16 s4, s0
+; CHECK-NEXT: vstrb.8 q1, [r6], #16
; CHECK-NEXT: le lr, .LBB8_4
; CHECK-NEXT: @ %bb.5: @ %middle.block
; CHECK-NEXT: cmp r12, r3
diff --git a/llvm/test/CodeGen/Thumb2/mve-div-expand.ll b/llvm/test/CodeGen/Thumb2/mve-div-expand.ll
index d72495b215bc..47fdf60a100a 100644
--- a/llvm/test/CodeGen/Thumb2/mve-div-expand.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-div-expand.ll
@@ -972,82 +972,74 @@ define arm_aapcs_vfpcc <8 x half> @frem_f16(<8 x half> %in1, <8 x half> %in2) {
; CHECK-NEXT: push {r4, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
; CHECK-NEXT: vpush {d8, d9, d10, d11, d12, d13}
-; CHECK-NEXT: vmov q4, q0
-; CHECK-NEXT: vmov q5, q1
-; CHECK-NEXT: vcvtb.f32.f16 s0, s16
-; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vmov q5, q0
+; CHECK-NEXT: vmov q4, q1
; CHECK-NEXT: vcvtb.f32.f16 s0, s20
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtb.f32.f16 s0, s16
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s16
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s20
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vcvtt.f32.f16 s0, s20
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vmov.16 q6[0], r4
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s17
+; CHECK-NEXT: vcvtb.f32.f16 s0, s21
; CHECK-NEXT: vmov.16 q6[1], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s21
+; CHECK-NEXT: vcvtb.f32.f16 s0, s17
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s17
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s21
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s21
; CHECK-NEXT: vmov.16 q6[2], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s18
+; CHECK-NEXT: vcvtb.f32.f16 s0, s22
; CHECK-NEXT: vmov.16 q6[3], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s22
+; CHECK-NEXT: vcvtb.f32.f16 s0, s18
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s18
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s22
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s22
; CHECK-NEXT: vmov.16 q6[4], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s19
+; CHECK-NEXT: vcvtb.f32.f16 s0, s23
; CHECK-NEXT: vmov.16 q6[5], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s23
+; CHECK-NEXT: vcvtb.f32.f16 s0, s19
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s19
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s23
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s23
; CHECK-NEXT: vmov.16 q6[6], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl fmodf
; CHECK-NEXT: vmov s0, r0
diff --git a/llvm/test/CodeGen/Thumb2/mve-fmath.ll b/llvm/test/CodeGen/Thumb2/mve-fmath.ll
index d93f31c9e217..1984478ba2bc 100644
--- a/llvm/test/CodeGen/Thumb2/mve-fmath.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-fmath.ll
@@ -123,8 +123,7 @@ define arm_aapcs_vfpcc <8 x half> @cos_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl cosf
; CHECK-NEXT: vmov s0, r0
@@ -138,8 +137,7 @@ define arm_aapcs_vfpcc <8 x half> @cos_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl cosf
@@ -153,8 +151,7 @@ define arm_aapcs_vfpcc <8 x half> @cos_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl cosf
@@ -168,8 +165,7 @@ define arm_aapcs_vfpcc <8 x half> @cos_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl cosf
@@ -254,8 +250,7 @@ define arm_aapcs_vfpcc <8 x half> @sin_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl sinf
