[RISCV] add the MC layer support of Zfinx extension

[achieveartificialintelligence]

Summary

This patch added the MC layer support of Zfinx extension.

Authored-by: StephenFan
Co-Authored-by: Shao-Ce Sun

[achieveartificialintelligence]

Fix

[achieveartificialintelligence]

Add arch info

[achieveartificialintelligence]

Fix

[MaskRay]

llvm/lib/Target/RISCV/Disassembler/RISCVDisassembler.cpp

164

Use functionName.

There is inconsistency in the code base but functionName is used much more than FunctionName.

186

https://llvm.org/docs/CodingStandards.html#don-t-use-braces-on-simple-single-statement-bodies-of-if-else-loop-statements

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

181

The indentation is inconsistent. Also applies to code above.

llvm/test/MC/RISCV/rv64zhinx-valid.s

3 ↗

(On Diff #379273)

llvm-mc can use %s instead of < %s

I heard that not using < is a bit more convenient for some Windows users. (They do suffer from llc and opt's prevailing < %s)

12 ↗

(On Diff #379273)

[[@LINE+1]] is a deprecated form.
Use [[#@LINE+1]]. Applies to many places elsewhere.

llvm/test/MC/RISCV/rvzhinx-aliases-valid.s

47

What is {{[[:space:]]}} used for?

[achieveartificialintelligence]

llvm/lib/Target/RISCV/Disassembler/RISCVDisassembler.cpp

164

I fix this in D112520

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

561

After D95588 is accepted, we will reuse this.

[achieveartificialintelligence]

Rebase

[achieveartificialintelligence]

Address @MaskRay 's comments.

[achieveartificialintelligence]

Rebase

[achieveartificialintelligence]

Rebase

[achieveartificialintelligence]

Ping

[achieveartificialintelligence]

Rebase and Ping.

[craig.topper]

llvm/lib/Support/RISCVISAInfo.cpp

68

Do we need to enforce that these can't be mixed with F, D, and Zfh?

llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp

1800

D and Zfh imply F. Is it enough to just check F?

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

103–104

Can we merge this with FPFMAD_rrr_frm_single and FPFMAH_rrr_frm_single by passing the 0b00/0b01/0b10 value from FPFMAS_rrr_frm/FPFMAD_rrr_frm/FPFMAH_rrr_frm?

This applies to most of the _single classes. We should share them if possible.

[craig.topper]

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

199

Aren't these aliases only valid for RV64. You need IN32X aliases for RV32 right?

[achieveartificialintelligence]

Address @craig.topper 's comments.

[achieveartificialintelligence]

Gentle Ping and Rebase.

[achieveartificialintelligence]

Update Part of Zfinx Codes

[achieveartificialintelligence]

I'll finish the rest of code, if this patch is acceptable.

[achieveartificialintelligence]

Ping.

[craig.topper]

llvm/lib/Support/RISCVISAInfo.cpp

290

@kito-cheng or @khchen is the right way to do this? Or should it be in updateImplication? Where do we have Zfhmin imply f or d imply f today?

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

34

Can you add let RenderMethod = "addRegOperands"; here. That should let you remove addGPRAsFPROperands from RISCVAsmParser.cpp. Similar for the other AsmOperandClasses.

[craig.topper]

llvm/lib/Target/RISCV/Disassembler/RISCVDisassembler.cpp

170

Can we make use of let DecoderMethod = on RegisterClass to avoid these aliases?

[achieveartificialintelligence]

Address @craig.topper 's comments.

[achieveartificialintelligence]

Rebase

[craig.topper]

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

123

Can we use DAGOperand instead of RegisterOperand? I think that should allow you to avoid adding GPROp, FPR32Op, FPR64Op, and FPR16Op. You can just use the RegisterClasses. Both RegisterClass and RegisterOperand inherit from DAGOperand.

[craig.topper]

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

228

Why do the aliases not apply to Zfinx?

[achieveartificialintelligence]

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

228

Because the first argument is FPFMA_rrr_frm Inst, if we use multiclass here, it'll be FFPFMA_rrr_frm Inst, FFPFMA_rrr_frm Inst_INX. I don't know is there a better way?

[achieveartificialintelligence]

Use DAGOperand instead of RegisterOperand

[achieveartificialintelligence]

Fix issue in RISCVISAInfo.cpp

[achieveartificialintelligence]

Inspired by D116019, use multiclass in Alias

[achieveartificialintelligence]

update

[achieveartificialintelligence]

Ping. Zfinx has been ratified, could we spend some time on this patch?

[achieveartificialintelligence]

update

[craig.topper]

llvm/lib/Support/RISCVISAInfo.cpp

687

Do we need the equivalent of this for Zfinx and Zdinx?

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

199

Don't we need IN32X aliases too?

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

574

Is possible to reference GPR here to avoid listing the allocation order again? Similar to how GPRNoX0 does (sub GPR, X0). Not sure if there is a way to be exactly another list.

[achieveartificialintelligence]

Address @craig.topper's comments

[achieveartificialintelligence]

Rebase

[achieveartificialintelligence]

rebase

[achieveartificialintelligence]

Ping. Any other suggestions?

[craig.topper]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

136

Does putting this back the way it was and using "(add GPR)" in GPRF16, GPRF32, and GPRF64 work?

[achieveartificialintelligence]

Address @craig.topper's comments. Thanks!

[asb]

Thanks for your work on this. The way you've managed to use multiclasses to handle this with the 'ExtInfo' definitions takes a bit of unpicking to follow, but it does a really good job of keeping the instruction definitions largely unmodified. Nice!

I've got a few requests/questions on the test coverage:

• I'm not sure the inclusion of the load/store instructions in the z[d,f,h]inx test cases has much value, given those instructions are unmodified?
• Given that the set of F/D/Zfh instructions that _aren't_ supported under z[d,f,h]inx is relatively short, it's probably worth being exhaustive in the *-invalid.s test cases. It's only 8 instructions for Zfinx for instance.
• The *-invalid.s test cases should have coverage for using a floating point register with an operation that is supported by Z[d,f,h]inx
• The zdinx test cases don't include any coverage for when odd registers are used, but I think it should

[asb]

Not an issue for this MC-layer patch, but I've created https://github.com/riscv/riscv-zfinx/issues/14 to point out what seems to be an incorrect statement about the status quo on the ABI for 32-bit floating point types on RV64 in the Zfinx spec.

[achieveartificialintelligence]

@asb Thank you very much! Could you please take a look at what needs to be modified now?

[achieveartificialintelligence]

Separate the Zhinxmin and Zhinx extensions. Inspired by D118581

[luismarques]

Assuming this will be merged soon, do you want to submit a backport request for the 14.0 branch?

[achieveartificialintelligence]

In D93298#3296515, @luismarques wrote:

Assuming this will be merged soon, do you want to submit a backport request for the 14.0 branch?

Yes.

[Jim]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

578

Is XLenRI correct for GPRPF64? RV32 has size 32.

[craig.topper]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

578

I'm not sure the RegInfos are correct for the other classes either. The RegInfos override the register size, the spill size, and spill alignment. I think those should all be based on the FP type.

So I don't think any of these should have a RegInfos.

[achieveartificialintelligence]

Address comments.

[achieveartificialintelligence]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

578

Without RegInfo, it would cause spill size to be changed for a large number of other tests.

[Jim]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

580

Is register pair only on RV32 for used as f64?

[achieveartificialintelligence]

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

580

Yes.

[asb]

I think all my comments have been addressed. @craig.topper - are you happy your RegInfo question is addressed?

[craig.topper]

In D93298#3326230, @asb wrote:

I think all my comments have been addressed. @craig.topper - are you happy your RegInfo question is addressed?

I'm happy.

[asb]

In that case, LGTM (needs a rebase though). Thanks for your patience on this @achieveartificialintelligence.

[achieveartificialintelligence]

Thanks all!

[krasimir]

It appears that this is causing an assertion segfault in a rustc test over at our experimental rust + llvm@head bot:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/8430#167e6de5-2dd5-41c3-87d7-b6e3f3908371/262-706
The test is https://github.com/rust-lang/rust/blob/master/src/test/assembly/asm/riscv-types.rs. These two lines appear to cause it (code compiles fine when removed):

• check_reg!(a0_f32 f32 "a0" "mv"); https://github.com/rust-lang/rust/blob/f838a425e3134d036a7d9632935111a569ac7446/src/test/assembly/asm/riscv-types.rs#L178
• check_reg!(a0_f64 f64 "a0" "mv"); https://github.com/rust-lang/rust/blob/f838a425e3134d036a7d9632935111a569ac7446/src/test/assembly/asm/riscv-types.rs#L192

The assertion:

Impossible reg-to-reg copy
UNREACHABLE executed at [...]/rust/src/llvm-project/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp:350

The IR for riscv-types.rs:

; ModuleID = 'riscv_types.4cedf4b7-cgu.0'
source_filename = "riscv_types.4cedf4b7-cgu.0"
target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
target triple = "riscv64"

@extern_static = external dso_local global i8
@alloc55 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"reg_i8" }>, align 1
@alloc56 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_i16" }>, align 1
@alloc57 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_i32" }>, align 1
@alloc58 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_f32" }>, align 1
@alloc59 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_i64" }>, align 1
@alloc60 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_f64" }>, align 1
@alloc61 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"reg_ptr" }>, align 1
@alloc62 = private unnamed_addr constant <{ [8 x i8] }> <{ [8 x i8] c"freg_f32" }>, align 1
@alloc63 = private unnamed_addr constant <{ [8 x i8] }> <{ [8 x i8] c"freg_f64" }>, align 1
@alloc64 = private unnamed_addr constant <{ [5 x i8] }> <{ [5 x i8] c"a0_i8" }>, align 1
@alloc65 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_i16" }>, align 1
@alloc66 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_i32" }>, align 1
@alloc67 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_f32" }>, align 1
@alloc68 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_i64" }>, align 1
@alloc69 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_f64" }>, align 1
@alloc70 = private unnamed_addr constant <{ [6 x i8] }> <{ [6 x i8] c"a0_ptr" }>, align 1
@alloc71 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"fa0_f32" }>, align 1
@alloc72 = private unnamed_addr constant <{ [7 x i8] }> <{ [7 x i8] c"fa0_f64" }>, align 1

; Function Attrs: nounwind
define dso_local void @sym_fn() unnamed_addr #0 {
start:
  tail call void asm sideeffect alignstack "call ${0:c}", "s,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(void ()* nonnull @extern_func) #1, !srcloc !1
  ret void
}

; Function Attrs: nounwind
define dso_local void @sym_static() unnamed_addr #0 {
start:
  tail call void asm sideeffect alignstack "lb t0, ${0:c}", "s,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i8* nonnull @extern_static) #1, !srcloc !2
  ret void
}

; Function Attrs: nounwind
define dso_local i8 @reg_i8(i8 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc55 to [0 x i8]*), i64 6) #1
  %0 = tail call i8 asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i8 %x) #1, !srcloc !3
  ret i8 %0
}

; Function Attrs: nounwind
define dso_local i16 @reg_i16(i16 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc56 to [0 x i8]*), i64 7) #1
  %0 = tail call i16 asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i16 %x) #1, !srcloc !3
  ret i16 %0
}

; Function Attrs: nounwind
define dso_local i32 @reg_i32(i32 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc57 to [0 x i8]*), i64 7) #1
  %0 = tail call i32 asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i32 %x) #1, !srcloc !3
  ret i32 %0
}

; Function Attrs: nounwind
define dso_local float @reg_f32(float %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc58 to [0 x i8]*), i64 7) #1
  %0 = tail call float asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(float %x) #1, !srcloc !3
  ret float %0
}

; Function Attrs: nounwind
define dso_local i64 @reg_i64(i64 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc59 to [0 x i8]*), i64 7) #1
  %0 = tail call i64 asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i64 %x) #1, !srcloc !3
  ret i64 %0
}

; Function Attrs: nounwind
define dso_local double @reg_f64(double %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc60 to [0 x i8]*), i64 7) #1
  %0 = tail call double asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(double %x) #1, !srcloc !3
  ret double %0
}

; Function Attrs: nounwind
define dso_local i8* @reg_ptr(i8* %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc61 to [0 x i8]*), i64 7) #1
  %0 = tail call i8* asm sideeffect alignstack "mv ${0}, ${1}", "=&r,r,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i8* %x) #1, !srcloc !3
  ret i8* %0
}

; Function Attrs: nounwind
define dso_local float @freg_f32(float %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [8 x i8] }>* @alloc62 to [0 x i8]*), i64 8) #1
  %0 = tail call float asm sideeffect alignstack "fmv.s ${0}, ${1}", "=&f,f,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(float %x) #1, !srcloc !3
  ret float %0
}

; Function Attrs: nounwind
define dso_local double @freg_f64(double %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [8 x i8] }>* @alloc63 to [0 x i8]*), i64 8) #1
  %0 = tail call double asm sideeffect alignstack "fmv.d ${0}, ${1}", "=&f,f,~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(double %x) #1, !srcloc !3
  ret double %0
}

; Function Attrs: nounwind
define dso_local i8 @a0_i8(i8 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [5 x i8] }>* @alloc64 to [0 x i8]*), i64 5) #1
  %0 = tail call i8 asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i8 %x) #1, !srcloc !4
  ret i8 %0
}

; Function Attrs: nounwind
define dso_local i16 @a0_i16(i16 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc65 to [0 x i8]*), i64 6) #1
  %0 = tail call i16 asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i16 %x) #1, !srcloc !4
  ret i16 %0
}

