[AArch64] Don't emit Neon in streaming[-compatible] functions with -fzero-call-used-regs #116995
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…zero-call-used-regs Previously, with `-fzero-call-used-regs` clang/LLVM would incorrectly emit Neon instructions in streaming functions, and streaming-compatible functions without SVE. With this change: * In streaming functions, Z/p registers will be zeroed * In streaming compatible functions w/o SVE, D registers will be zeroed - (As Neon vector instructions are illegal including `movi v..`)
Member
|
@llvm/pr-subscribers-backend-aarch64 Author: Benjamin Maxwell (MacDue) ChangesPreviously, with With this change:
Patch is 24.83 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/116995.diff 3 Files Affected:
diff --git a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
index 216244950ba9ee..d6673969aa3056 100644
--- a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -1010,7 +1010,7 @@ void AArch64FrameLowering::emitZeroCallUsedRegs(BitVector RegsToZero,
BitVector GPRsToZero(TRI.getNumRegs());
BitVector FPRsToZero(TRI.getNumRegs());
- bool HasSVE = STI.hasSVE();
+ bool HasSVE = STI.isSVEorStreamingSVEAvailable();
for (MCRegister Reg : RegsToZero.set_bits()) {
if (TRI.isGeneralPurposeRegister(MF, Reg)) {
// For GPRs, we only care to clear out the 64-bit register.
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
index a470c03efd5eb4..41cfecb60561c9 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -9694,19 +9694,26 @@ void AArch64InstrInfo::buildClearRegister(Register Reg, MachineBasicBlock &MBB,
MachineBasicBlock::iterator Iter,
DebugLoc &DL,
bool AllowSideEffects) const {
- const MachineFunction &MF = *MBB.getParent();
+ MachineFunction &MF = *MBB.getParent();
const AArch64Subtarget &STI = MF.getSubtarget<AArch64Subtarget>();
const AArch64RegisterInfo &TRI = *STI.getRegisterInfo();
if (TRI.isGeneralPurposeRegister(MF, Reg)) {
BuildMI(MBB, Iter, DL, get(AArch64::MOVZXi), Reg).addImm(0).addImm(0);
- } else if (STI.hasSVE()) {
+ } else if (STI.isSVEorStreamingSVEAvailable()) {
BuildMI(MBB, Iter, DL, get(AArch64::DUP_ZI_D), Reg)
.addImm(0)
.addImm(0);
- } else {
+ } else if (STI.isNeonAvailable()) {
BuildMI(MBB, Iter, DL, get(AArch64::MOVIv2d_ns), Reg)
.addImm(0);
+ } else {
+ // This is a streaming-compatible function without SVE. We don't have full
+ // Neon (just FPRs), so we can at most use the first 64-bit sub-register.
+ // So given `movi v..` would be illegal use `fmov d..` instead.
+ assert(STI.hasNEON() && "Expected to have NEON.");
+ Register Reg64 = TRI.getSubReg(Reg, AArch64::dsub);
+ BuildMI(MBB, Iter, DL, get(AArch64::FMOVD0), Reg64);
}
}
diff --git a/llvm/test/CodeGen/AArch64/zero-call-used-regs.ll b/llvm/test/CodeGen/AArch64/zero-call-used-regs.ll
index 75a7c7f4a0511e..4799ea3bcd19f6 100644
--- a/llvm/test/CodeGen/AArch64/zero-call-used-regs.ll
+++ b/llvm/test/CodeGen/AArch64/zero-call-used-regs.ll
@@ -1,6 +1,10 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,DEFAULT
-; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-unknown-unknown -mattr=+sve | FileCheck %s --check-prefixes=CHECK,SVE
+; RUN: llc < %s -verify-machineinstrs -mtriple=aarch64-unknown-unknown -mattr=+sve | FileCheck %s --check-prefixes=CHECK,SVE-OR-SME
+; RUN: llc -mattr=+sme -force-streaming < %s | FileCheck %s --check-prefixes=CHECK,SVE-OR-SME
+; RUN: llc -force-streaming-compatible < %s | FileCheck %s --check-prefixes=CHECK,STREAMING-COMPAT
+
+target triple = "aarch64-unknown-linux-gnu"
@result = dso_local global i32 0, align 4
@@ -156,32 +160,55 @@ define dso_local i32 @all_arg(i32 noundef %a, i32 noundef %b, i32 noundef %c) lo