; CHECK-NEXT: vmov s0, r0
@@ -269,8 +264,7 @@ define arm_aapcs_vfpcc <8 x half> @sin_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl sinf
@@ -284,8 +278,7 @@ define arm_aapcs_vfpcc <8 x half> @sin_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl sinf
@@ -299,8 +292,7 @@ define arm_aapcs_vfpcc <8 x half> @sin_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl sinf
@@ -385,8 +377,7 @@ define arm_aapcs_vfpcc <8 x half> @exp_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl expf
; CHECK-NEXT: vmov s0, r0
@@ -400,8 +391,7 @@ define arm_aapcs_vfpcc <8 x half> @exp_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl expf
@@ -415,8 +405,7 @@ define arm_aapcs_vfpcc <8 x half> @exp_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl expf
@@ -430,8 +419,7 @@ define arm_aapcs_vfpcc <8 x half> @exp_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl expf
@@ -516,8 +504,7 @@ define arm_aapcs_vfpcc <8 x half> @exp2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl exp2f
; CHECK-NEXT: vmov s0, r0
@@ -531,8 +518,7 @@ define arm_aapcs_vfpcc <8 x half> @exp2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl exp2f
@@ -546,8 +532,7 @@ define arm_aapcs_vfpcc <8 x half> @exp2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl exp2f
@@ -561,8 +546,7 @@ define arm_aapcs_vfpcc <8 x half> @exp2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl exp2f
@@ -647,8 +631,7 @@ define arm_aapcs_vfpcc <8 x half> @log_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl logf
; CHECK-NEXT: vmov s0, r0
@@ -662,8 +645,7 @@ define arm_aapcs_vfpcc <8 x half> @log_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl logf
@@ -677,8 +659,7 @@ define arm_aapcs_vfpcc <8 x half> @log_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl logf
@@ -692,8 +673,7 @@ define arm_aapcs_vfpcc <8 x half> @log_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl logf
@@ -778,8 +758,7 @@ define arm_aapcs_vfpcc <8 x half> @log2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log2f
; CHECK-NEXT: vmov s0, r0
@@ -793,8 +772,7 @@ define arm_aapcs_vfpcc <8 x half> @log2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log2f
@@ -808,8 +786,7 @@ define arm_aapcs_vfpcc <8 x half> @log2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log2f
@@ -823,8 +800,7 @@ define arm_aapcs_vfpcc <8 x half> @log2_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log2f
@@ -909,8 +885,7 @@ define arm_aapcs_vfpcc <8 x half> @log10_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s16
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log10f
; CHECK-NEXT: vmov s0, r0
@@ -924,8 +899,7 @@ define arm_aapcs_vfpcc <8 x half> @log10_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s17
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov.16 q5[2], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log10f
@@ -939,8 +913,7 @@ define arm_aapcs_vfpcc <8 x half> @log10_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s18
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov.16 q5[4], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log10f
@@ -954,8 +927,7 @@ define arm_aapcs_vfpcc <8 x half> @log10_float16_t(<8 x half> %src) {
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s19