; Function Attrs: nounwind
define dso_local i32 @a0_i32(i32 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc66 to [0 x i8]*), i64 6) #1
  %0 = tail call i32 asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i32 %x) #1, !srcloc !4
  ret i32 %0
}

; Function Attrs: nounwind
define dso_local float @a0_f32(float %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc67 to [0 x i8]*), i64 6) #1
  %0 = tail call float asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(float %x) #1, !srcloc !4
  ret float %0
}

; Function Attrs: nounwind
define dso_local i64 @a0_i64(i64 %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc68 to [0 x i8]*), i64 6) #1
  %0 = tail call i64 asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i64 %x) #1, !srcloc !4
  ret i64 %0
}

; Function Attrs: nounwind
define dso_local double @a0_f64(double %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc69 to [0 x i8]*), i64 6) #1
  %0 = tail call double asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(double %x) #1, !srcloc !4
  ret double %0
}

; Function Attrs: nounwind
define dso_local i8* @a0_ptr(i8* %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [6 x i8] }>* @alloc70 to [0 x i8]*), i64 6) #1
  %0 = tail call i8* asm sideeffect alignstack "mv a0, a0", "={x10},{x10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(i8* %x) #1, !srcloc !4
  ret i8* %0
}

; Function Attrs: nounwind
define dso_local float @fa0_f32(float %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc71 to [0 x i8]*), i64 7) #1
  %0 = tail call float asm sideeffect alignstack "fmv.s fa0, fa0", "={f10},{f10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(float %x) #1, !srcloc !4
  ret float %0
}

; Function Attrs: nounwind
define dso_local double @fa0_f64(double %x) unnamed_addr #0 {
start:
  tail call void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1 bitcast (<{ [7 x i8] }>* @alloc72 to [0 x i8]*), i64 7) #1
  %0 = tail call double asm sideeffect alignstack "fmv.d fa0, fa0", "={f10},{f10},~{vtype},~{vl},~{vxsat},~{vxrm},~{memory}"(double %x) #1, !srcloc !4
  ret double %0
}

; Function Attrs: nounwind
declare dso_local void @extern_func() unnamed_addr #0

; Function Attrs: nounwind
declare dso_local void @dont_merge([0 x i8]* noalias noundef nonnull readonly align 1, i64) unnamed_addr #0

attributes #0 = { nounwind "target-cpu"="generic-rv64" }
attributes #1 = { nounwind }

!llvm.module.flags = !{!0}

!0 = !{i32 1, !"Code Model", i32 3}
!1 = !{i32 1048}
!2 = !{i32 1281}
!3 = !{i32 1670}
!4 = !{i32 2113}

@achieveartificialintelligence could you please take a look? Any ideas what's going wrong here?

[jrtc27]

In D93298#3331641, @krasimir wrote:

It appears that this is causing an assertion segfault in a rustc test over at our experimental rust + llvm@head bot:
https://buildkite.com/llvm-project/rust-llvm-integrate-prototype/builds/8430#167e6de5-2dd5-41c3-87d7-b6e3f3908371/262-706
The test is https://github.com/rust-lang/rust/blob/master/src/test/assembly/asm/riscv-types.rs. These two lines appear to cause it (code compiles fine when removed):

• check_reg!(a0_f32 f32 "a0" "mv"); https://github.com/rust-lang/rust/blob/f838a425e3134d036a7d9632935111a569ac7446/src/test/assembly/asm/riscv-types.rs#L178
• check_reg!(a0_f64 f64 "a0" "mv"); https://github.com/rust-lang/rust/blob/f838a425e3134d036a7d9632935111a569ac7446/src/test/assembly/asm/riscv-types.rs#L192

The assertion:

Impossible reg-to-reg copy
UNREACHABLE executed at [...]/rust/src/llvm-project/llvm/lib/Target/RISCV/RISCVInstrInfo.cpp:350

...

This is just MC layer support. Inline assembly interacts with CodeGen; I assume the register classes/constraints still need updating to support Zfinx.

llvm/test/MC/RISCV/rv32i-invalid.s

1

This is an unrelated change

[achieveartificialintelligence]

@krasimir Since I don't have a rust environment, can you help me to test if D120130 works?

[jrtc27]

In D93298#3332020, @achieveartificialintelligence wrote:

@krasimir Since I don't have a rust environment, can you help me to test if D120130 works?

Hm, actually, looking at the log, it doesn't appear to be trying to use Zfinx? My guess is that putting a float in a GPR has broken as that's now legal but should only be legal for Zfinx?

[xbolva00]

It appears that this is causing an assertion segfault in a rustc test over at our experimental rust + llvm@head bot:

I dont think that patch author is required to debug this issue for "experimental rust + llvm@head" - downstream.

Since I don't have a rust environment

There is no requirement, indeed. LLVM devs are not responsible to keep ""experimental rust + llvm@head" working.

[jrtc27]

In D93298#3332038, @xbolva00 wrote:

It appears that this is causing an assertion segfault in a rustc test over at our experimental rust + llvm@head bot:

I dont think that patch author is required to debug this issue for "experimental rust + llvm@head" - downstream.

Since I don't have a rust environment

There is no requirement, indeed. LLVM devs are not responsible to keep ""experimental rust + llvm@head" working.

They're required to assist with fixing regressions that show when compiling legitimate IR. An IR reproducer was given, so no Rust toolchain knowledge should be needed, at that point it's the LLVM dev's responsibility, and if they can't fix it in a timely manner then it should be backed out.

[xbolva00]

In D93298#3332058, @jrtc27 wrote:

In D93298#3332038, @xbolva00 wrote:

It appears that this is causing an assertion segfault in a rustc test over at our experimental rust + llvm@head bot:

I dont think that patch author is required to debug this issue for "experimental rust + llvm@head" - downstream.

Since I don't have a rust environment

There is no requirement, indeed. LLVM devs are not responsible to keep ""experimental rust + llvm@head" working.

They're required to assist with fixing regressions that show when compiling legitimate IR. An IR reproducer was given, so no Rust toolchain knowledge should be needed, at that point it's the LLVM dev's responsibility, and if they can't fix it in a timely manner then it should be backed out.

Then yeah, if IR is provided and is standard one, this is okay. Consider to add that IR as testcase for the fix.

[nikic]

Because I happened to also run into this, reduced IR for -mtriple=riscv32 -mattr=+d is:

define float @test(float %x) {
  %1 = tail call float asm sideeffect alignstack "mv a0, a0", "={x10},{x10}"(float 0.000000e+00)
  ret float 0.000000e+00
}

[krasimir]

@achieveartificialintelligence any progress with the problematic IR? Have you been able to reproduce using nikic's reduced example?

[achieveartificialintelligence]

In D93298#3342452, @krasimir wrote:

@achieveartificialintelligence any progress with the problematic IR? Have you been able to reproduce using nikic's reduced example?

Sorry, I don't have a solution right now.

[krasimir]

In D93298#3342480, @achieveartificialintelligence wrote:

In D93298#3342452, @krasimir wrote:

@achieveartificialintelligence any progress with the problematic IR? Have you been able to reproduce using nikic's reduced example?

Sorry, I don't have a solution right now.

Could we revert this until then?

[craig.topper]

Here's a fix that I tested on the original repreoduce. It detects that we picked one of the new register classes and tries to redirect back to the normal GPR register class if the GPR register class has the same width. I also checked for MVT::Other to use the GPR class if there was no type.

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 3daf2d03de03..4984872139c5 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -10893,7 +10893,28 @@ RISCVTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
     }
   }
 
-  return TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
+  std::pair<Register, const TargetRegisterClass*> Res;
+  Res = TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
+
+  if (Res.second == &RISCV::GPRF32RegClass) {
+    if (!Subtarget.is64Bit() || VT == MVT::Other)
+      return std::make_pair(Res.first, &RISCV::GPRRegClass);
+    return std::make_pair(0, nullptr);
+  }
+
+  if (Res.second == &RISCV::GPRF64RegClass) {
+    if (Subtarget.is64Bit() || VT == MVT::Other)
+      return std::make_pair(Res.first, &RISCV::GPRRegClass);
+    return std::make_pair(0, nullptr);
+  }
+
+  if (Res.second == &RISCV::GPRF16RegClass) {
+    if (VT == MVT::Other)
+      return std::make_pair(Res.first, &RISCV::GPRRegClass);
+    return std::make_pair(0, nullptr);
+  }
+
+  return Res;
 }

[achieveartificialintelligence]

Thanks for @craig.topper's patch!

[achieveartificialintelligence]

Can you help test if this patch works fine now? @krasimir

[achieveartificialintelligence]

Rebase

[jrtc27]

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

10923–10924 ↗

(On Diff #411719)

[achieveartificialintelligence]

Address @jrtc27's comments

[achieveartificialintelligence]

Are there any other errors now? @krasimir @nikic

[craig.topper]

llvm/test/CodeGen/RISCV/zfinx-types.ll

5 ↗

(On Diff #411375)

I'm not sure the Zfh or D RUN lines are really doing anything. Especially the Zfh.

The same bug did happen with double, D extension and RV64 so should probably test that.

[achieveartificialintelligence]

Address @craig.topper's comment

[craig.topper]

llvm/test/CodeGen/RISCV/zfinx-types.ll

5 ↗

(On Diff #411375)

I see you added rv64, but didn't add a test case that uses double instead of float.

[krasimir]

@achieveartificialintelligence thank you for looking into this, sorry for the late reply!

It looks like the latest version addresses @nikic's IR reproducer with -mtriple=riscv32.

There seems to be an error trying this out with -mtriple=riscv64:

% cat test.ll
define float @test(float %x) {
  %1 = tail call float asm sideeffect alignstack "mv a0, a0", "={x10},{x10}"(float 0.000000e+00)
  ret float 0.000000e+00
}
% llc -mtriple=riscv32 -mattr=+d test.ll
% llc -mtriple=riscv64 -mattr=+d test.ll
error: couldn't allocate output register for constraint '{x10}'

I'm not sure if this is intended to work -mtriple=riscv64, but judging by the newly added test RUN: llc -mtriple=riscv64 ... line it seems like it should?

[craig.topper]

In D93298#3354313, @krasimir wrote:

@achieveartificialintelligence thank you for looking into this, sorry for the late reply!

It looks like the latest version addresses @nikic's IR reproducer with -mtriple=riscv32.

There seems to be an error trying this out with -mtriple=riscv64:

% cat test.ll
define float @test(float %x) {
  %1 = tail call float asm sideeffect alignstack "mv a0, a0", "={x10},{x10}"(float 0.000000e+00)
  ret float 0.000000e+00
}
% llc -mtriple=riscv32 -mattr=+d test.ll
% llc -mtriple=riscv64 -mattr=+d test.ll
error: couldn't allocate output register for constraint '{x10}'

I'm not sure if this is intended to work -mtriple=riscv64, but judging by the newly added test RUN: llc -mtriple=riscv64 ... line it seems like it should?

My patch was only partially correct. I'll fix it

[craig.topper]

In D93298#3355088, @craig.topper wrote:

In D93298#3354313, @krasimir wrote:

@achieveartificialintelligence thank you for looking into this, sorry for the late reply!

It looks like the latest version addresses @nikic's IR reproducer with -mtriple=riscv32.

There seems to be an error trying this out with -mtriple=riscv64:

% cat test.ll
define float @test(float %x) {
  %1 = tail call float asm sideeffect alignstack "mv a0, a0", "={x10},{x10}"(float 0.000000e+00)
  ret float 0.000000e+00
}
% llc -mtriple=riscv32 -mattr=+d test.ll
% llc -mtriple=riscv64 -mattr=+d test.ll
error: couldn't allocate output register for constraint '{x10}'

I'm not sure if this is intended to work -mtriple=riscv64, but judging by the newly added test RUN: llc -mtriple=riscv64 ... line it seems like it should?

My patch was only partially correct. I'll fix it

I just pushed 6cb42cd6669785f3b611106e1b6b38bbe65733a9 to hopefully fix this.

[jrtc27]

Does a double with r for RV32 work with that fix? That's supposed to give the low half of the register. You might need to also deal with the register pair class?

[jrtc27]

Never mind, I see you added a test for that case

[krasimir]

I just pushed 6cb42cd6669785f3b611106e1b6b38bbe65733a9 to hopefully fix this.

