[RISCV] Add GetVTypeMinimalPredicates for the operation supported by zvfhmin. NFC. #143847
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…zvfhmin. NFC. This patch adds a new `GetVTypeMinimalPredicates` for `f16` operation supported by `Zvfhmin`. This is a refactor patch for implementing vector compute support for bf16 (Zvfbfa), so that we can add `bf16` type to `GetVTypePredicates` for vector computation.
|
@llvm/pr-subscribers-backend-risc-v Author: Jim Lin (tclin914) ChangesThis patch adds a new Full diff: https://github.com/llvm/llvm-project/pull/143847.diff 3 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
index 281f8d55932b9..1e48d42ac8f38 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
@@ -783,6 +783,15 @@ class GetVTypePredicates<VTypeInfo vti> {
true : [HasVInstructions]);
}
+class GetVTypeMinimalPredicates<VTypeInfo vti> {
+ list<Predicate> Predicates = !cond(!eq(vti.Scalar, f16) : [HasVInstructionsF16Minimal],
+ !eq(vti.Scalar, bf16) : [HasVInstructionsBF16Minimal],
+ !eq(vti.Scalar, f32) : [HasVInstructionsAnyF],
+ !eq(vti.Scalar, f64) : [HasVInstructionsF64],
+ !eq(vti.SEW, 64) : [HasVInstructionsI64],
+ true : [HasVInstructions]);
+}
+
class VPseudoUSLoadNoMask<VReg RetClass,
int EEW,
DAGOperand sewop = sew> :
@@ -4568,7 +4577,7 @@ multiclass VPatUnaryS_M<string intrinsic_name,
multiclass VPatUnaryV_V_AnyMask<string intrinsic, string instruction,
list<VTypeInfo> vtilist> {
foreach vti = vtilist in {
- let Predicates = GetVTypePredicates<vti>.Predicates in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
def : VPatUnaryAnyMask<intrinsic, instruction, "VM",
vti.Vector, vti.Vector, vti.Mask,
vti.Log2SEW, vti.LMul, vti.RegClass, vti.RegClass>;
@@ -4887,7 +4896,7 @@ multiclass VPatBinaryV_VV_INT<string intrinsic, string instruction,
list<VTypeInfo> vtilist> {
foreach vti = vtilist in {
defvar ivti = GetIntVTypeInfo<vti>.Vti;
- let Predicates = GetVTypePredicates<vti>.Predicates in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
defm : VPatBinary<intrinsic,
instruction # "_VV_" # vti.LMul.MX # "_E" # vti.SEW,
vti.Vector, vti.Vector, ivti.Vector, vti.Mask,
@@ -4950,7 +4959,7 @@ multiclass VPatBinaryV_VX_RM<string intrinsic, string instruction,
multiclass VPatBinaryV_VX_INT<string intrinsic, string instruction,
list<VTypeInfo> vtilist> {
foreach vti = vtilist in
- let Predicates = GetVTypePredicates<vti>.Predicates in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
defm : VPatBinary<intrinsic, instruction # "_VX_" # vti.LMul.MX,
vti.Vector, vti.Vector, XLenVT, vti.Mask,
vti.Log2SEW, vti.RegClass,
@@ -4960,7 +4969,7 @@ multiclass VPatBinaryV_VX_INT<string intrinsic, string instruction,
multiclass VPatBinaryV_VI<string intrinsic, string instruction,
list<VTypeInfo> vtilist, Operand imm_type> {
foreach vti = vtilist in
- let Predicates = GetVTypePredicates<vti>.Predicates in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
defm : VPatBinary<intrinsic, instruction # "_VI_" # vti.LMul.MX,
vti.Vector, vti.Vector, XLenVT, vti.Mask,
vti.Log2SEW, vti.RegClass,
@@ -5887,12 +5896,11 @@ multiclass VPatConversionWF_VF<string intrinsic, string instruction,
defvar fvti = fvtiToFWti.Vti;
defvar fwti = fvtiToFWti.Wti;
// Define vfwcvt.f.f.v for f16 when Zvfhmin is enable.
- let Predicates = !if(!eq(fvti.Scalar, f16), [HasVInstructionsF16Minimal],
- !listconcat(GetVTypePredicates<fvti>.Predicates,
- GetVTypePredicates<fwti>.Predicates)) in
- defm : VPatConversion<intrinsic, instruction, "V",
- fwti.Vector, fvti.Vector, fwti.Mask, fvti.Log2SEW,
- fvti.LMul, fwti.RegClass, fvti.RegClass, isSEWAware>;
+ let Predicates = !listconcat(GetVTypeMinimalPredicates<fvti>.Predicates,
+ GetVTypeMinimalPredicates<fwti>.Predicates) in
+ defm : VPatConversion<intrinsic, instruction, "V",
+ fwti.Vector, fvti.Vector, fwti.Mask, fvti.Log2SEW,
+ fvti.LMul, fwti.RegClass, fvti.RegClass, isSEWAware>;
}
}
@@ -5979,8 +5987,9 @@ multiclass VPatConversionVF_WF_RM<string intrinsic, string instruction,
foreach fvtiToFWti = wlist in {
defvar fvti = fvtiToFWti.Vti;
defvar fwti = fvtiToFWti.Wti;
- let Predicates = !listconcat(GetVTypePredicates<fvti>.Predicates,
- GetVTypePredicates<fwti>.Predicates) in
+ // Define vfncvt.f.f.w for f16 when Zvfhmin is enable.