; DEFAULT-NEXT: movi v7.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: all_arg:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: mul w8, w1, w0
-; SVE-NEXT: mov x1, #0 // =0x0
-; SVE-NEXT: mov x3, #0 // =0x0
-; SVE-NEXT: mov x4, #0 // =0x0
-; SVE-NEXT: mov x5, #0 // =0x0
-; SVE-NEXT: mov x6, #0 // =0x0
-; SVE-NEXT: mov x7, #0 // =0x0
-; SVE-NEXT: mov x18, #0 // =0x0
-; SVE-NEXT: mov z0.d, #0 // =0x0
-; SVE-NEXT: orr w0, w8, w2
-; SVE-NEXT: mov x2, #0 // =0x0
-; SVE-NEXT: mov x8, #0 // =0x0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: mov z2.d, #0 // =0x0
-; SVE-NEXT: mov z3.d, #0 // =0x0
-; SVE-NEXT: mov z4.d, #0 // =0x0
-; SVE-NEXT: mov z5.d, #0 // =0x0
-; SVE-NEXT: mov z6.d, #0 // =0x0
-; SVE-NEXT: mov z7.d, #0 // =0x0
-; SVE-NEXT: pfalse p0.b
-; SVE-NEXT: pfalse p1.b
-; SVE-NEXT: pfalse p2.b
-; SVE-NEXT: pfalse p3.b
-; SVE-NEXT: ret
+; SVE-OR-SME-LABEL: all_arg:
+; SVE-OR-SME: // %bb.0: // %entry
+; SVE-OR-SME-NEXT: mul w8, w1, w0
+; SVE-OR-SME-NEXT: mov x1, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x3, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x4, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x5, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x6, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x7, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x18, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z0.d, #0 // =0x0
+; SVE-OR-SME-NEXT: orr w0, w8, w2
+; SVE-OR-SME-NEXT: mov x2, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x8, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z1.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z2.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z3.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z4.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z5.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z6.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z7.d, #0 // =0x0
+; SVE-OR-SME-NEXT: pfalse p0.b
+; SVE-OR-SME-NEXT: pfalse p1.b
+; SVE-OR-SME-NEXT: pfalse p2.b
+; SVE-OR-SME-NEXT: pfalse p3.b
+; SVE-OR-SME-NEXT: ret
+;
+; STREAMING-COMPAT-LABEL: all_arg:
+; STREAMING-COMPAT: // %bb.0: // %entry
+; STREAMING-COMPAT-NEXT: mul w8, w1, w0
+; STREAMING-COMPAT-NEXT: mov x1, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x3, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x4, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x5, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x6, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x7, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x18, #0 // =0x0
+; STREAMING-COMPAT-NEXT: fmov d0, xzr
+; STREAMING-COMPAT-NEXT: orr w0, w8, w2
+; STREAMING-COMPAT-NEXT: mov x2, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x8, #0 // =0x0
+; STREAMING-COMPAT-NEXT: fmov d1, xzr
+; STREAMING-COMPAT-NEXT: fmov d2, xzr
+; STREAMING-COMPAT-NEXT: fmov d3, xzr
+; STREAMING-COMPAT-NEXT: fmov d4, xzr
+; STREAMING-COMPAT-NEXT: fmov d5, xzr
+; STREAMING-COMPAT-NEXT: fmov d6, xzr
+; STREAMING-COMPAT-NEXT: fmov d7, xzr
+; STREAMING-COMPAT-NEXT: ret
entry:
%mul = mul nsw i32 %b, %a
@@ -238,69 +265,117 @@ define dso_local i32 @all(i32 noundef %a, i32 noundef %b, i32 noundef %c) local_
; DEFAULT-NEXT: movi v31.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: all:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: mul w8, w1, w0
-; SVE-NEXT: mov x1, #0 // =0x0
-; SVE-NEXT: mov x3, #0 // =0x0
-; SVE-NEXT: mov x4, #0 // =0x0
-; SVE-NEXT: mov x5, #0 // =0x0
-; SVE-NEXT: mov x6, #0 // =0x0
-; SVE-NEXT: mov x7, #0 // =0x0
-; SVE-NEXT: mov x9, #0 // =0x0
-; SVE-NEXT: mov x10, #0 // =0x0
-; SVE-NEXT: orr w0, w8, w2
-; SVE-NEXT: mov x2, #0 // =0x0