Thank you Craig! That fixed it.

llvm/lib/Support/RISCVISAInfo.cpp

static const RISCVSupportedExtension SupportedExperimentalExtensions[] = {
     {"zbr", RISCVExtensionVersion{0, 93}},
     {"zbs", RISCVExtensionVersion{1, 0}},
     {"zbt", RISCVExtensionVersion{0, 93}},
 
     {"zvlsseg", RISCVExtensionVersion{0, 10}},
 
     {"zfhmin", RISCVExtensionVersion{0, 1}},
     {"zfh", RISCVExtensionVersion{0, 1}},
+
+    {"zfinx", RISCVExtensionVersion{1, 0}},
+    {"zdinx", RISCVExtensionVersion{1, 0}},
+    {"zhinxmin", RISCVExtensionVersion{1, 0}},
+    {"zhinx", RISCVExtensionVersion{1, 0}},
 };
 
 static bool stripExperimentalPrefix(StringRef &Ext) {
   return Ext.consume_front("experimental-");
 }
 
 // This function finds the first character that doesn't belong to a version
 // (e.g. zbe0p93 is extension 'zbe' of version '0p93'). So the function will
void RISCVISAInfo::toFeatures(
     std::vector<StringRef> &Features,
     std::function<StringRef(const Twine &)> StrAlloc) const {
   for (auto &Ext : Exts) {
     StringRef ExtName = Ext.first;
 
     if (ExtName == "i")
       continue;
 
     if (ExtName == "zvlsseg") {
       Features.push_back("+experimental-v");
       Features.push_back("+experimental-zvlsseg");
     } else if (isExperimentalExtension(ExtName)) {
       Features.push_back(StrAlloc("+experimental-" + ExtName));
     } else {
       Features.push_back(StrAlloc("+" + ExtName));
     }
   }
if (HasE && !IsRv32)
     return createStringError(
         errc::invalid_argument,
         "standard user-level extension 'e' requires 'rv32'");
 
   // It's illegal to specify the 'd' (double-precision floating point)
   // extension without also specifying the 'f' (single precision
   // floating-point) extension.
   // TODO: This has been removed in later specs, which specify that D implies F
   if (HasD && !HasF)
     return createStringError(errc::invalid_argument,
                              "d requires f extension to also be specified");
 
   // Additional dependency checks.
   // TODO: The 'q' extension requires rv64.
   // TODO: It is illegal to specify 'e' extensions with 'f' and 'd'.
 
   return Error::success();
 }
 
 static const char *ImpliedExtsV[] = {"zvlsseg"};
 static const char *ImpliedExtsZfh[] = {"zfhmin"};
+static const char *ImpliedExtsZhinx[] = {"zhinxmin"};
 
 struct ImpliedExtsEntry {
   StringLiteral Name;
   ArrayRef<const char *> Exts;
 
   bool operator<(const ImpliedExtsEntry &Other) const {
     return Name < Other.Name;
   }
 
   bool operator<(StringRef Other) const { return Name < Other; }
 };
 
 static constexpr ImpliedExtsEntry ImpliedExts[] = {
     {{"v"}, {ImpliedExtsV}},
     {{"zfh"}, {ImpliedExtsZfh}},
+    {{"zhinx"}, {ImpliedExtsZhinx}},
 };
 
 void RISCVISAInfo::updateImplication() {
   bool HasE = Exts.count("e") == 1;
   bool HasI = Exts.count("i") == 1;
 
   // If not in e extension and i extension does not exist, i extension is
   // implied

llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp

#include "RISCVGenAsmMatcher.inc"
   OperandMatchResultTy parseOperandWithModifier(OperandVector &Operands);
   OperandMatchResultTy parseBareSymbol(OperandVector &Operands);
   OperandMatchResultTy parseCallSymbol(OperandVector &Operands);
   OperandMatchResultTy parsePseudoJumpSymbol(OperandVector &Operands);
   OperandMatchResultTy parseJALOffset(OperandVector &Operands);
   OperandMatchResultTy parseVTypeI(OperandVector &Operands);
   OperandMatchResultTy parseMaskReg(OperandVector &Operands);
   OperandMatchResultTy parseInsnDirectiveOpcode(OperandVector &Operands);
+  OperandMatchResultTy parseGPRAsFPR(OperandVector &Operands);
 
   bool parseOperand(OperandVector &Operands, StringRef Mnemonic);
 
   bool parseDirectiveOption();
   bool parseDirectiveAttribute();
   bool parseDirectiveInsn(SMLoc L);
 
   void setFeatureBits(uint64_t Feature, StringRef FeatureString) {
enum class KindTy {
     Register,
     Immediate,
     SystemRegister,
     VType,
   } Kind;
 
   bool IsRV64;
 
+  bool IsGPRAsFPR;
+
   struct RegOp {
     MCRegister RegNum;
   };
 
   struct ImmOp {
     const MCExpr *Val;
   };
 
public:
   bool isMem() const override { return false; }
   bool isSystemRegister() const { return Kind == KindTy::SystemRegister; }
 
   bool isGPR() const {
     return Kind == KindTy::Register &&
            RISCVMCRegisterClasses[RISCV::GPRRegClassID].contains(Reg.RegNum);
   }
 
+  bool isGPRAsFPR() const { return isGPR() && IsGPRAsFPR; }
+
+  bool isGPRF64AsFPR() const { return isGPR() && IsGPRAsFPR && IsRV64; }
+
+  bool isGPRPF64AsFPR() const {
+    return isGPR() && IsGPRAsFPR && !IsRV64 && !((Reg.RegNum - RISCV::X0) & 1);
+  }
+
   static bool evaluateConstantImm(const MCExpr *Expr, int64_t &Imm,
                                   RISCVMCExpr::VariantKind &VK) {
     if (auto *RE = dyn_cast<RISCVMCExpr>(Expr)) {
       VK = RE->getKind();
       return RE->evaluateAsConstant(Imm);
     }
 
     if (auto CE = dyn_cast<MCConstantExpr>(Expr)) {
static std::unique_ptr<RISCVOperand> createToken(StringRef Str, SMLoc S,
     Op->Tok = Str;
     Op->StartLoc = S;
     Op->EndLoc = S;
     Op->IsRV64 = IsRV64;
     return Op;
   }
 
   static std::unique_ptr<RISCVOperand> createReg(unsigned RegNo, SMLoc S,
-                                                 SMLoc E, bool IsRV64) {
+                                                 SMLoc E, bool IsRV64,
+                                                 bool IsGPRAsFPR = false) {
     auto Op = std::make_unique<RISCVOperand>(KindTy::Register);
     Op->Reg.RegNum = RegNo;
     Op->StartLoc = S;
     Op->EndLoc = E;
     Op->IsRV64 = IsRV64;
+    Op->IsGPRAsFPR = IsGPRAsFPR;
     return Op;
   }
 
   static std::unique_ptr<RISCVOperand> createImm(const MCExpr *Val, SMLoc S,
                                                  SMLoc E, bool IsRV64) {
     auto Op = std::make_unique<RISCVOperand>(KindTy::Immediate);
     Op->Imm.Val = Val;
     Op->StartLoc = S;

 }
 
 static MCRegister convertFPR64ToFPR32(MCRegister Reg) {
   assert(Reg >= RISCV::F0_D && Reg <= RISCV::F31_D && "Invalid register");
   return Reg - RISCV::F0_D + RISCV::F0_F;
 }
 
 static MCRegister convertVRToVRMx(const MCRegisterInfo &RI, MCRegister Reg,
                                   unsigned Kind) {
   unsigned RegClassID;
   if (Kind == MCK_VRM2)
     RegClassID = RISCV::VRM2RegClassID;
   else if (Kind == MCK_VRM4)
     RegClassID = RISCV::VRM4RegClassID;
   else if (Kind == MCK_VRM8)
     RegClassID = RISCV::VRM8RegClassID;
   else
unsigned RISCVAsmParser::validateTargetOperandClass(MCParsedAsmOperand &AsmOp,
   }
   // As the parser couldn't differentiate an FPR16 from an FPR64, coerce the
   // register from FPR64 to FPR16 if necessary.
   if (IsRegFPR64 && Kind == MCK_FPR16) {
     Op.Reg.RegNum = convertFPR64ToFPR16(Reg);
     return Match_Success;
   }
   // As the parser couldn't differentiate an VRM2/VRM4/VRM8 from an VR, coerce
   // the register from VR to VRM2/VRM4/VRM8 if necessary.
   if (IsRegVR && (Kind == MCK_VRM2 || Kind == MCK_VRM4 || Kind == MCK_VRM8)) {
     Op.Reg.RegNum = convertVRToVRMx(*getContext().getRegisterInfo(), Reg, Kind);
     if (Op.Reg.RegNum == 0)
       return Match_InvalidOperand;
     return Match_Success;
   }
   return Match_InvalidOperand;
 }
 
 bool RISCVAsmParser::generateImmOutOfRangeError(
     OperandVector &Operands, uint64_t ErrorInfo, int64_t Lower, int64_t Upper,
     Twine Msg = "immediate must be an integer in the range") {
   SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
   return Error(ErrorLoc, Msg + " [" + Twine(Lower) + ", " + Twine(Upper) + "]");
 }
 
case AsmToken::Identifier:
     SMLoc E = SMLoc::getFromPointer(S.getPointer() + Name.size());
     getLexer().Lex();
     Operands.push_back(RISCVOperand::createReg(RegNo, S, E, isRV64()));
   }
 
   return MatchOperand_Success;
 }
 
+OperandMatchResultTy RISCVAsmParser::parseGPRAsFPR(OperandVector &Operands) {
+  switch (getLexer().getKind()) {
+  default:
+    return MatchOperand_NoMatch;
+  case AsmToken::Identifier:
+    StringRef Name = getLexer().getTok().getIdentifier();
+    MCRegister RegNo;
+    matchRegisterNameHelper(isRV32E(), RegNo, Name);
+
+    if (RegNo == RISCV::NoRegister)
+      return MatchOperand_NoMatch;
+    SMLoc S = getLoc();
+    SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
+    getLexer().Lex();
+    Operands.push_back(RISCVOperand::createReg(
+        RegNo, S, E, isRV64(), !getSTI().hasFeature(RISCV::FeatureStdExtF)));
+  }
+  return MatchOperand_Success;
+}
+
 OperandMatchResultTy
 RISCVAsmParser::parseMemOpBaseReg(OperandVector &Operands) {
   if (getLexer().isNot(AsmToken::LParen)) {
     Error(getLoc(), "expected '('");
     return MatchOperand_ParseFail;
   }
 
   getParser().Lex(); // Eat '('

llvm/lib/Target/RISCV/Disassembler/RISCVDisassembler.cpp

static DecodeStatus DecodeGPRCRegisterClass(MCInst &Inst, uint64_t RegNo,
   if (RegNo >= 8)
     return MCDisassembler::Fail;
 
   MCRegister Reg = RISCV::X8 + RegNo;
   Inst.addOperand(MCOperand::createReg(Reg));
   return MCDisassembler::Success;
 }
 
+static DecodeStatus DecodeGPRPF64RegisterClass(MCInst &Inst, uint64_t RegNo,
+                                               uint64_t Address,
+                                               const void *Decoder) {
+  if (RegNo >= 32 || RegNo & 1)
+    return MCDisassembler::Fail;
+
+  MCRegister Reg = RISCV::X0 + RegNo;
+  Inst.addOperand(MCOperand::createReg(Reg));
+  return MCDisassembler::Success;
+}
+
 static DecodeStatus DecodeVRRegisterClass(MCInst &Inst, uint64_t RegNo,
                                           uint64_t Address,
                                           const void *Decoder) {
   if (RegNo >= 32)
     return MCDisassembler::Fail;
 
   MCRegister Reg = RISCV::V0 + RegNo;
   Inst.addOperand(MCOperand::createReg(Reg));
   return MCDisassembler::Success;
 }
 
 static DecodeStatus DecodeVRM2RegisterClass(MCInst &Inst, uint64_t RegNo,
                                             uint64_t Address,
                                             const void *Decoder) {
   if (RegNo >= 32)
     return MCDisassembler::Fail;
 
   if (RegNo % 2)
     return MCDisassembler::Fail;
 
DecodeStatus RISCVDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
 
   // It's a 32 bit instruction if bit 0 and 1 are 1.
   if ((Bytes[0] & 0x3) == 0x3) {
     if (Bytes.size() < 4) {
       Size = 0;
       return MCDisassembler::Fail;
     }
     Insn = support::endian::read32le(Bytes.data());
+    if (STI.getFeatureBits()[RISCV::FeatureStdExtZdinx] &&
+        !STI.getFeatureBits()[RISCV::Feature64Bit]) {
+      LLVM_DEBUG(dbgs() << "Trying RV32Zdinx table (Double in Integer and"
+                           "rv32)\n");
+      Result = decodeInstruction(DecoderTableRV32Zdinx32, MI, Insn, Address,
+                                 this, STI);
+      if (Result != MCDisassembler::Fail) {
+        Size = 4;
+        return Result;
+      }
+    }
+
+    if (STI.getFeatureBits()[RISCV::FeatureStdExtZfinx]) {
+      LLVM_DEBUG(dbgs() << "Trying RVZfinx table (Float in Integer):\n");
+      Result = decodeInstruction(DecoderTableRVZfinx32, MI, Insn, Address, this,
+                                 STI);
+      if (Result != MCDisassembler::Fail) {
+        Size = 4;
+        return Result;
+      }
+    }
     LLVM_DEBUG(dbgs() << "Trying RISCV32 table :\n");
     Result = decodeInstruction(DecoderTable32, MI, Insn, Address, this, STI);
     Size = 4;
   } else {
     if (Bytes.size() < 2) {
       Size = 0;
       return MCDisassembler::Fail;
     }

llvm/lib/Target/RISCV/RISCV.td

 def FeatureStdExtZfh
     : SubtargetFeature<"experimental-zfh", "HasStdExtZfh", "true",
                        "'Zfh' (Half-Precision Floating-Point)",
                        [FeatureStdExtZfhmin, FeatureStdExtF]>;
 def HasStdExtZfh : Predicate<"Subtarget->hasStdExtZfh()">,
                              AssemblerPredicate<(all_of FeatureStdExtZfh),
                              "'Zfh' (Half-Precision Floating-Point)">;
 