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov.16 q5[6], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: bl log10f
@@ -1039,82 +1011,74 @@ define arm_aapcs_vfpcc <8 x half> @pow_float16_t(<8 x half> %src1, <8 x half> %s
; CHECK-NEXT: push {r4, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
; CHECK-NEXT: vpush {d8, d9, d10, d11, d12, d13}
-; CHECK-NEXT: vmov q4, q0
-; CHECK-NEXT: vmov q5, q1
-; CHECK-NEXT: vcvtb.f32.f16 s0, s16
-; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vmov q5, q0
+; CHECK-NEXT: vmov q4, q1
; CHECK-NEXT: vcvtb.f32.f16 s0, s20
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtb.f32.f16 s0, s16
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s16
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r4, s0
-; CHECK-NEXT: vmovx.f16 s0, s20
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vcvtt.f32.f16 s0, s20
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s16
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vmov.16 q6[0], r4
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s17
+; CHECK-NEXT: vcvtb.f32.f16 s0, s21
; CHECK-NEXT: vmov.16 q6[1], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s21
+; CHECK-NEXT: vcvtb.f32.f16 s0, s17
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s17
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s21
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s21
; CHECK-NEXT: vmov.16 q6[2], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s17
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s18
+; CHECK-NEXT: vcvtb.f32.f16 s0, s22
; CHECK-NEXT: vmov.16 q6[3], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s22
+; CHECK-NEXT: vcvtb.f32.f16 s0, s18
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s18
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s22
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s22
; CHECK-NEXT: vmov.16 q6[4], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s18
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s19
+; CHECK-NEXT: vcvtb.f32.f16 s0, s23
; CHECK-NEXT: vmov.16 q6[5], r0
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s0, s23
+; CHECK-NEXT: vcvtb.f32.f16 s0, s19
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
-; CHECK-NEXT: vmovx.f16 s2, s19
; CHECK-NEXT: vcvtb.f16.f32 s0, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s2
; CHECK-NEXT: vmov r0, s0
-; CHECK-NEXT: vmovx.f16 s0, s23
-; CHECK-NEXT: vcvtb.f32.f16 s0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s23
; CHECK-NEXT: vmov.16 q6[6], r0
-; CHECK-NEXT: vmov r0, s2
+; CHECK-NEXT: vmov r0, s0
+; CHECK-NEXT: vcvtt.f32.f16 s0, s19
; CHECK-NEXT: vmov r1, s0
; CHECK-NEXT: bl powf
; CHECK-NEXT: vmov s0, r0
diff --git a/llvm/test/CodeGen/Thumb2/mve-masked-ldst.ll b/llvm/test/CodeGen/Thumb2/mve-masked-ldst.ll
index 795e218d7fe2..a3f06e5d2537 100644
--- a/llvm/test/CodeGen/Thumb2/mve-masked-ldst.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-masked-ldst.ll
@@ -876,11 +876,11 @@ define void @foo_v4f32_v4f16(<4 x float> *%dest, <4 x i16> *%mask, <4 x half> *%
; CHECK-NEXT: lsls.w r1, lr, #29
; CHECK-NEXT: bmi .LBB18_7
; CHECK-NEXT: .LBB18_4:
-; CHECK-NEXT: vmov q2, q0
+; CHECK-NEXT: vmov q1, q0
; CHECK-NEXT: lsls.w r1, lr, #28
; CHECK-NEXT: bmi .LBB18_8
; CHECK-NEXT: .LBB18_5:
-; CHECK-NEXT: vmov q1, q2
+; CHECK-NEXT: vmov q2, q1
; CHECK-NEXT: b .LBB18_9
; CHECK-NEXT: .LBB18_6:
; CHECK-NEXT: vmov q0, q1
@@ -889,37 +889,35 @@ define void @foo_v4f32_v4f16(<4 x float> *%dest, <4 x i16> *%mask, <4 x half> *%
; CHECK-NEXT: .LBB18_7: @ %cond.load4
; CHECK-NEXT: vmovx.f16 s4, s0
; CHECK-NEXT: vmov r1, s0
+; CHECK-NEXT: vldr.16 s8, [r2, #4]
; CHECK-NEXT: vmov r3, s4
-; CHECK-NEXT: vldr.16 s4, [r2, #4]
-; CHECK-NEXT: vmov.16 q2[0], r1
+; CHECK-NEXT: vmov.16 q1[0], r1
; CHECK-NEXT: vmovx.f16 s0, s1
-; CHECK-NEXT: vmov.16 q2[1], r3
-; CHECK-NEXT: vmov r1, s4
-; CHECK-NEXT: vmov.16 q2[2], r1
+; CHECK-NEXT: vmov.16 q1[1], r3
+; CHECK-NEXT: vmov r1, s8
+; CHECK-NEXT: vmov.16 q1[2], r1
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q2[3], r1
+; CHECK-NEXT: vmov.16 q1[3], r1
; CHECK-NEXT: lsls.w r1, lr, #28
; CHECK-NEXT: bpl .LBB18_5
; CHECK-NEXT: .LBB18_8: @ %cond.load7
-; CHECK-NEXT: vmovx.f16 s0, s8
-; CHECK-NEXT: vmov r3, s8
+; CHECK-NEXT: vmovx.f16 s0, s4
+; CHECK-NEXT: vmov r3, s4
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q1[0], r3
+; CHECK-NEXT: vmov.16 q2[0], r3
; CHECK-NEXT: vldr.16 s0, [r2, #6]
-; CHECK-NEXT: vmov.16 q1[1], r1
-; CHECK-NEXT: vmov r1, s9
-; CHECK-NEXT: vmov.16 q1[2], r1
+; CHECK-NEXT: vmov.16 q2[1], r1
+; CHECK-NEXT: vmov r1, s5
+; CHECK-NEXT: vmov.16 q2[2], r1
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q1[3], r1
+; CHECK-NEXT: vmov.16 q2[3], r1
; CHECK-NEXT: .LBB18_9: @ %else8
; CHECK-NEXT: vmrs r2, p0
; CHECK-NEXT: movs r1, #0
-; CHECK-NEXT: vmovx.f16 s0, s5
-; CHECK-NEXT: vmovx.f16 s8, s4
-; CHECK-NEXT: vcvtb.f32.f16 s3, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s5
-; CHECK-NEXT: vcvtb.f32.f16 s1, s8
-; CHECK-NEXT: vcvtb.f32.f16 s0, s4
+; CHECK-NEXT: vcvtt.f32.f16 s3, s9
+; CHECK-NEXT: vcvtb.f32.f16 s2, s9
+; CHECK-NEXT: vcvtt.f32.f16 s1, s8
+; CHECK-NEXT: vcvtb.f32.f16 s0, s8
; CHECK-NEXT: and r3, r2, #1
; CHECK-NEXT: rsbs r3, r3, #0
; CHECK-NEXT: bfi r1, r3, #0, #1
@@ -1004,11 +1002,11 @@ define void @foo_v4f32_v4f16_unaligned(<4 x float> *%dest, <4 x i16> *%mask, <4
; CHECK-NEXT: lsls.w r1, lr, #29
; CHECK-NEXT: bmi .LBB19_7
; CHECK-NEXT: .LBB19_4:
-; CHECK-NEXT: vmov q2, q0
+; CHECK-NEXT: vmov q1, q0
; CHECK-NEXT: lsls.w r1, lr, #28
; CHECK-NEXT: bmi .LBB19_8
; CHECK-NEXT: .LBB19_5:
-; CHECK-NEXT: vmov q1, q2
+; CHECK-NEXT: vmov q2, q1
; CHECK-NEXT: b .LBB19_9
; CHECK-NEXT: .LBB19_6:
; CHECK-NEXT: vmov q0, q1
@@ -1017,37 +1015,35 @@ define void @foo_v4f32_v4f16_unaligned(<4 x float> *%dest, <4 x i16> *%mask, <4
; CHECK-NEXT: .LBB19_7: @ %cond.load4
; CHECK-NEXT: vmovx.f16 s4, s0
; CHECK-NEXT: vmov r1, s0
+; CHECK-NEXT: vldr.16 s8, [r2, #4]
; CHECK-NEXT: vmov r3, s4
-; CHECK-NEXT: vldr.16 s4, [r2, #4]
-; CHECK-NEXT: vmov.16 q2[0], r1
+; CHECK-NEXT: vmov.16 q1[0], r1
; CHECK-NEXT: vmovx.f16 s0, s1
-; CHECK-NEXT: vmov.16 q2[1], r3
-; CHECK-NEXT: vmov r1, s4
-; CHECK-NEXT: vmov.16 q2[2], r1
+; CHECK-NEXT: vmov.16 q1[1], r3
+; CHECK-NEXT: vmov r1, s8
+; CHECK-NEXT: vmov.16 q1[2], r1
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q2[3], r1
+; CHECK-NEXT: vmov.16 q1[3], r1
; CHECK-NEXT: lsls.w r1, lr, #28
; CHECK-NEXT: bpl .LBB19_5
; CHECK-NEXT: .LBB19_8: @ %cond.load7
-; CHECK-NEXT: vmovx.f16 s0, s8
-; CHECK-NEXT: vmov r3, s8
+; CHECK-NEXT: vmovx.f16 s0, s4
+; CHECK-NEXT: vmov r3, s4
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q1[0], r3
+; CHECK-NEXT: vmov.16 q2[0], r3