-; SVE-NEXT: mov x8, #0 // =0x0
-; SVE-NEXT: mov x11, #0 // =0x0
-; SVE-NEXT: mov x12, #0 // =0x0
-; SVE-NEXT: mov x13, #0 // =0x0
-; SVE-NEXT: mov x14, #0 // =0x0
-; SVE-NEXT: mov x15, #0 // =0x0
-; SVE-NEXT: mov x16, #0 // =0x0
-; SVE-NEXT: mov x17, #0 // =0x0
-; SVE-NEXT: mov x18, #0 // =0x0
-; SVE-NEXT: mov z0.d, #0 // =0x0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: mov z2.d, #0 // =0x0
-; SVE-NEXT: mov z3.d, #0 // =0x0
-; SVE-NEXT: mov z4.d, #0 // =0x0
-; SVE-NEXT: mov z5.d, #0 // =0x0
-; SVE-NEXT: mov z6.d, #0 // =0x0
-; SVE-NEXT: mov z7.d, #0 // =0x0
-; SVE-NEXT: mov z16.d, #0 // =0x0
-; SVE-NEXT: mov z17.d, #0 // =0x0
-; SVE-NEXT: mov z18.d, #0 // =0x0
-; SVE-NEXT: mov z19.d, #0 // =0x0
-; SVE-NEXT: mov z20.d, #0 // =0x0
-; SVE-NEXT: mov z21.d, #0 // =0x0
-; SVE-NEXT: mov z22.d, #0 // =0x0
-; SVE-NEXT: mov z23.d, #0 // =0x0
-; SVE-NEXT: mov z24.d, #0 // =0x0
-; SVE-NEXT: mov z25.d, #0 // =0x0
-; SVE-NEXT: mov z26.d, #0 // =0x0
-; SVE-NEXT: mov z27.d, #0 // =0x0
-; SVE-NEXT: mov z28.d, #0 // =0x0
-; SVE-NEXT: mov z29.d, #0 // =0x0
-; SVE-NEXT: mov z30.d, #0 // =0x0
-; SVE-NEXT: mov z31.d, #0 // =0x0
-; SVE-NEXT: pfalse p0.b
-; SVE-NEXT: pfalse p1.b
-; SVE-NEXT: pfalse p2.b
-; SVE-NEXT: pfalse p3.b
-; SVE-NEXT: pfalse p4.b
-; SVE-NEXT: pfalse p5.b
-; SVE-NEXT: pfalse p6.b
-; SVE-NEXT: pfalse p7.b
-; SVE-NEXT: pfalse p8.b
-; SVE-NEXT: pfalse p9.b
-; SVE-NEXT: pfalse p10.b
-; SVE-NEXT: pfalse p11.b
-; SVE-NEXT: pfalse p12.b
-; SVE-NEXT: pfalse p13.b
-; SVE-NEXT: pfalse p14.b
-; SVE-NEXT: pfalse p15.b
-; SVE-NEXT: ret
+; SVE-OR-SME-LABEL: all:
+; SVE-OR-SME: // %bb.0: // %entry
+; SVE-OR-SME-NEXT: mul w8, w1, w0
+; SVE-OR-SME-NEXT: mov x1, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x3, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x4, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x5, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x6, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x7, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x9, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x10, #0 // =0x0
+; SVE-OR-SME-NEXT: orr w0, w8, w2
+; SVE-OR-SME-NEXT: mov x2, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x8, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x11, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x12, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x13, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x14, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x15, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x16, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x17, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x18, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z0.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z1.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z2.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z3.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z4.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z5.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z6.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z7.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z16.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z17.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z18.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z19.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z20.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z21.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z22.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z23.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z24.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z25.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z26.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z27.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z28.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z29.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z30.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z31.d, #0 // =0x0