+def FeatureStdExtZfinx
+    : SubtargetFeature<"experimental-zfinx", "HasStdExtZfinx", "true",
+                       "'Zfinx' (Float in Integer)">;
+def HasStdExtZfinx : Predicate<"Subtarget->hasStdExtZfinx()">,
+                               AssemblerPredicate<(all_of FeatureStdExtZfinx),
+                               "'Zfinx' (Float in Integer)">;
+
+def FeatureStdExtZdinx
+    : SubtargetFeature<"experimental-zdinx", "HasStdExtZdinx", "true",
+                       "'Zdinx' (Double in Integer)",
+                       [FeatureStdExtZfinx]>;
+def HasStdExtZdinx : Predicate<"Subtarget->hasStdExtZdinx()">,
+                               AssemblerPredicate<(all_of FeatureStdExtZdinx),
+                               "'Zdinx' (Double in Integer)">;
+
+def FeatureStdExtZhinxmin
+    : SubtargetFeature<"experimental-zhinxmin", "HasStdExtZhinxmin", "true",
+                       "'Zhinxmin' (Half Float in Integer Minimal)",
+                       [FeatureStdExtZfinx]>;
+def HasStdExtZhinxmin : Predicate<"Subtarget->hasStdExtZhinxmin()">,
+                                  AssemblerPredicate<(all_of FeatureStdExtZhinxmin),
+                                  "'Zhinxmin' (Half Float in Integer Minimal)">;
+
+def FeatureStdExtZhinx
+    : SubtargetFeature<"experimental-zhinx", "HasStdExtZhinx", "true",
+                       "'Zhinx' (Half Float in Integer)",
+                       [FeatureStdExtZhinxmin]>;
+def HasStdExtZhinx : Predicate<"Subtarget->hasStdExtZhinx()">,
+                               AssemblerPredicate<(all_of FeatureStdExtZhinx),
+                               "'Zhinx' (Half Float in Integer)">;
+
 def FeatureStdExtC
     : SubtargetFeature<"c", "HasStdExtC", "true",
                        "'C' (Compressed Instructions)">;
 def HasStdExtC : Predicate<"Subtarget->hasStdExtC()">,
                            AssemblerPredicate<(all_of FeatureStdExtC),
                            "'C' (Compressed Instructions)">;
 
 def FeatureStdExtZba

llvm/lib/Target/RISCV/RISCVInstrInfoD.td

 def SDT_RISCVSplitF64     : SDTypeProfile<2, 1, [SDTCisVT<0, i32>,
                                                  SDTCisVT<1, i32>,
                                                  SDTCisVT<2, f64>]>;
 
 def RISCVBuildPairF64 : SDNode<"RISCVISD::BuildPairF64", SDT_RISCVBuildPairF64>;
 def RISCVSplitF64     : SDNode<"RISCVISD::SplitF64", SDT_RISCVSplitF64>;
 
 //===----------------------------------------------------------------------===//
+// Zdinx
+//===----------------------------------------------------------------------===//
+
+def GPRPF64AsFPR : AsmOperandClass {
+  let Name = "GPRPF64AsFPR";
+  let ParserMethod = "parseGPRAsFPR";
+  let RenderMethod = "addRegOperands";
+}
+
+def GPRF64AsFPR : AsmOperandClass {
+  let Name = "GPRF64AsFPR";
+  let ParserMethod = "parseGPRAsFPR";
+  let RenderMethod = "addRegOperands";
+}
+
+def FPR64INX : RegisterOperand<GPRF64> {
+  let ParserMatchClass = GPRF64AsFPR;
+  let DecoderMethod = "DecodeGPRRegisterClass";
+}
+
+def FPR64IN32X : RegisterOperand<GPRPF64> {
+  let ParserMatchClass = GPRPF64AsFPR;
+}
+
+def D       : ExtInfo<"",       [HasStdExtD],             FPR64>;
+def ZDINX   : ExtInfo<"_INX",   [HasStdExtZdinx, IsRV64], FPR64INX,
+                      "RVZfinx">;
+def ZDIN32X : ExtInfo<"_IN32X", [HasStdExtZdinx, IsRV32], FPR64IN32X,
+                      "RV32Zdinx">;
+
+defvar DINX = [D, ZDINX, ZDIN32X];
+
+//===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtD] in {
 def FLD : FPLoad_r<0b011, "fld", FPR64, WriteFLD64>;
 
 // Operands for stores are in the order srcreg, base, offset rather than
 // reflecting the order these fields are specified in the instruction
 // encoding.
 def FSD : FPStore_r<0b011, "fsd", FPR64, WriteFST64>;
+} // Predicates = [HasStdExtD]
 
 let SchedRW = [WriteFMA64, ReadFMA64, ReadFMA64, ReadFMA64] in {
-def FMADD_D  : FPFMA_rrr_frm<OPC_MADD,  0b01, "fmadd.d",  FPR64>;
-def FMSUB_D  : FPFMA_rrr_frm<OPC_MSUB,  0b01, "fmsub.d",  FPR64>;
-def FNMSUB_D : FPFMA_rrr_frm<OPC_NMSUB, 0b01, "fnmsub.d", FPR64>;
-def FNMADD_D : FPFMA_rrr_frm<OPC_NMADD, 0b01, "fnmadd.d", FPR64>;
+defm FMADD_D  : FPFMA_rrr_frm_m<OPC_MADD,  0b01, "fmadd.d",  DINX>;
+defm FMSUB_D  : FPFMA_rrr_frm_m<OPC_MSUB,  0b01, "fmsub.d",  DINX>;
+defm FNMSUB_D : FPFMA_rrr_frm_m<OPC_NMSUB, 0b01, "fnmsub.d", DINX>;
+defm FNMADD_D : FPFMA_rrr_frm_m<OPC_NMADD, 0b01, "fnmadd.d", DINX>;
 }
 
-def : FPFMADynFrmAlias<FMADD_D,  "fmadd.d",  FPR64>;
-def : FPFMADynFrmAlias<FMSUB_D,  "fmsub.d",  FPR64>;
-def : FPFMADynFrmAlias<FNMSUB_D, "fnmsub.d", FPR64>;
-def : FPFMADynFrmAlias<FNMADD_D, "fnmadd.d", FPR64>;
-
-def FADD_D : FPALU_rr_frm<0b0000001, "fadd.d", FPR64>,
-             Sched<[WriteFALU64, ReadFALU64, ReadFALU64]>;
-def FSUB_D : FPALU_rr_frm<0b0000101, "fsub.d", FPR64>,
-             Sched<[WriteFALU64, ReadFALU64, ReadFALU64]>;
-def FMUL_D : FPALU_rr_frm<0b0001001, "fmul.d", FPR64>,
-             Sched<[WriteFMul64, ReadFMul64, ReadFMul64]>;
-def FDIV_D : FPALU_rr_frm<0b0001101, "fdiv.d", FPR64>,
-             Sched<[WriteFDiv64, ReadFDiv64, ReadFDiv64]>;
-
-def        : FPALUDynFrmAlias<FADD_D, "fadd.d", FPR64>;
-def        : FPALUDynFrmAlias<FSUB_D, "fsub.d", FPR64>;
-def        : FPALUDynFrmAlias<FMUL_D, "fmul.d", FPR64>;
-def        : FPALUDynFrmAlias<FDIV_D, "fdiv.d", FPR64>;
+defm : FPFMADynFrmAlias_m<FMADD_D,  "fmadd.d",  DINX>;
+defm : FPFMADynFrmAlias_m<FMSUB_D,  "fmsub.d",  DINX>;
+defm : FPFMADynFrmAlias_m<FNMSUB_D, "fnmsub.d", DINX>;
+defm : FPFMADynFrmAlias_m<FNMADD_D, "fnmadd.d", DINX>;
+
+let SchedRW = [WriteFALU64, ReadFALU64, ReadFALU64] in {
+defm FADD_D : FPALU_rr_frm_m<0b0000001, "fadd.d", DINX>;
+defm FSUB_D : FPALU_rr_frm_m<0b0000101, "fsub.d", DINX>;
+}
+let SchedRW = [WriteFMul64, ReadFMul64, ReadFMul64] in
+defm FMUL_D : FPALU_rr_frm_m<0b0001001, "fmul.d", DINX>;
+
+let SchedRW = [WriteFDiv64, ReadFDiv64, ReadFDiv64] in
+defm FDIV_D : FPALU_rr_frm_m<0b0001101, "fdiv.d", DINX>;
 
+defm : FPALUDynFrmAlias_m<FADD_D, "fadd.d", DINX>;
+defm : FPALUDynFrmAlias_m<FSUB_D, "fsub.d", DINX>;
+defm : FPALUDynFrmAlias_m<FMUL_D, "fmul.d", DINX>;
+defm : FPALUDynFrmAlias_m<FDIV_D, "fdiv.d", DINX>;
+
+let Predicates = [HasStdExtD] in {
 def FSQRT_D : FPUnaryOp_r_frm<0b0101101, 0b00000, FPR64, FPR64, "fsqrt.d">,
               Sched<[WriteFSqrt64, ReadFSqrt64]>;
 def         : FPUnaryOpDynFrmAlias<FSQRT_D, "fsqrt.d", FPR64, FPR64>;
 
 let SchedRW = [WriteFSGNJ64, ReadFSGNJ64, ReadFSGNJ64],
     mayRaiseFPException = 0 in {
 def FSGNJ_D  : FPALU_rr<0b0010001, 0b000, "fsgnj.d", FPR64>;
 def FSGNJN_D : FPALU_rr<0b0010001, 0b001, "fsgnjn.d", FPR64>;
def FMV_D_X : FPUnaryOp_r<0b1111001, 0b00000, 0b000, FPR64, GPR, "fmv.d.x">,
               Sched<[WriteFMovI64ToF64, ReadFMovI64ToF64]>;
 } // Predicates = [HasStdExtD, IsRV64]
 
 //===----------------------------------------------------------------------===//
 // Assembler Pseudo Instructions (User-Level ISA, Version 2.2, Chapter 20)
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtD] in {
 def : InstAlias<"fld $rd, (${rs1})",  (FLD FPR64:$rd,  GPR:$rs1, 0), 0>;
 def : InstAlias<"fsd $rs2, (${rs1})", (FSD FPR64:$rs2, GPR:$rs1, 0), 0>;
 
 def : InstAlias<"fmv.d $rd, $rs",  (FSGNJ_D  FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 def : InstAlias<"fabs.d $rd, $rs", (FSGNJX_D FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 def : InstAlias<"fneg.d $rd, $rs", (FSGNJN_D FPR64:$rd, FPR64:$rs, FPR64:$rs)>;
 
 // fgt.d/fge.d are recognised by the GNU assembler but the canonical
 // flt.d/fle.d forms will always be printed. Therefore, set a zero weight.
 def : InstAlias<"fgt.d $rd, $rs, $rt",
                 (FLT_D GPR:$rd, FPR64:$rt, FPR64:$rs), 0>;
 def : InstAlias<"fge.d $rd, $rs, $rt",
                 (FLE_D GPR:$rd, FPR64:$rt, FPR64:$rs), 0>;
 
 def PseudoFLD  : PseudoFloatLoad<"fld", FPR64>;
 def PseudoFSD  : PseudoStore<"fsd", FPR64>;
 } // Predicates = [HasStdExtD]
 
 //===----------------------------------------------------------------------===//
 // Pseudo-instructions and codegen patterns
 //===----------------------------------------------------------------------===//
 
 class PatFpr64Fpr64<SDPatternOperator OpNode, RVInstR Inst>
     : Pat<(OpNode FPR64:$rs1, FPR64:$rs2), (Inst $rs1, $rs2)>;
 
 class PatFpr64Fpr64DynFrm<SDPatternOperator OpNode, RVInstRFrm Inst>
     : Pat<(OpNode FPR64:$rs1, FPR64:$rs2), (Inst $rs1, $rs2, 0b111)>;

llvm/lib/Target/RISCV/RISCVInstrInfoF.td

 def riscv_any_fcvt_wu_rv64 : PatFrags<(ops node:$src, node:$frm),
                                       [(riscv_strict_fcvt_wu_rv64 node:$src, node:$frm),
                                        (riscv_fcvt_wu_rv64 node:$src, node:$frm)]>;
 
 //===----------------------------------------------------------------------===//
 // Operand and SDNode transformation definitions.
 //===----------------------------------------------------------------------===//
 
+// Zfinx
+
+def GPRAsFPR : AsmOperandClass {
+  let Name = "GPRAsFPR";
+  let ParserMethod = "parseGPRAsFPR";
+  let RenderMethod = "addRegOperands";
+}
+
+def FPR32INX : RegisterOperand<GPRF32> {
+  let ParserMatchClass = GPRAsFPR;
+  let DecoderMethod = "DecodeGPRRegisterClass";
+}
+
+class ExtInfo<string suffix, list<Predicate> pres, DAGOperand reg,
+              string space = ""> {
+  string Suffix = suffix;
+  list<Predicate> Predicates = pres;
+  DAGOperand Reg = reg;
+  string Space = space;
+}
+
+def F     : ExtInfo<"",     [HasStdExtF],     FPR32>;
+def ZFINX : ExtInfo<"_INX", [HasStdExtZfinx], FPR32INX, "RVZfinx">;
+
+defvar FINX = [F, ZFINX];
+
 // Floating-point rounding mode
 
 def FRMArg : AsmOperandClass {
   let Name = "FRMArg";
   let RenderMethod = "addFRMArgOperands";
   let DiagnosticType = "InvalidFRMArg";
 }
 
 def frmarg : Operand<XLenVT> {
   let ParserMatchClass = FRMArg;
   let PrintMethod = "printFRMArg";
   let DecoderMethod = "decodeFRMArg";
 }
 