; CHECK-NEXT: vldr.16 s0, [r2, #6]
-; CHECK-NEXT: vmov.16 q1[1], r1
-; CHECK-NEXT: vmov r1, s9
-; CHECK-NEXT: vmov.16 q1[2], r1
+; CHECK-NEXT: vmov.16 q2[1], r1
+; CHECK-NEXT: vmov r1, s5
+; CHECK-NEXT: vmov.16 q2[2], r1
; CHECK-NEXT: vmov r1, s0
-; CHECK-NEXT: vmov.16 q1[3], r1
+; CHECK-NEXT: vmov.16 q2[3], r1
; CHECK-NEXT: .LBB19_9: @ %else8
; CHECK-NEXT: vmrs r2, p0
; CHECK-NEXT: movs r1, #0
-; CHECK-NEXT: vmovx.f16 s0, s5
-; CHECK-NEXT: vmovx.f16 s8, s4
-; CHECK-NEXT: vcvtb.f32.f16 s3, s0
-; CHECK-NEXT: vcvtb.f32.f16 s2, s5
-; CHECK-NEXT: vcvtb.f32.f16 s1, s8
-; CHECK-NEXT: vcvtb.f32.f16 s0, s4
+; CHECK-NEXT: vcvtt.f32.f16 s3, s9
+; CHECK-NEXT: vcvtb.f32.f16 s2, s9
+; CHECK-NEXT: vcvtt.f32.f16 s1, s8
+; CHECK-NEXT: vcvtb.f32.f16 s0, s8
; CHECK-NEXT: and r3, r2, #1
; CHECK-NEXT: rsbs r3, r3, #0
; CHECK-NEXT: bfi r1, r3, #0, #1
diff --git a/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll b/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
index a78647dfabc3..d7792628b575 100644
--- a/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-vcvt16.ll
@@ -4,11 +4,9 @@
define arm_aapcs_vfpcc <4 x float> @fpext_4(<4 x half> %src1) {
; CHECK-LABEL: fpext_4:
; CHECK: @ %bb.0: @ %entry
-; CHECK-NEXT: vmovx.f16 s4, s1
-; CHECK-NEXT: vmovx.f16 s8, s0
-; CHECK-NEXT: vcvtb.f32.f16 s7, s4
+; CHECK-NEXT: vcvtt.f32.f16 s7, s1
; CHECK-NEXT: vcvtb.f32.f16 s6, s1
-; CHECK-NEXT: vcvtb.f32.f16 s5, s8
+; CHECK-NEXT: vcvtt.f32.f16 s5, s0
; CHECK-NEXT: vcvtb.f32.f16 s4, s0
; CHECK-NEXT: vmov q0, q1
; CHECK-NEXT: bx lr
@@ -20,17 +18,13 @@ entry:
define arm_aapcs_vfpcc <8 x float> @fpext_8(<8 x half> %src1) {
; CHECK-LABEL: fpext_8:
; CHECK: @ %bb.0: @ %entry
-; CHECK-NEXT: vmovx.f16 s8, s1
-; CHECK-NEXT: vmovx.f16 s6, s0
-; CHECK-NEXT: vcvtb.f32.f16 s11, s8
-; CHECK-NEXT: vmovx.f16 s4, s3
+; CHECK-NEXT: vcvtt.f32.f16 s11, s1
; CHECK-NEXT: vcvtb.f32.f16 s10, s1
-; CHECK-NEXT: vmovx.f16 s12, s2
-; CHECK-NEXT: vcvtb.f32.f16 s9, s6
+; CHECK-NEXT: vcvtt.f32.f16 s9, s0
; CHECK-NEXT: vcvtb.f32.f16 s8, s0
-; CHECK-NEXT: vcvtb.f32.f16 s7, s4
+; CHECK-NEXT: vcvtt.f32.f16 s7, s3
; CHECK-NEXT: vcvtb.f32.f16 s6, s3
-; CHECK-NEXT: vcvtb.f32.f16 s5, s12
+; CHECK-NEXT: vcvtt.f32.f16 s5, s2
; CHECK-NEXT: vcvtb.f32.f16 s4, s2
; CHECK-NEXT: vmov q0, q2
; CHECK-NEXT: bx lr
@@ -274,23 +268,16 @@ entry:
define arm_aapcs_vfpcc <8 x float> @load_shuffleext_16(<16 x half>* %src) {
; CHECK-LABEL: load_shuffleext_16:
; CHECK: @ %bb.0: @ %entry
-; CHECK-NEXT: .vsave {d8}
-; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vld20.16 {q2, q3}, [r0]
; CHECK-NEXT: vld21.16 {q2, q3}, [r0]
-; CHECK-NEXT: vmovx.f16 s0, s9
-; CHECK-NEXT: vmovx.f16 s6, s8
-; CHECK-NEXT: vcvtb.f32.f16 s3, s0
-; CHECK-NEXT: vmovx.f16 s4, s11
+; CHECK-NEXT: vcvtt.f32.f16 s3, s9
; CHECK-NEXT: vcvtb.f32.f16 s2, s9
-; CHECK-NEXT: vmovx.f16 s16, s10
-; CHECK-NEXT: vcvtb.f32.f16 s1, s6
+; CHECK-NEXT: vcvtt.f32.f16 s1, s8
; CHECK-NEXT: vcvtb.f32.f16 s0, s8
-; CHECK-NEXT: vcvtb.f32.f16 s7, s4
+; CHECK-NEXT: vcvtt.f32.f16 s7, s11
; CHECK-NEXT: vcvtb.f32.f16 s6, s11
-; CHECK-NEXT: vcvtb.f32.f16 s5, s16
+; CHECK-NEXT: vcvtt.f32.f16 s5, s10
; CHECK-NEXT: vcvtb.f32.f16 s4, s10
-; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: bx lr
entry:
%wide.load = load <16 x half>, <16 x half>* %src, align 4
More information about the llvm-commits
mailing list