+; SVE-OR-SME-NEXT: pfalse p0.b
+; SVE-OR-SME-NEXT: pfalse p1.b
+; SVE-OR-SME-NEXT: pfalse p2.b
+; SVE-OR-SME-NEXT: pfalse p3.b
+; SVE-OR-SME-NEXT: pfalse p4.b
+; SVE-OR-SME-NEXT: pfalse p5.b
+; SVE-OR-SME-NEXT: pfalse p6.b
+; SVE-OR-SME-NEXT: pfalse p7.b
+; SVE-OR-SME-NEXT: pfalse p8.b
+; SVE-OR-SME-NEXT: pfalse p9.b
+; SVE-OR-SME-NEXT: pfalse p10.b
+; SVE-OR-SME-NEXT: pfalse p11.b
+; SVE-OR-SME-NEXT: pfalse p12.b
+; SVE-OR-SME-NEXT: pfalse p13.b
+; SVE-OR-SME-NEXT: pfalse p14.b
+; SVE-OR-SME-NEXT: pfalse p15.b
+; SVE-OR-SME-NEXT: ret
+;
+; STREAMING-COMPAT-LABEL: all:
+; STREAMING-COMPAT: // %bb.0: // %entry
+; STREAMING-COMPAT-NEXT: mul w8, w1, w0
+; STREAMING-COMPAT-NEXT: mov x1, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x3, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x4, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x5, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x6, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x7, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x9, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x10, #0 // =0x0
+; STREAMING-COMPAT-NEXT: orr w0, w8, w2
+; STREAMING-COMPAT-NEXT: mov x2, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x8, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x11, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x12, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x13, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x14, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x15, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x16, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x17, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x18, #0 // =0x0
+; STREAMING-COMPAT-NEXT: fmov d0, xzr
+; STREAMING-COMPAT-NEXT: fmov d1, xzr
+; STREAMING-COMPAT-NEXT: fmov d2, xzr
+; STREAMING-COMPAT-NEXT: fmov d3, xzr
+; STREAMING-COMPAT-NEXT: fmov d4, xzr
+; STREAMING-COMPAT-NEXT: fmov d5, xzr
+; STREAMING-COMPAT-NEXT: fmov d6, xzr
+; STREAMING-COMPAT-NEXT: fmov d7, xzr
+; STREAMING-COMPAT-NEXT: fmov d16, xzr
+; STREAMING-COMPAT-NEXT: fmov d17, xzr
+; STREAMING-COMPAT-NEXT: fmov d18, xzr
+; STREAMING-COMPAT-NEXT: fmov d19, xzr
+; STREAMING-COMPAT-NEXT: fmov d20, xzr
+; STREAMING-COMPAT-NEXT: fmov d21, xzr
+; STREAMING-COMPAT-NEXT: fmov d22, xzr
+; STREAMING-COMPAT-NEXT: fmov d23, xzr
+; STREAMING-COMPAT-NEXT: fmov d24, xzr
+; STREAMING-COMPAT-NEXT: fmov d25, xzr
+; STREAMING-COMPAT-NEXT: fmov d26, xzr
+; STREAMING-COMPAT-NEXT: fmov d27, xzr
+; STREAMING-COMPAT-NEXT: fmov d28, xzr
+; STREAMING-COMPAT-NEXT: fmov d29, xzr
+; STREAMING-COMPAT-NEXT: fmov d30, xzr
+; STREAMING-COMPAT-NEXT: fmov d31, xzr
+; STREAMING-COMPAT-NEXT: ret
entry:
%mul = mul nsw i32 %b, %a
@@ -355,12 +430,19 @@ define dso_local double @used_arg_float(double noundef %a, float noundef %b) loc
; DEFAULT-NEXT: movi v1.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: used_arg_float:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: fcvt d1, s1
-; SVE-NEXT: fmul d0, d1, d0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: ret
+; SVE-OR-SME-LABEL: used_arg_float:
+; SVE-OR-SME: // %bb.0: // %entry
+; SVE-OR-SME-NEXT: fcvt d1, s1
+; SVE-OR-SME-NEXT: fmul d0, d1, d0
+; SVE-OR-SME-NEXT: mov z1.d, #0 // =0x0
+; SVE-OR-SME-NEXT: ret