 //===----------------------------------------------------------------------===//
 // Instruction class templates
 //===----------------------------------------------------------------------===//
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in
 class FPLoad_r<bits<3> funct3, string opcodestr, RegisterClass rty,
                SchedWrite sw>
     : RVInstI<funct3, OPC_LOAD_FP, (outs rty:$rd),
               (ins GPR:$rs1, simm12:$imm12),
               opcodestr, "$rd, ${imm12}(${rs1})">,
       Sched<[sw, ReadFMemBase]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
 class FPStore_r<bits<3> funct3, string opcodestr, RegisterClass rty,
                 SchedWrite sw>
     : RVInstS<funct3, OPC_STORE_FP, (outs),
               (ins rty:$rs2, GPR:$rs1, simm12:$imm12),
               opcodestr, "$rs2, ${imm12}(${rs1})">,
       Sched<[sw, ReadStoreData, ReadFMemBase]>;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, mayRaiseFPException = 1,
     UseNamedOperandTable = 1, hasPostISelHook = 1 in
 class FPFMA_rrr_frm<RISCVOpcode opcode, bits<2> funct2, string opcodestr,
-                    RegisterClass rty>
+                    DAGOperand rty>
     : RVInstR4Frm<funct2, opcode, (outs rty:$rd),
                   (ins rty:$rs1, rty:$rs2, rty:$rs3, frmarg:$frm),
                   opcodestr, "$rd, $rs1, $rs2, $rs3, $frm">;
 
+multiclass FPFMA_rrr_frm_m<RISCVOpcode opcode, bits<2> funct2,
+                           string opcodestr, list<ExtInfo> Exts> {
+  foreach Ext = Exts in
+    let Predicates = Ext.Predicates, DecoderNamespace = Ext.Space in
+    def Ext.Suffix : FPFMA_rrr_frm<opcode, funct2, opcodestr, Ext.Reg>;
+}
+
 class FPFMADynFrmAlias<FPFMA_rrr_frm Inst, string OpcodeStr,
-                       RegisterClass rty>
+                       DAGOperand rty>
     : InstAlias<OpcodeStr#" $rd, $rs1, $rs2, $rs3",
                 (Inst rty:$rd, rty:$rs1, rty:$rs2, rty:$rs3, 0b111)>;
+multiclass FPFMADynFrmAlias_m<FPFMA_rrr_frm Inst, string OpcodeStr,
+                              list<ExtInfo> Exts> {
+  foreach Ext = Exts in
+    let Predicates = Ext.Predicates in
+    def : FPFMADynFrmAlias<!cast<FPFMA_rrr_frm>(Inst#Ext.Suffix), OpcodeStr,
+                           Ext.Reg>;
+}
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, mayRaiseFPException = 1 in
 class FPALU_rr<bits<7> funct7, bits<3> funct3, string opcodestr,
                RegisterClass rty>
     : RVInstR<funct7, funct3, OPC_OP_FP, (outs rty:$rd),
               (ins rty:$rs1, rty:$rs2), opcodestr, "$rd, $rs1, $rs2">;
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, mayRaiseFPException = 1,
     UseNamedOperandTable = 1, hasPostISelHook = 1 in
-class FPALU_rr_frm<bits<7> funct7, string opcodestr, RegisterClass rty>
+class FPALU_rr_frm<bits<7> funct7, string opcodestr, DAGOperand rty>
     : RVInstRFrm<funct7, OPC_OP_FP, (outs rty:$rd),
                  (ins rty:$rs1, rty:$rs2, frmarg:$frm), opcodestr,
                   "$rd, $rs1, $rs2, $frm">;
 
+multiclass FPALU_rr_frm_m<bits<7> funct7, string opcodestr, list<ExtInfo> Exts> {
+  foreach Ext = Exts in
+    let Predicates = Ext.Predicates, DecoderNamespace = Ext.Space in
+    def Ext.Suffix : FPALU_rr_frm<funct7, opcodestr, Ext.Reg>;
+}
+
 class FPALUDynFrmAlias<FPALU_rr_frm Inst, string OpcodeStr,
-                       RegisterClass rty>
+                       DAGOperand rty>
     : InstAlias<OpcodeStr#" $rd, $rs1, $rs2",
                 (Inst rty:$rd, rty:$rs1, rty:$rs2, 0b111)>;
+multiclass FPALUDynFrmAlias_m<FPALU_rr_frm Inst, string OpcodeStr,
+                              list<ExtInfo> Exts> {
+  foreach Ext = Exts in
+    let Predicates = Ext.Predicates in
+    def : FPALUDynFrmAlias<!cast<FPALU_rr_frm>(Inst#Ext.Suffix), OpcodeStr,
+                           Ext.Reg>;
+}
 
 let hasSideEffects = 0, mayLoad = 0, mayStore = 0, mayRaiseFPException = 1 in
 class FPUnaryOp_r<bits<7> funct7, bits<5> rs2val, bits<3> funct3,
                   RegisterClass rdty, RegisterClass rs1ty, string opcodestr>
     : RVInstR<funct7, funct3, OPC_OP_FP, (outs rdty:$rd), (ins rs1ty:$rs1),
               opcodestr, "$rd, $rs1"> {
   let rs2 = rs2val;
 }

 
 let Predicates = [HasStdExtF] in {
 def FLW : FPLoad_r<0b010, "flw", FPR32, WriteFLD32>;
 
 // Operands for stores are in the order srcreg, base, offset rather than
 // reflecting the order these fields are specified in the instruction
 // encoding.
 def FSW : FPStore_r<0b010, "fsw", FPR32, WriteFST32>;
+} // Predicates = [HasStdExtF]
 
 let SchedRW = [WriteFMA32, ReadFMA32, ReadFMA32, ReadFMA32] in {
-def FMADD_S  : FPFMA_rrr_frm<OPC_MADD,  0b00, "fmadd.s",  FPR32>;
-def FMSUB_S  : FPFMA_rrr_frm<OPC_MSUB,  0b00, "fmsub.s",  FPR32>;
-def FNMSUB_S : FPFMA_rrr_frm<OPC_NMSUB, 0b00, "fnmsub.s", FPR32>;
-def FNMADD_S : FPFMA_rrr_frm<OPC_NMADD, 0b00, "fnmadd.s", FPR32>;
-}
-
-def : FPFMADynFrmAlias<FMADD_S,  "fmadd.s",  FPR32>;
-def : FPFMADynFrmAlias<FMSUB_S,  "fmsub.s",  FPR32>;
-def : FPFMADynFrmAlias<FNMSUB_S, "fnmsub.s", FPR32>;
-def : FPFMADynFrmAlias<FNMADD_S, "fnmadd.s", FPR32>;
-
-def FADD_S : FPALU_rr_frm<0b0000000, "fadd.s", FPR32>,
-             Sched<[WriteFALU32, ReadFALU32, ReadFALU32]>;
-def FSUB_S : FPALU_rr_frm<0b0000100, "fsub.s", FPR32>,
-             Sched<[WriteFALU32, ReadFALU32, ReadFALU32]>;
-def FMUL_S : FPALU_rr_frm<0b0001000, "fmul.s", FPR32>,
-             Sched<[WriteFMul32, ReadFMul32, ReadFMul32]>;
-def FDIV_S : FPALU_rr_frm<0b0001100, "fdiv.s", FPR32>,
-             Sched<[WriteFDiv32, ReadFDiv32, ReadFDiv32]>;
-
-def        : FPALUDynFrmAlias<FADD_S, "fadd.s", FPR32>;
-def        : FPALUDynFrmAlias<FSUB_S, "fsub.s", FPR32>;
-def        : FPALUDynFrmAlias<FMUL_S, "fmul.s", FPR32>;
-def        : FPALUDynFrmAlias<FDIV_S, "fdiv.s", FPR32>;
+defm FMADD_S  : FPFMA_rrr_frm_m<OPC_MADD,  0b00, "fmadd.s",  FINX>;
+defm FMSUB_S  : FPFMA_rrr_frm_m<OPC_MSUB,  0b00, "fmsub.s",  FINX>;
+defm FNMSUB_S : FPFMA_rrr_frm_m<OPC_NMSUB, 0b00, "fnmsub.s", FINX>;
+defm FNMADD_S : FPFMA_rrr_frm_m<OPC_NMADD, 0b00, "fnmadd.s", FINX>;
+}
+
+defm : FPFMADynFrmAlias_m<FMADD_S,  "fmadd.s",  FINX>;
+defm : FPFMADynFrmAlias_m<FMSUB_S,  "fmsub.s",  FINX>;
+defm : FPFMADynFrmAlias_m<FNMSUB_S, "fnmsub.s", FINX>;
+defm : FPFMADynFrmAlias_m<FNMADD_S, "fnmadd.s", FINX>;
+
+let SchedRW = [WriteFALU32, ReadFALU32, ReadFALU32] in {
+defm FADD_S : FPALU_rr_frm_m<0b0000000, "fadd.s", FINX>;
+defm FSUB_S : FPALU_rr_frm_m<0b0000100, "fsub.s", FINX>;
+}
+let SchedRW = [WriteFMul32, ReadFMul32, ReadFMul32] in
+defm FMUL_S : FPALU_rr_frm_m<0b0001000, "fmul.s", FINX>;
 
+let SchedRW = [WriteFDiv32, ReadFDiv32, ReadFDiv32] in
+defm FDIV_S : FPALU_rr_frm_m<0b0001100, "fdiv.s", FINX>;
+
+defm : FPALUDynFrmAlias_m<FADD_S, "fadd.s", FINX>;
+defm : FPALUDynFrmAlias_m<FSUB_S, "fsub.s", FINX>;
+defm : FPALUDynFrmAlias_m<FMUL_S, "fmul.s", FINX>;
+defm : FPALUDynFrmAlias_m<FDIV_S, "fdiv.s", FINX>;
+
+let Predicates = [HasStdExtF] in {
 def FSQRT_S : FPUnaryOp_r_frm<0b0101100, 0b00000, FPR32, FPR32, "fsqrt.s">,
               Sched<[WriteFSqrt32, ReadFSqrt32]>;
 def         : FPUnaryOpDynFrmAlias<FSQRT_S, "fsqrt.s", FPR32, FPR32>;
 
 let SchedRW = [WriteFSGNJ32, ReadFSGNJ32, ReadFSGNJ32],
     mayRaiseFPException = 0 in {
 def FSGNJ_S  : FPALU_rr<0b0010000, 0b000, "fsgnj.s", FPR32>;
 def FSGNJN_S : FPALU_rr<0b0010000, 0b001, "fsgnjn.s", FPR32>;

llvm/lib/Target/RISCV/RISCVInstrInfoZfh.td

def SDT_RISCVFMV_X_ANYEXTH
     : SDTypeProfile<1, 1, [SDTCisVT<0, XLenVT>, SDTCisVT<1, f16>]>;
 
 def riscv_fmv_h_x
     : SDNode<"RISCVISD::FMV_H_X", SDT_RISCVFMV_H_X>;
 def riscv_fmv_x_anyexth
     : SDNode<"RISCVISD::FMV_X_ANYEXTH", SDT_RISCVFMV_X_ANYEXTH>;
 
 //===----------------------------------------------------------------------===//
+// Zhinx and Zhinxmin
+//===----------------------------------------------------------------------===//
+
+def FPR16INX : RegisterOperand<GPRF16> {
+  let ParserMatchClass = GPRAsFPR;
+  let DecoderMethod = "DecodeGPRRegisterClass";
+}
+
+def ZFH   : ExtInfo<"",     [HasStdExtZfh],   FPR16>;
+def ZHINX : ExtInfo<"_INX", [HasStdExtZhinx], FPR16INX, "RVZfinx">;
+
+defvar HINX = [ZFH, ZHINX];
+
+//===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtZfhmin] in {
 def FLH : FPLoad_r<0b001, "flh", FPR16, WriteFLD16>;
 
 // Operands for stores are in the order srcreg, base, offset rather than
 // reflecting the order these fields are specified in the instruction
 // encoding.
 def FSH : FPStore_r<0b001, "fsh", FPR16, WriteFST16>;
 } // Predicates = [HasStdExtZfhmin]
 
-let Predicates = [HasStdExtZfh] in {
 let SchedRW = [WriteFMA16, ReadFMA16, ReadFMA16, ReadFMA16] in {
-def FMADD_H  : FPFMA_rrr_frm<OPC_MADD,  0b10, "fmadd.h",  FPR16>;
-def FMSUB_H  : FPFMA_rrr_frm<OPC_MSUB,  0b10, "fmsub.h",  FPR16>;
-def FNMSUB_H : FPFMA_rrr_frm<OPC_NMSUB, 0b10, "fnmsub.h", FPR16>;
-def FNMADD_H : FPFMA_rrr_frm<OPC_NMADD, 0b10, "fnmadd.h", FPR16>;
+defm FMADD_H  : FPFMA_rrr_frm_m<OPC_MADD,  0b10, "fmadd.h",  HINX>;
+defm FMSUB_H  : FPFMA_rrr_frm_m<OPC_MSUB,  0b10, "fmsub.h",  HINX>;
+defm FNMSUB_H : FPFMA_rrr_frm_m<OPC_NMSUB, 0b10, "fnmsub.h", HINX>;
+defm FNMADD_H : FPFMA_rrr_frm_m<OPC_NMADD, 0b10, "fnmadd.h", HINX>;
+}
+
+defm : FPFMADynFrmAlias_m<FMADD_H,  "fmadd.h",  HINX>;
+defm : FPFMADynFrmAlias_m<FMSUB_H,  "fmsub.h",  HINX>;
+defm : FPFMADynFrmAlias_m<FNMSUB_H, "fnmsub.h", HINX>;
+defm : FPFMADynFrmAlias_m<FNMADD_H, "fnmadd.h", HINX>;
+
+let SchedRW = [WriteFALU16, ReadFALU16, ReadFALU16] in {
+defm FADD_H : FPALU_rr_frm_m<0b0000010, "fadd.h", HINX>;
+defm FSUB_H : FPALU_rr_frm_m<0b0000110, "fsub.h", HINX>;
 }
+let SchedRW = [WriteFMul16, ReadFMul16, ReadFMul16] in
+defm FMUL_H : FPALU_rr_frm_m<0b0001010, "fmul.h", HINX>;
+
+let SchedRW = [WriteFDiv16, ReadFDiv16, ReadFDiv16] in
+defm FDIV_H : FPALU_rr_frm_m<0b0001110, "fdiv.h", HINX>;
 
-def : FPFMADynFrmAlias<FMADD_H,  "fmadd.h",  FPR16>;
-def : FPFMADynFrmAlias<FMSUB_H,  "fmsub.h",  FPR16>;
-def : FPFMADynFrmAlias<FNMSUB_H, "fnmsub.h", FPR16>;
-def : FPFMADynFrmAlias<FNMADD_H, "fnmadd.h", FPR16>;
-
-def FADD_H : FPALU_rr_frm<0b0000010, "fadd.h", FPR16>,
-             Sched<[WriteFALU16, ReadFALU16, ReadFALU16]>;
-def FSUB_H : FPALU_rr_frm<0b0000110, "fsub.h", FPR16>,
-             Sched<[WriteFALU16, ReadFALU16, ReadFALU16]>;
-def FMUL_H : FPALU_rr_frm<0b0001010, "fmul.h", FPR16>,
-             Sched<[WriteFMul16, ReadFMul16, ReadFMul16]>;
-def FDIV_H : FPALU_rr_frm<0b0001110, "fdiv.h", FPR16>,
-             Sched<[WriteFDiv16, ReadFDiv16, ReadFDiv16]>;
-
-def        : FPALUDynFrmAlias<FADD_H, "fadd.h", FPR16>;
-def        : FPALUDynFrmAlias<FSUB_H, "fsub.h", FPR16>;
-def        : FPALUDynFrmAlias<FMUL_H, "fmul.h", FPR16>;
-def        : FPALUDynFrmAlias<FDIV_H, "fdiv.h", FPR16>;
+defm : FPALUDynFrmAlias_m<FADD_H, "fadd.h", HINX>;
+defm : FPALUDynFrmAlias_m<FSUB_H, "fsub.h", HINX>;
+defm : FPALUDynFrmAlias_m<FMUL_H, "fmul.h", HINX>;
+defm : FPALUDynFrmAlias_m<FDIV_H, "fdiv.h", HINX>;
 
+let Predicates = [HasStdExtZfh] in {
 def FSQRT_H : FPUnaryOp_r_frm<0b0101110, 0b00000, FPR16, FPR16, "fsqrt.h">,
               Sched<[WriteFSqrt16, ReadFSqrt16]>;
 def         : FPUnaryOpDynFrmAlias<FSQRT_H, "fsqrt.h", FPR16, FPR16>;
 
 let SchedRW = [WriteFSGNJ16, ReadFSGNJ16, ReadFSGNJ16],
     mayRaiseFPException = 0 in {
 def FSGNJ_H  : FPALU_rr<0b0010010, 0b000, "fsgnj.h", FPR16>;
 def FSGNJN_H : FPALU_rr<0b0010010, 0b001, "fsgnjn.h", FPR16>;

llvm/lib/Target/RISCV/RISCVRegisterInfo.td

 def sub_vrm1_1 : SubRegIndex<64, 64>;
 def sub_vrm1_2 : ComposedSubRegIndex<sub_vrm2_1, sub_vrm1_0>;
 def sub_vrm1_3 : ComposedSubRegIndex<sub_vrm2_1, sub_vrm1_1>;
 def sub_vrm1_4 : ComposedSubRegIndex<sub_vrm2_2, sub_vrm1_0>;
 def sub_vrm1_5 : ComposedSubRegIndex<sub_vrm2_2, sub_vrm1_1>;
 def sub_vrm1_6 : ComposedSubRegIndex<sub_vrm2_3, sub_vrm1_0>;
 def sub_vrm1_7 : ComposedSubRegIndex<sub_vrm2_3, sub_vrm1_1>;
 
+def sub_32_hi  : SubRegIndex<32, 32>;
 } // Namespace = "RISCV"
 
 // Integer registers
 // CostPerUse is set higher for registers that may not be compressible as they
 // are not part of GPRC, the most restrictive register class used by the
 // compressed instruction set. This will influence the greedy register
 // allocator to reduce the use of registers that can't be encoded in 16 bit
 // instructions. This affects register allocation even when compressed
let RegAltNameIndices = [ABIRegAltName] in {
   def X29 : RISCVReg<29,"x29", ["t4"]>, DwarfRegNum<[29]>;
   def X30 : RISCVReg<30,"x30", ["t5"]>, DwarfRegNum<[30]>;
   def X31 : RISCVReg<31,"x31", ["t6"]>, DwarfRegNum<[31]>;
   }
 }
 
 def XLenVT : ValueTypeByHwMode<[RV32, RV64],
                                [i32,  i64]>;
 def XLenRI : RegInfoByHwMode<
       [RV32,              RV64],
       [RegInfo<32,32,32>, RegInfo<64,64,64>]>;
 
 // The order of registers represents the preferred allocation sequence.
 // Registers are listed in the order caller-save, callee-save, specials.
 def GPR : RegisterClass<"RISCV", [XLenVT], 32, (add
     (sequence "X%u", 10, 17),
     (sequence "X%u", 5, 7),
     (sequence "X%u", 28, 31),
     (sequence "X%u", 8, 9),
     (sequence "X%u", 18, 27),
     (sequence "X%u", 0, 4)
   )> {
   let RegInfos = XLenRI;
 }
 
 def GPRX0 : RegisterClass<"RISCV", [XLenVT], 32, (add X0)> {
   let RegInfos = XLenRI;
 }
 
 // The order of registers represents the preferred allocation sequence.
 // Registers are listed in the order caller-save, callee-save, specials.
 def GPRNoX0 : RegisterClass<"RISCV", [XLenVT], 32, (add

 
 defvar VMaskVTs = [vbool1_t, vbool2_t, vbool4_t, vbool8_t, vbool16_t,
                    vbool32_t, vbool64_t];
 
 def VMV0 : RegisterClass<"RISCV", VMaskVTs, 64, (add V0)> {
   let Size = 64;
 }
 
+let RegAltNameIndices = [ABIRegAltName] in {
+  foreach Index = [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22,
+                   24, 26, 28, 30] in {
+    defvar Reg = !cast<Register>("X"#Index);
+    def X#Index#_PD : RISCVRegWithSubRegs<Index, Reg.AsmName,
+                       [!cast<Register>("X"#Index),
+                        !cast<Register>("X"#!add(Index, 1))],
+                       Reg.AltNames> {
+      let SubRegIndices = [sub_32, sub_32_hi];
+    }
+  }
+}
+
+def GPRF16 : RegisterClass<"RISCV", [f16], 16, (add
+    (sequence "X%u", 10, 17),
+    (sequence "X%u", 5, 7),
+    (sequence "X%u", 28, 31),
+    (sequence "X%u", 8, 9),
+    (sequence "X%u", 18, 27),
+    (sequence "X%u", 0, 4)
+)> {
+  let RegInfos = XLenRI;
+}
+
+def GPRF32 : RegisterClass<"RISCV", [f32], 32, (add
+    (sequence "X%u", 10, 17),
+    (sequence "X%u", 5, 7),
+    (sequence "X%u", 28, 31),
+    (sequence "X%u", 8, 9),
+    (sequence "X%u", 18, 27),
+    (sequence "X%u", 0, 4)
+)> {
+  let RegInfos = XLenRI;
+}
+
+def GPRF64 : RegisterClass<"RISCV", [f64], 64, (add
+    (sequence "X%u", 10, 17),
+    (sequence "X%u", 5, 7),
+    (sequence "X%u", 28, 31),
+    (sequence "X%u", 8, 9),
+    (sequence "X%u", 18, 27),
+    (sequence "X%u", 0, 4)
+)> {
+  let RegInfos = XLenRI;
+}
+
+def GPRPF64 : RegisterClass<"RISCV", [f64], 64, (add
+    X10_PD, X12_PD, X14_PD, X16_PD,
+    X6_PD,
+    X28_PD, X30_PD,
+    X8_PD,
+    X18_PD, X20_PD, X22_PD, X24_PD, X26_PD,
+    X0_PD, X2_PD, X4_PD
+)> {
+  let RegInfos = XLenRI;
+}
+
 // The register class is added for inline assembly for vector mask types.
 def VM : VReg<VMaskVTs,
            (add (sequence "V%u", 8, 31),
                 (sequence "V%u", 0, 7)), 1>;
 
 foreach m = LMULList.m in {
   foreach nf = NFList<m>.L in {
     def "VRN" # nf # "M" # m # "NoV0": VReg<[untyped],

 // NOTE: The alignment and size are bogus values. The Size needs to be non-zero
 // or tablegen will use "untyped" to determine the size which will assert.
 let isAllocatable = 0 in
 def AnyReg : RegisterClass<"RISCV", [untyped], 32,
                            (add (sequence "X%u", 0, 31),
                                 (sequence "F%u_D", 0, 31),
                                 (sequence "V%u", 0, 31))> {
   let Size = 32;
 }

llvm/lib/Target/RISCV/RISCVSubtarget.h

class RISCVSubtarget : public RISCVGenSubtargetInfo {
   bool HasStdExtZbp = false;
   bool HasStdExtZbr = false;
   bool HasStdExtZbs = false;
   bool HasStdExtZbt = false;
   bool HasStdExtV = false;
   bool HasStdExtZvlsseg = false;
   bool HasStdExtZfhmin = false;
   bool HasStdExtZfh = false;
+  bool HasStdExtZfinx = false;
+  bool HasStdExtZdinx = false;
+  bool HasStdExtZhinxmin = false;
+  bool HasStdExtZhinx = false;
   bool HasRV64 = false;
   bool IsRV32E = false;
   bool EnableLinkerRelax = false;
   bool EnableRVCHintInstrs = true;
   bool EnableSaveRestore = false;
   unsigned XLen = 32;
   MVT XLenVT = MVT::i32;
   uint8_t MaxInterleaveFactor = 2;
   RISCVABI::ABI TargetABI = RISCVABI::ABI_Unknown;
   BitVector UserReservedRegister;
   RISCVFrameLowering FrameLowering;
   RISCVInstrInfo InstrInfo;
   RISCVRegisterInfo RegInfo;
   RISCVTargetLowering TLInfo;
   SelectionDAGTargetInfo TSInfo;
 
public:
   void ParseSubtargetFeatures(StringRef CPU, StringRef TuneCPU, StringRef FS);
 
   const RISCVFrameLowering *getFrameLowering() const override {
     return &FrameLowering;
   }
   const RISCVInstrInfo *getInstrInfo() const override { return &InstrInfo; }
   const RISCVRegisterInfo *getRegisterInfo() const override {
     return &RegInfo;
   }
   const RISCVTargetLowering *getTargetLowering() const override {
     return &TLInfo;
   }
   const SelectionDAGTargetInfo *getSelectionDAGInfo() const override {
     return &TSInfo;
   }
   bool enableMachineScheduler() const override { return true; }
   bool hasStdExtM() const { return HasStdExtM; }
public:
   bool hasStdExtZbp() const { return HasStdExtZbp; }
   bool hasStdExtZbr() const { return HasStdExtZbr; }
   bool hasStdExtZbs() const { return HasStdExtZbs; }
   bool hasStdExtZbt() const { return HasStdExtZbt; }
   bool hasStdExtV() const { return HasStdExtV; }
   bool hasStdExtZvlsseg() const { return HasStdExtZvlsseg; }
   bool hasStdExtZfhmin() const { return HasStdExtZfhmin; }
   bool hasStdExtZfh() const { return HasStdExtZfh; }
+  bool hasStdExtZfinx() const { return HasStdExtZfinx; }
+  bool hasStdExtZdinx() const { return HasStdExtZdinx; }
+  bool hasStdExtZhinxmin() const { return HasStdExtZhinxmin; }
+  bool hasStdExtZhinx() const { return HasStdExtZhinx; }
   bool is64Bit() const { return HasRV64; }
   bool isRV32E() const { return IsRV32E; }
   bool enableLinkerRelax() const { return EnableLinkerRelax; }
   bool enableRVCHintInstrs() const { return EnableRVCHintInstrs; }
   bool enableSaveRestore() const { return EnableSaveRestore; }
   MVT getXLenVT() const { return XLenVT; }
   unsigned getXLen() const { return XLen; }
   RISCVABI::ABI getTargetABI() const { return TargetABI; }
public:
   // of 0 means nothing is known about that particular limit beyond what's
   // implied by the architecture.
   unsigned getMaxRVVVectorSizeInBits() const;
   unsigned getMinRVVVectorSizeInBits() const;
   unsigned getMaxLMULForFixedLengthVectors() const;
   unsigned getMaxELENForFixedLengthVectors() const;
   bool useRVVForFixedLengthVectors() const;
 };
 } // End llvm namespace
 