+;
+; STREAMING-COMPAT-LABEL: used_arg_float:
+; STREAMING-COMPAT: // %bb.0: // %entry
+; STREAMING-COMPAT-NEXT: fcvt d1, s1
+; STREAMING-COMPAT-NEXT: fmul d0, d1, d0
+; STREAMING-COMPAT-NEXT: fmov d1, xzr
+; STREAMING-COMPAT-NEXT: ret
entry:
%conv = fpext float %b to double
@@ -376,12 +458,19 @@ define dso_local double @used_float(double noundef %a, float noundef %b) local_u
; DEFAULT-NEXT: movi v1.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: used_float:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: fcvt d1, s1
-; SVE-NEXT: fmul d0, d1, d0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: ret
+; SVE-OR-SME-LABEL: used_float:
+; SVE-OR-SME: // %bb.0: // %entry
+; SVE-OR-SME-NEXT: fcvt d1, s1
+; SVE-OR-SME-NEXT: fmul d0, d1, d0
+; SVE-OR-SME-NEXT: mov z1.d, #0 // =0x0
+; SVE-OR-SME-NEXT: ret
+;
+; STREAMING-COMPAT-LABEL: used_float:
+; STREAMING-COMPAT: // %bb.0: // %entry
+; STREAMING-COMPAT-NEXT: fcvt d1, s1
+; STREAMING-COMPAT-NEXT: fmul d0, d1, d0
+; STREAMING-COMPAT-NEXT: fmov d1, xzr
+; STREAMING-COMPAT-NEXT: ret
entry:
%conv = fpext float %b to double
@@ -468,32 +557,55 @@ define dso_local double @all_arg_float(double noundef %a, float noundef %b) loca
; DEFAULT-NEXT: movi v7.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: all_arg_float:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: fcvt d1, s1
-; SVE-NEXT: fmul d0, d1, d0
-; SVE-NEXT: mov x0, #0 // =0x0
-; SVE-NEXT: mov x1, #0 // =0x0
-; SVE-NEXT: mov x2, #0 // =0x0
-; SVE-NEXT: mov x3, #0 // =0x0
-; SVE-NEXT: mov x4, #0 // =0x0
-; SVE-NEXT: mov x5, #0 // =0x0
-; SVE-NEXT: mov x6, #0 // =0x0
-; SVE-NEXT: mov x7, #0 // =0x0
-; SVE-NEXT: mov x8, #0 // =0x0
-; SVE-NEXT: mov x18, #0 // =0x0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: mov z2.d, #0 // =0x0
-; SVE-NEXT: mov z3.d, #0 // =0x0
-; SVE-NEXT: mov z4.d, #0 // =0x0
-; SVE-NEXT: mov z5.d, #0 // =0x0
-; SVE-NEXT: mov z6.d, #0 // =0x0
-; SVE-NEXT: mov z7.d, #0 // =0x0
-; SVE-NEXT: pfalse p0.b
-; SVE-NEXT: pfalse p1.b
-; SVE-NEXT: pfalse p2.b
-; SVE-NEXT: pfalse p3.b
-; SVE-NEXT: ret
+; SVE-OR-SME-LABEL: all_arg_float:
+; SVE-OR-SME: // %bb.0: // %entry
+; SVE-OR-SME-NEXT: fcvt d1, s1
+; SVE-OR-SME-NEXT: fmul d0, d1, d0
+; SVE-OR-SME-NEXT: mov x0, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x1, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x2, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x3, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x4, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x5, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x6, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x7, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x8, #0 // =0x0
+; SVE-OR-SME-NEXT: mov x18, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z1.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z2.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z3.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z4.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z5.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z6.d, #0 // =0x0
+; SVE-OR-SME-NEXT: mov z7.d, #0 // =0x0
+; SVE-OR-SME-NEXT: pfalse p0.b
+; SVE-OR-SME-NEXT: pfalse p1.b
+; SVE-OR-SME-NEXT: pfalse p2.b
+; SVE-OR-SME-NEXT: pfalse p3.b
+; SVE-OR-SME-NEXT: ret
+;
+; STREAMING-COMPAT-LABEL: all_arg_float:
+; STREAMING-COMPAT: // %bb.0: // %entry
+; STREAMING-COMPAT-NEXT: fcvt d1, s1
+; STREAMING-COMPAT-NEXT: fmul d0, d1, d0