 #endif

llvm/test/MC/RISCV/attribute-arch.s

 # CHECK: attribute      5, "rv32i2p0_zbt0p93"
 
 .attribute arch, "rv32ifzfhmin0p1"
 # CHECK: attribute      5, "rv32i2p0_f2p0_zfhmin0p1"
 
 .attribute arch, "rv32ifzfh0p1"
 # CHECK: attribute      5, "rv32i2p0_f2p0_zfh0p1_zfhmin0p1"
 
+.attribute arch, "rv32izfinx1p0"
+# CHECK: attribute      5, "rv32i2p0_zfinx1p0"
+
+.attribute arch, "rv32izdinx1p0"
+# CHECK: attribute      5, "rv32i2p0_zdinx1p0"
+
+.attribute arch, "rv32izhinxmin1p0"
+# CHECK: attribute      5, "rv32i2p0_zhinxmin1p0"
+
+.attribute arch, "rv32izhinx1p0"
+# CHECK: attribute      5, "rv32i2p0_zhinx1p0_zhinxmin1p0"
+
 .attribute arch, "rv32iv0p10zvlsseg0p10"
 # CHECK: attribute      5, "rv32i2p0_v0p10_zvlsseg0p10"

llvm/test/MC/RISCV/rv32i-invalid.s

-# RUN: not llvm-mc -triple riscv32 < %s 2>&1 | FileCheck %s
+# RUN: not llvm-mc -triple riscv32 %s 2>&1 | FileCheck %s
 
 # Out of range immediates
 ## fencearg
 fence iorw, iore # CHECK: :[[@LINE]]:13: error: operand must be formed of letters selected in-order from 'iorw'
 fence wr, wr # CHECK: :[[@LINE]]:7: error: operand must be formed of letters selected in-order from 'iorw'
 fence rw, rr # CHECK: :[[@LINE]]:11: error: operand must be formed of letters selected in-order from 'iorw'
 fence 1, rw # CHECK: :[[@LINE]]:7: error: operand must be formed of letters selected in-order from 'iorw'
 fence unknown, unknown # CHECK: :[[@LINE]]:7: error: operand must be formed of letters selected in-order from 'iorw'

 jal gp, %lo(a) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %hi(2) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %hi(b) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %pcrel_hi(3) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %pcrel_hi(c) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %pcrel_lo(4) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 jal gp, %pcrel_lo(d) # CHECK: :[[@LINE]]:9: error: immediate must be a multiple of 2 bytes in the range [-1048576, 1048574]
 
 # Bare symbol names when an operand modifier is required and unsupported
 # operand modifiers.
 
 lui a0, foo # CHECK: :[[@LINE]]:9: error: operand must be a symbol with %hi/%tprel_hi modifier or an integer in the range [0, 1048575]
 lui a0, %lo(foo) # CHECK: :[[@LINE]]:9: error: operand must be a symbol with %hi/%tprel_hi modifier or an integer in the range [0, 1048575]
 lui a0, %pcrel_lo(foo) # CHECK: :[[@LINE]]:9: error: operand must be a symbol with %hi/%tprel_hi modifier or an integer in the range [0, 1048575]
 lui a0, %pcrel_hi(foo) # CHECK: :[[@LINE]]:9: error: operand must be a symbol with %hi/%tprel_hi modifier or an integer in the range [0, 1048575]
 
 auipc a0, foo # CHECK: :[[@LINE]]:11: error: operand must be a symbol with a %pcrel_hi/%got_pcrel_hi/%tls_ie_pcrel_hi/%tls_gd_pcrel_hi modifier or an integer in the range [0, 1048575]

 ori a0, a1 # CHECK: :[[@LINE]]:1: error: too few operands for instruction
 xor s2, s2 # CHECK: :[[@LINE]]:1: error: too few operands for instruction
 
 # Instruction not in the base ISA
 mul a4, ra, s0 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'M' (Integer Multiplication and Division)
 amomaxu.w s5, s4, (s3) # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'A' (Atomic Instructions)
 fadd.s ft0, ft1, ft2 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'F' (Single-Precision Floating-Point)
 fadd.h ft0, ft1, ft2 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zfh' (Half-Precision Floating-Point)
+fadd.s a0, a1, a2 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zfinx' (Float in Integer)
+fadd.d a0, a2, a4 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zdinx' (Double in Integer)
+fadd.h a0, a1, a2 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zhinx' (Half Float in Integer)
 
 # Using floating point registers when integer registers are expected
 addi a2, ft0, 24 # CHECK: :[[@LINE]]:10: error: invalid operand for instruction
 
 # fence.tso accepts no operands
 fence.tso rw, rw # CHECK: :[[@LINE]]:11: error: invalid operand for instruction

llvm/test/MC/RISCV/rv32zdinx-invalid.s

+# RUN: not llvm-mc -triple riscv32 -mattr=+experimental-zdinx %s 2>&1 | FileCheck %s
+
+# Invalid instructions
+flw fa4, 12(sp) # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'F' (Single-Precision Floating-Point)
+fsw a5, 12(sp) # CHECK: :[[@LINE]]:5: error: invalid operand for instruction
+fmv.x.w s0, s1 # CHECK: :[[@LINE]]:13: error: invalid operand for instruction
+
+# Invalid register names
+lw f15, 100(a0) # CHECK: :[[@LINE]]:4: error: invalid operand for instruction
+lw t1, 100(a10) # CHECK: :[[@LINE]]:12: error: expected register
+fadd.d a100, a2, a3 # CHECK: :[[@LINE]]:8: error: invalid operand for instruction

llvm/test/MC/RISCV/rv32zdinx-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zdinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv32 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zdinx -M no-aliases -d -r - \
+# RUN:     | FileCheck --check-prefix=CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zdinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv64 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zdinx -M no-aliases -d -r - \
+# RUN:     | FileCheck --check-prefix=CHECK-ASM-AND-OBJ %s
+
+# CHECK-ASM-AND-OBJ: lw s0, 12(a0)
+# CHECK-ASM: encoding: [0x03,0x24,0xc5,0x00]
+lw s0, 12(a0)
+# CHECK-ASM-AND-OBJ: lw s1, 4(ra)
+# CHECK-ASM: encoding: [0x83,0xa4,0x40,0x00]
+lw s1, +4(ra)
+# CHECK-ASM-AND-OBJ: lw s2, -2048(a3)
+# CHECK-ASM: encoding: [0x03,0xa9,0x06,0x80]
+lw s2, -2048(x13)
+# CHECK-ASM-AND-OBJ: lw s3, -2048(s1)
+# CHECK-ASM: encoding: [0x83,0xa9,0x04,0x80]
+lw s3, %lo(2048)(s1)
+# CHECK-ASM-AND-OBJ: lw s4, 2047(s2)
+# CHECK-ASM: encoding: [0x03,0x2a,0xf9,0x7f]
+lw s4, 2047(s2)
+# CHECK-ASM-AND-OBJ: lw s5, 0(s3)
+# CHECK-ASM: encoding: [0x83,0xaa,0x09,0x00]
+lw s5, 0(s3)
+
+# CHECK-ASM-AND-OBJ: sw s6, 2047(s4)
+# CHECK-ASM: encoding: [0xa3,0x2f,0x6a,0x7f]
+sw s6, 2047(s4)
+# CHECK-ASM-AND-OBJ: sw s7, -2048(s5)
+# CHECK-ASM: encoding: [0x23,0xa0,0x7a,0x81]
+sw s7, -2048(s5)
+# CHECK-ASM-AND-OBJ: sw s8, -2048(s6)
+# CHECK-ASM: encoding: [0x23,0x20,0x8b,0x81]
+sw s8, %lo(2048)(s6)
+# CHECK-ASM-AND-OBJ: sw s9, 999(s7)
+# CHECK-ASM: encoding: [0xa3,0xa3,0x9b,0x3f]
+sw s9, 999(s7)
+
+# CHECK-ASM-AND-OBJ: fmadd.d a0, a2, a4, a6, dyn
+# CHECK-ASM: encoding: [0x43,0x75,0xe6,0x82]
+fmadd.d x10, x12, x14, x16, dyn
+# CHECK-ASM-AND-OBJ: fmsub.d a0, a2, a4, a6, dyn
+# CHECK-ASM: encoding: [0x47,0x75,0xe6,0x82]
+fmsub.d x10, x12, x14, x16, dyn
+# CHECK-ASM-AND-OBJ: fnmsub.d a0, a2, a4, a6, dyn
+# CHECK-ASM: encoding: [0x4b,0x75,0xe6,0x82]
+fnmsub.d x10, x12, x14, x16, dyn
+# CHECK-ASM-AND-OBJ: fnmadd.d a0, a2, a4, a6, dyn
+# CHECK-ASM: encoding: [0x4f,0x75,0xe6,0x82]
+fnmadd.d x10, x12, x14, x16, dyn
+
+# CHECK-ASM-AND-OBJ: fadd.d s10, t3, t5, dyn
+# CHECK-ASM: encoding: [0x53,0x7d,0xee,0x03]
+fadd.d x26, x28, x30, dyn
+# CHECK-ASM-AND-OBJ: fsub.d s10, t3, t5, dyn
+# CHECK-ASM: encoding: [0x53,0x7d,0xee,0x0b]
+fsub.d x26, x28, x30, dyn
+# CHECK-ASM-AND-OBJ: fmul.d s10, t3, t5, dyn
+# CHECK-ASM: encoding: [0x53,0x7d,0xee,0x13]
+fmul.d x26, x28, x30, dyn
+# CHECK-ASM-AND-OBJ: fdiv.d s10, t3, t5, dyn
+# CHECK-ASM: encoding: [0x53,0x7d,0xee,0x1b]
+fdiv.d x26, x28, x30, dyn

llvm/test/MC/RISCV/rv32zfinx-invalid.s

+# RUN: not llvm-mc -triple riscv32 -mattr=+experimental-zfinx %s 2>&1 | FileCheck %s
+
+# Invalid instructions
+flw fa4, 12(sp) # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'F' (Single-Precision Floating-Point)
+fsw a5, 12(sp) # CHECK: :[[@LINE]]:5: error: invalid operand for instruction
+fmv.x.d s0, s1 # CHECK: :[[@LINE]]:13: error: invalid operand for instruction
+fadd.d t1, t3, t5 # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zdinx' (Double in Integer)
+
+# Invalid register names
+lw f15, 100(a0) # CHECK: :[[@LINE]]:4: error: invalid operand for instruction
+lw t1, 100(a10) # CHECK: :[[@LINE]]:12: error: expected register
+fadd.s a100, a2, a3 # CHECK: :[[@LINE]]:8: error: invalid operand for instruction

llvm/test/MC/RISCV/rv32zfinx-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zfinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zfinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv32 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zfinx -M no-aliases -d -r - \
+# RUN:     | FileCheck --check-prefix=CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv64 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zfinx -M no-aliases -d -r - \
+# RUN:     | FileCheck --check-prefix=CHECK-ASM-AND-OBJ %s
+
+# CHECK-ASM-AND-OBJ: lw s0, 12(a0)
+# CHECK-ASM: encoding: [0x03,0x24,0xc5,0x00]
+lw s0, 12(a0)
+# CHECK-ASM-AND-OBJ: lw s1, 4(ra)
+# CHECK-ASM: encoding: [0x83,0xa4,0x40,0x00]
+lw s1, +4(ra)
+# CHECK-ASM-AND-OBJ: lw s2, -2048(a3)
+# CHECK-ASM: encoding: [0x03,0xa9,0x06,0x80]
+lw s2, -2048(x13)
+# CHECK-ASM-AND-OBJ: lw s3, -2048(s1)
+# CHECK-ASM: encoding: [0x83,0xa9,0x04,0x80]
+lw s3, %lo(2048)(s1)
+# CHECK-ASM-AND-OBJ: lw s4, 2047(s2)
+# CHECK-ASM: encoding: [0x03,0x2a,0xf9,0x7f]
+lw s4, 2047(s2)
+# CHECK-ASM-AND-OBJ: lw s5, 0(s3)
+# CHECK-ASM: encoding: [0x83,0xaa,0x09,0x00]
+lw s5, 0(s3)
+
+# CHECK-ASM-AND-OBJ: sw s6, 2047(s4)
+# CHECK-ASM: encoding: [0xa3,0x2f,0x6a,0x7f]
+sw s6, 2047(s4)
+# CHECK-ASM-AND-OBJ: sw s7, -2048(s5)
+# CHECK-ASM: encoding: [0x23,0xa0,0x7a,0x81]
+sw s7, -2048(s5)
+# CHECK-ASM-AND-OBJ: sw s8, -2048(s6)
+# CHECK-ASM: encoding: [0x23,0x20,0x8b,0x81]
+sw s8, %lo(2048)(s6)
+# CHECK-ASM-AND-OBJ: sw s9, 999(s7)
+# CHECK-ASM: encoding: [0xa3,0xa3,0x9b,0x3f]
+sw s9, 999(s7)
+
+# CHECK-ASM-AND-OBJ: fmadd.s a0, a1, a2, a3, dyn
+# CHECK-ASM: encoding: [0x43,0xf5,0xc5,0x68]
+fmadd.s x10, x11, x12, x13, dyn
+# CHECK-ASM-AND-OBJ: fmsub.s a4, a5, a6, a7, dyn
+# CHECK-ASM: encoding: [0x47,0xf7,0x07,0x89]
+fmsub.s x14, x15, x16, x17, dyn
+# CHECK-ASM-AND-OBJ: fnmsub.s s2, s3, s4, s5, dyn
+# CHECK-ASM: encoding: [0x4b,0xf9,0x49,0xa9]
+fnmsub.s x18, x19, x20, x21, dyn
+# CHECK-ASM-AND-OBJ: fnmadd.s s6, s7, s8, s9, dyn
+# CHECK-ASM: encoding: [0x4f,0xfb,0x8b,0xc9]
+fnmadd.s x22, x23, x24, x25, dyn
+
+# CHECK-ASM-AND-OBJ: fadd.s s10, s11, t3, dyn
+# CHECK-ASM: encoding: [0x53,0xfd,0xcd,0x01]
+fadd.s x26, x27, x28, dyn
+# CHECK-ASM-AND-OBJ: fsub.s t4, t5, t6, dyn
+# CHECK-ASM: encoding: [0xd3,0x7e,0xff,0x09]
+fsub.s x29, x30, x31, dyn
+# CHECK-ASM-AND-OBJ: fmul.s s0, s1, s2, dyn
+# CHECK-ASM: encoding: [0x53,0xf4,0x24,0x11]
+fmul.s s0, s1, s2, dyn
+# CHECK-ASM-AND-OBJ: fdiv.s s3, s4, s5, dyn
+# CHECK-ASM: encoding: [0xd3,0x79,0x5a,0x19]
+fdiv.s s3, s4, s5, dyn

llvm/test/MC/RISCV/rv32zhinx-invalid.s

+# RUN: not llvm-mc -triple riscv32 -mattr=+experimental-zhinx %s 2>&1 | FileCheck %s
+
+# Invalid instructions
+flw fa4, 12(sp) # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'F' (Single-Precision Floating-Point)
+fsw a5, 12(sp) # CHECK: :[[@LINE]]:5: error: invalid operand for instruction
+fmv.x.h s0, s1 # CHECK: :[[@LINE]]:13: error: invalid operand for instruction
+
+# Invalid register names
+lw f15, 100(a0) # CHECK: :[[@LINE]]:4: error: invalid operand for instruction
+lw t1, 100(a10) # CHECK: :[[@LINE]]:12: error: expected register
+fadd.h a100, a2, a3 # CHECK: :[[@LINE]]:8: error: invalid operand for instruction

llvm/test/MC/RISCV/rv32zhinx-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zhinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zhinx -riscv-no-aliases -show-encoding \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM,CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv32 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zhinx -M no-aliases -d -r - \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM-AND-OBJ %s
+# RUN: llvm-mc -filetype=obj -triple=riscv64 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump --mattr=+experimental-zhinx -M no-aliases -d -r - \
+# RUN:     | FileCheck -check-prefixes=CHECK-ASM-AND-OBJ %s
+
+# CHECK-ASM-AND-OBJ: lh s0, 12(a0)
+# CHECK-ASM: encoding: [0x03,0x14,0xc5,0x00]
+lh s0, 12(a0)
+# CHECK-ASM-AND-OBJ: lh s1, 4(ra)
+# CHECK-ASM: encoding: [0x83,0x94,0x40,0x00]
+lh s1, +4(ra)
+# CHECK-ASM-AND-OBJ: lh s2, -2048(a3)
+# CHECK-ASM: encoding: [0x03,0x99,0x06,0x80]
+lh s2, -2048(x13)
+# CHECK-ASM-AND-OBJ: lh s3, -2048(s1)
+# CHECK-ASM: encoding: [0x83,0x99,0x04,0x80]
+lh s3, %lo(2048)(s1)
+# CHECK-ASM-AND-OBJ: lh s4, 2047(s2)
+# CHECK-ASM: encoding: [0x03,0x1a,0xf9,0x7f]
+lh s4, 2047(s2)
+# CHECK-ASM-AND-OBJ: lh s5, 0(s3)
+# CHECK-ASM: encoding: [0x83,0x9a,0x09,0x00]
+lh s5, 0(s3)
+
+# CHECK-ASM-AND-OBJ: sh s6, 2047(s4)
+# CHECK-ASM: encoding: [0xa3,0x1f,0x6a,0x7f]
+sh s6, 2047(s4)
+# CHECK-ASM-AND-OBJ: sh s7, -2048(s5)
+# CHECK-ASM: encoding: [0x23,0x90,0x7a,0x81]
+sh s7, -2048(s5)
+# CHECK-ASM-AND-OBJ: sh s0, -2048(s6)
+# CHECK-ASM: encoding: [0x23,0x10,0x8b,0x80]
+sh x8, %lo(2048)(s6)
+# CHECK-ASM-AND-OBJ: sh s1, 999(s7)
+# CHECK-ASM: encoding: [0xa3,0x93,0x9b,0x3e]
+sh x9, 999(s7)
+
+# CHECK-ASM-AND-OBJ: fmadd.h a0, a1, a2, a3, dyn
+# CHECK-ASM: encoding: [0x43,0xf5,0xc5,0x6c]
+fmadd.h x10, x11, x12, x13, dyn
+# CHECK-ASM-AND-OBJ: fmsub.h a4, a5, a6, a7, dyn
+# CHECK-ASM: encoding: [0x47,0xf7,0x07,0x8d]
+fmsub.h x14, x15, x16, x17, dyn
+# CHECK-ASM-AND-OBJ: fnmsub.h s2, s3, s4, s5, dyn
+# CHECK-ASM: encoding: [0x4b,0xf9,0x49,0xad]
+fnmsub.h x18, x19, x20, x21, dyn
+# CHECK-ASM-AND-OBJ: fnmadd.h s6, s7, s8, s9, dyn
+# CHECK-ASM: encoding: [0x4f,0xfb,0x8b,0xcd]
+fnmadd.h x22, x23, x24, x25, dyn
+
+# CHECK-ASM-AND-OBJ: fadd.h s10, s11, t3, dyn
+# CHECK-ASM: encoding: [0x53,0xfd,0xcd,0x05]
+fadd.h x26, x27, x28, dyn
+# CHECK-ASM-AND-OBJ: fsub.h t4, t5, t6, dyn
+# CHECK-ASM: encoding: [0xd3,0x7e,0xff,0x0d]
+fsub.h x29, x30, x31, dyn
+# CHECK-ASM-AND-OBJ: fmul.h s0, s1, s2, dyn
+# CHECK-ASM: encoding: [0x53,0xf4,0x24,0x15]
+fmul.h s0, s1, s2, dyn
+# CHECK-ASM-AND-OBJ: fdiv.h s3, s4, s5, dyn
+# CHECK-ASM: encoding: [0xd3,0x79,0x5a,0x1d]
+fdiv.h s3, s4, s5, dyn

llvm/test/MC/RISCV/rv32zhinxmin-invalid.s

+# RUN: not llvm-mc -triple riscv32 -mattr=+experimental-zhinxmin %s 2>&1 | FileCheck %s
+
+# Invalid instructions
+flw fa4, 12(sp) # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'F' (Single-Precision Floating-Point)
+fsw a5, 12(sp) # CHECK: :[[@LINE]]:5: error: invalid operand for instruction
+fmv.x.h s0, s1 # CHECK: :[[@LINE]]:13: error: invalid operand for instruction
+
+# Invalid register names
+lw f15, 100(a0) # CHECK: :[[@LINE]]:4: error: invalid operand for instruction
+lw t1, 100(a10) # CHECK: :[[@LINE]]:12: error: expected register
+fadd.h a100, a2, a3 # CHECK: :[[@LINE]]:8: error: invalid operand for instruction
+
+# Valid in Zhinx
+fmadd.h x10, x11, x12, x13, dyn # CHECK: :[[@LINE]]:1: error: instruction requires the following: 'Zhinx' (Half Float in Integer)

llvm/test/MC/RISCV/rvzdinx-aliases-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zdinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zdinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zdinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zdinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zdinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zdinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zdinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zdinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zdinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+
+##===----------------------------------------------------------------------===##
+## Aliases which omit the rounding mode.
+##===----------------------------------------------------------------------===##
+
+# CHECK-INST: fmadd.d a0, a2, a4, a6, dyn
+# CHECK-ALIAS: fmadd.d a0, a2, a4, a6
+fmadd.d x10, x12, x14, x16
+# CHECK-INST: fmsub.d a0, a2, a4, a6, dyn
+# CHECK-ALIAS: fmsub.d a0, a2, a4, a6
+fmsub.d x10, x12, x14, x16
+# CHECK-INST: fnmsub.d a0, a2, a4, a6, dyn
+# CHECK-ALIAS: fnmsub.d a0, a2, a4, a6
+fnmsub.d x10, x12, x14, x16
+# CHECK-INST: fnmadd.d a0, a2, a4, a6, dyn
+# CHECK-ALIAS: fnmadd.d a0, a2, a4, a6
+fnmadd.d x10, x12, x14, x16
+# CHECK-INST: fadd.d a0, a2, a4, dyn
+# CHECK-ALIAS: fadd.d a0, a2, a4
+fadd.d x10, x12, x14
+# CHECK-INST: fsub.d a0, a2, a4, dyn
+# CHECK-ALIAS: fsub.d a0, a2, a4
+fsub.d x10, x12, x14
+# CHECK-INST: fmul.d a0, a2, a4, dyn
+# CHECK-ALIAS: fmul.d a0, a2, a4
+fmul.d x10, x12, x14
+# CHECK-INST: fdiv.d a0, a2, a4, dyn
+# CHECK-ALIAS: fdiv.d a0, a2, a4
+fdiv.d x10, x12, x14

llvm/test/MC/RISCV/rvzfinx-aliases-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zfinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zfinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zfinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zfinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zfinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zfinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zfinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zfinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zfinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+
+##===----------------------------------------------------------------------===##
+## Aliases which omit the rounding mode.
+##===----------------------------------------------------------------------===##
+
+# CHECK-INST: fmadd.s a0, a1, a2, a3, dyn
+# CHECK-ALIAS: fmadd.s a0, a1, a2, a3
+fmadd.s x10, x11, x12, x13
+# CHECK-INST: fmsub.s a4, a5, a6, a7, dyn
+# CHECK-ALIAS: fmsub.s a4, a5, a6, a7
+fmsub.s x14, x15, x16, x17
+# CHECK-INST: fnmsub.s s2, s3, s4, s5, dyn
+# CHECK-ALIAS: fnmsub.s s2, s3, s4, s5
+fnmsub.s x18, x19, x20, x21
+# CHECK-INST: fnmadd.s s6, s7, s8, s9, dyn
+# CHECK-ALIAS: fnmadd.s s6, s7, s8, s9
+fnmadd.s x22, x23, x24, x25
+# CHECK-INST: fadd.s s10, s11, t3, dyn
+# CHECK-ALIAS: fadd.s s10, s11, t3
+fadd.s x26, x27, x28
+# CHECK-INST: fsub.s t4, t5, t6, dyn
+# CHECK-ALIAS: fsub.s t4, t5, t6
+fsub.s x29, x30, x31
+# CHECK-INST: fmul.s s0, s1, s2, dyn
+# CHECK-ALIAS: fmul.s s0, s1, s2
+fmul.s s0, s1, s2
+# CHECK-INST: fdiv.s s3, s4, s5, dyn
+# CHECK-ALIAS: fdiv.s s3, s4, s5
+fdiv.s s3, s4, s5

llvm/test/MC/RISCV/rvzhinx-aliases-valid.s

+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zhinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv32 -mattr=+experimental-zhinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zhinx -riscv-no-aliases \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc %s -triple=riscv64 -mattr=+experimental-zhinx \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zhinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv32 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zhinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zhinx -M no-aliases - \
+# RUN:     | FileCheck -check-prefix=CHECK-INST %s
+# RUN: llvm-mc -filetype=obj -triple riscv64 -mattr=+experimental-zhinx %s \
+# RUN:     | llvm-objdump -d --mattr=+experimental-zhinx - \
+# RUN:     | FileCheck -check-prefix=CHECK-ALIAS %s
+
+##===----------------------------------------------------------------------===##
+## Aliases which omit the rounding mode.
+##===----------------------------------------------------------------------===##
+
+# CHECK-INST: fmadd.h a0, a1, a2, a3, dyn
+# CHECK-ALIAS: fmadd.h a0, a1, a2, a3
+fmadd.h x10, x11, x12, x13
+# CHECK-INST: fmsub.h a4, a5, a6, a7, dyn
+# CHECK-ALIAS: fmsub.h a4, a5, a6, a7
+fmsub.h x14, x15, x16, x17
+# CHECK-INST: fnmsub.h s2, s3, s4, s5, dyn
+# CHECK-ALIAS: fnmsub.h s2, s3, s4, s5
+fnmsub.h x18, x19, x20, x21
+# CHECK-INST: fnmadd.h s6, s7, s8, s9, dyn
+# CHECK-ALIAS: fnmadd.h s6, s7, s8, s9
+fnmadd.h x22, x23, x24, x25
+# CHECK-INST: fadd.h s10, s11, t3, dyn
+# CHECK-ALIAS: fadd.h s10, s11, t3
+fadd.h x26, x27, x28
+# CHECK-INST: fsub.h t4, t5, t6, dyn
+# CHECK-ALIAS: fsub.h t4, t5, t6
+fsub.h x29, x30, x31
+# CHECK-INST: fmul.h s0, s1, s2, dyn
+# CHECK-ALIAS: fmul.h s0, s1, s2
+fmul.h s0, s1, s2
+# CHECK-INST: fdiv.h s3, s4, s5, dyn
+# CHECK-ALIAS: fdiv.h s3, s4, s5
+fdiv.h s3, s4, s5