+; STREAMING-COMPAT-NEXT: mov x0, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x1, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x2, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x3, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x4, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x5, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x6, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x7, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x8, #0 // =0x0
+; STREAMING-COMPAT-NEXT: mov x18, #0 // =0x0
+; STREAMING-COMPAT-NEXT: fmov d1, xzr
+; STREAMING-COMPAT-NEXT: fmov d2, xzr
+; STREAMING-COMPAT-NEXT: fmov d3, xzr
+; STREAMING-COMPAT-NEXT: fmov d4, xzr
+; STREAMING-COMPAT-NEXT: fmov d5, xzr
+; STREAMING-COMPAT-NEXT: fmov d6, xzr
+; STREAMING-COMPAT-NEXT: fmov d7, xzr
+; STREAMING-COMPAT-NEXT: ret
entry:
%conv = fpext float %b to double
@@ -550,69 +662,117 @@ define dso_local double @all_float(double noundef %a, float noundef %b) local_un
; DEFAULT-NEXT: movi v31.2d, #0000000000000000
; DEFAULT-NEXT: ret
;
-; SVE-LABEL: all_float:
-; SVE: // %bb.0: // %entry
-; SVE-NEXT: fcvt d1, s1
-; SVE-NEXT: fmul d0, d1, d0
-; SVE-NEXT: mov x0, #0 // =0x0
-; SVE-NEXT: mov x1, #0 // =0x0
-; SVE-NEXT: mov x2, #0 // =0x0
-; SVE-NEXT: mov x3, #0 // =0x0
-; SVE-NEXT: mov x4, #0 // =0x0
-; SVE-NEXT: mov x5, #0 // =0x0
-; SVE-NEXT: mov x6, #0 // =0x0
-; SVE-NEXT: mov x7, #0 // =0x0
-; SVE-NEXT: mov x8, #0 // =0x0
-; SVE-NEXT: mov x9, #0 // =0x0
-; SVE-NEXT: mov x10, #0 // =0x0
-; SVE-NEXT: mov x11, #0 // =0x0
-; SVE-NEXT: mov x12, #0 // =0x0
-; SVE-NEXT: mov x13, #0 // =0x0
-; SVE-NEXT: mov x14, #0 // =0x0
-; SVE-NEXT: mov x15, #0 // =0x0
-; SVE-NEXT: mov x16, #0 // =0x0
-; SVE-NEXT: mov x17, #0 // =0x0
-; SVE-NEXT: mov x18, #0 // =0x0
-; SVE-NEXT: mov z1.d, #0 // =0x0
-; SVE-NEXT: mov z2.d, #0 // =0x0
-; SVE-NEXT: mov z3.d, #0 // =0x0
-; SVE-NEXT: mov z4.d, #0 // =0x0
-; SVE-NEXT: mov z5.d, #0 // =0x0
-; SVE-NEXT: mov z6.d, #0 // =0x0
-; SVE-NEXT: mov z7.d, #0 // =0x0
-; SVE-NEXT: mov z16.d, #0 // =0x0
-; SVE-NEXT: mov z17.d, #0 // =0x0
-; SVE-NEXT: mov z18.d, #0 // =0x0
-; SVE-NEXT: mov z19.d, #0 // =0x0
-; SVE-NEXT: mov z20.d, #0 // =0x0
-; SVE-NEXT: mov z21.d, #0 // =0x0
-; SVE-NEXT: mov z22.d, #0 // =0x0
-; SVE-NEXT: mov z23.d, #0 // =0x0
-; SVE-NEXT: mov z24.d, #0 // =0x0
-; SVE-NEXT: mov z25.d, #0 // =0x0
-; SVE-NEXT: mov z26.d, #0 // =0x0
-; SVE-NEXT: mov z27.d, #0 // =0x0
-; SVE-NEXT: mov z28.d, #0 // =0x0
-; SVE-NEXT: mov z29.d, #0 // =0x0
-; SVE-NEXT: mov z30.d, #0 // =0x0
-; SVE-NEXT: mov z31.d, #0 // =0x0
-; SVE-NEXT: pfalse p0.b
-; SVE-NEXT: pfalse p1.b
-; SVE-NEXT: pfalse p2.b
-; SVE-NEXT: pfalse p3.b
-; SVE-NEXT: pfalse p4.b
-; SVE-NEXT: pfalse p5.b
-; SVE-NEXT: pfalse p6.b
-; SVE-NEXT: pfalse p7.b
-; SVE-NEXT: pfalse p8.b
-; SVE-NEXT:...
[truncated]
|
MacDue
commented
Nov 20, 2024
Comment on lines
+9711
to
+9713
| // This is a streaming-compatible function without SVE. We don't have full | ||
| // Neon (just FPRs), so we can at most use the first 64-bit sub-register. | ||
| // So given `movi v..` would be illegal use `fmov d..` instead. |
Member
Author
There was a problem hiding this comment.
I'm pretty sure this is correct. I don't think there's a way to zero the full 128 bits without Neon or SVE in a streaming-compatible function (we at most have the D registers).
Collaborator
There was a problem hiding this comment.
fmovd (and all other instructions that set a d register) will implicitly zero the upper bits of the register too, so it should be OK even if other parts of the register do get used.
sdesmalen-arm
approved these changes
Nov 20, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, with
-fzero-call-used-regsclang/LLVM would incorrectly emit Neon instructions in streaming functions, and streaming-compatible functions without SVE.With this change:
movi v..)