+ let Predicates = !listconcat(GetVTypeMinimalPredicates<fvti>.Predicates,
+ GetVTypeMinimalPredicates<fwti>.Predicates) in
defm : VPatConversionRoundingMode<intrinsic, instruction, "W",
fvti.Vector, fwti.Vector, fvti.Mask, fvti.Log2SEW,
fvti.LMul, fvti.RegClass, fwti.RegClass,
@@ -7005,8 +7014,7 @@ defm : VPatBinaryV_VM_XM_IM<"int_riscv_vmerge", "PseudoVMERGE">;
// 11.16. Vector Integer Move Instructions
//===----------------------------------------------------------------------===//
foreach vti = AllVectors in {
- let Predicates = !if(!eq(vti.Scalar, f16), [HasVInstructionsF16Minimal],
- GetVTypePredicates<vti>.Predicates) in {
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in {
def : Pat<(vti.Vector (int_riscv_vmv_v_v (vti.Vector vti.RegClass:$passthru),
(vti.Vector vti.RegClass:$rs1),
VLOpFrag)),
@@ -7201,8 +7209,7 @@ defm : VPatConversionVI_VF<"int_riscv_vfclass", "PseudoVFCLASS">;
// NOTE: Clang previously used int_riscv_vfmerge for vector-vector, but now uses
// int_riscv_vmerge. Support both for compatibility.
foreach vti = AllFloatVectors in {
- let Predicates = !if(!eq(vti.Scalar, f16), [HasVInstructionsF16Minimal],
- GetVTypePredicates<vti>.Predicates) in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
defm : VPatBinaryCarryInTAIL<"int_riscv_vmerge", "PseudoVMERGE", "VVM",
vti.Vector,
vti.Vector, vti.Vector, vti.Mask,
@@ -7281,16 +7288,8 @@ defm : VPatConversionVF_WI_RM<"int_riscv_vfncvt_f_xu_w", "PseudoVFNCVT_F_XU",
isSEWAware=1>;
defm : VPatConversionVF_WI_RM<"int_riscv_vfncvt_f_x_w", "PseudoVFNCVT_F_X",
isSEWAware=1>;
-defvar WidenableFloatVectorsExceptF16 = !filter(fvtiToFWti, AllWidenableFloatVectors,
- !ne(fvtiToFWti.Vti.Scalar, f16));
-defm : VPatConversionVF_WF_RM<"int_riscv_vfncvt_f_f_w", "PseudoVFNCVT_F_F",
- WidenableFloatVectorsExceptF16, isSEWAware=1>;
-// Define vfncvt.f.f.w for f16 when Zvfhmin is enable.
-defvar F16WidenableFloatVectors = !filter(fvtiToFWti, AllWidenableFloatVectors,
- !eq(fvtiToFWti.Vti.Scalar, f16));
-let Predicates = [HasVInstructionsF16Minimal] in
defm : VPatConversionVF_WF_RM<"int_riscv_vfncvt_f_f_w", "PseudoVFNCVT_F_F",
- F16WidenableFloatVectors, isSEWAware=1>;
+ AllWidenableFloatVectors, isSEWAware=1>;
defm : VPatConversionVF_WF_BF_RM<"int_riscv_vfncvtbf16_f_f_w",
"PseudoVFNCVTBF16_F_F", isSEWAware=1>;
defm : VPatConversionVF_WF<"int_riscv_vfncvt_rod_f_f_w", "PseudoVFNCVT_ROD_F_F",
@@ -7425,10 +7424,7 @@ defm : VPatBinaryV_VV_INT_EEW<"int_riscv_vrgatherei16_vv", "PseudoVRGATHEREI16",
eew=16, vtilist=AllIntegerVectors>;
defm : VPatBinaryV_VV_VX_VI_INT<"int_riscv_vrgather", "PseudoVRGATHER",
- AllFloatVectorsExceptFP16, uimm5>;
-let Predicates = [HasVInstructionsF16Minimal] in
- defm : VPatBinaryV_VV_VX_VI_INT<"int_riscv_vrgather", "PseudoVRGATHER",
- AllFP16Vectors, uimm5>;
+ AllFloatVectors, uimm5>;
defm : VPatBinaryV_VV_VX_VI_INT<"int_riscv_vrgather", "PseudoVRGATHER",
AllBFloatVectors, uimm5>;
defm : VPatBinaryV_VV_INT_EEW<"int_riscv_vrgatherei16_vv", "PseudoVRGATHEREI16",
@@ -7437,9 +7433,7 @@ defm : VPatBinaryV_VV_INT_EEW<"int_riscv_vrgatherei16_vv", "PseudoVRGATHEREI16",
// 16.5. Vector Compress Instruction
//===----------------------------------------------------------------------===//
defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllIntegerVectors>;
-defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllFloatVectorsExceptFP16>;
-let Predicates = [HasVInstructionsF16Minimal] in
- defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllFP16Vectors>;
+defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllFloatVectors>;
defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllBFloatVectors>;
// Include the non-intrinsic ISel patterns
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
index e318a78285a2e..520959b0896f7 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
@@ -864,8 +864,7 @@ multiclass VPatAVGADD_VV_VX_RM<SDNode vop, int vxrm, string suffix = ""> {
// 7.4. Vector Unit-Stride Instructions
foreach vti = AllVectors in
- let Predicates = !if(!eq(vti.Scalar, f16), [HasVInstructionsF16Minimal],
- GetVTypePredicates<vti>.Predicates) in
+ let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in
defm : VPatUSLoadStoreSDNode<vti.Vector, vti.RegClass, vti.Log2SEW, vti.LMul,
vti.AVL, vti.RegClass>;
foreach mti = AllMasks in
@@ -1449,9 +1448,8 @@ defm : VPatNConvertI2FPSDNode_W_RM<any_uint_to_fp, "PseudoVFNCVT_F_XU_W">;
foreach fvtiToFWti = AllWidenableFloatVectors in {
defvar fvti = fvtiToFWti.Vti;
defvar fwti = fvtiToFWti.Wti;
- let Predicates = !if(!eq(fvti.Scalar, f16), [HasVInstructionsF16Minimal],
- !listconcat(GetVTypePredicates<fvti>.Predicates,
- GetVTypePredicates<fwti>.Predicates)) in
+ let Predicates = !listconcat(GetVTypeMinimalPredicates<fvti>.Predicates,
+ GetVTypeMinimalPredicates<fwti>.Predicates) in
def : Pat<(fvti.Vector (fpround (fwti.Vector fwti.RegClass:$rs1))),
(!cast<Instruction>("PseudoVFNCVT_F_F_W_"#fvti.LMul.MX#"_E"#fvti.SEW)
(fvti.Vector (IMPLICIT_DEF)),
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
index 1da4adc8c3125..52a1c2edd76f2 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
@@ -2687,9 +2687,9 @@ defm : VPatWConvertI2FPVL_V<any_riscv_sint_to_fp_vl, "PseudoVFWCVT_F_X_V">;
foreach fvtiToFWti = AllWidenableFloatVectors in {
defvar fvti = fvtiToFWti.Vti;
defvar fwti = fvtiToFWti.Wti;
- let Predicates = !if(!eq(fvti.Scalar, f16), [HasVInstructionsF16Minimal],
- !listconcat(GetVTypePredicates<fvti>.Predicates,
- GetVTypePredicates<fwti>.Predicates)) in
+ // Define vfwcvt.f.f.v for f16 when Zvfhmin is enable.
+ let Predicates = !listconcat(GetVTypeMinimalPredicates<fvti>.Predicates,
+ GetVTypeMinimalPredicates<fwti>.Predicates) in
def : Pat<(fwti.Vector (any_riscv_fpextend_vl
(fvti.Vector fvti.RegClass:$rs1),
(fvti.Mask VMV0:$vm),
@@ -2730,10 +2730,9 @@ defm : VPatNConvertI2FP_RM_VL_W<riscv_vfcvt_rm_f_x_vl, "PseudoVFNCVT_F_X_W">;
foreach fvtiToFWti = AllWidenableFloatVectors in {
defvar fvti = fvtiToFWti.Vti;
defvar fwti = fvtiToFWti.Wti;
- // Define vfwcvt.f.f.v for f16 when Zvfhmin is enable.
- let Predicates = !if(!eq(fvti.Scalar, f16), [HasVInstructionsF16Minimal],
- !listconcat(GetVTypePredicates<fvti>.Predicates,
- GetVTypePredicates<fwti>.Predicates)) in {
+ // Define vfncvt.f.f.w for f16 when Zvfhmin is enable.
+ let Predicates = !listconcat(GetVTypeMinimalPredicates<fvti>.Predicates,
+ GetVTypeMinimalPredicates<fwti>.Predicates) in {
def : Pat<(fvti.Vector (any_riscv_fpround_vl
(fwti.Vector fwti.RegClass:$rs1),
(fwti.Mask VMV0:$vm), VLOpFrag)),
|
lukel97
reviewed
Jun 12, 2025
topperc
reviewed
Jun 13, 2025
| @@ -4568,7 +4577,7 @@ multiclass VPatUnaryS_M<string intrinsic_name, | |||
| multiclass VPatUnaryV_V_AnyMask<string intrinsic, string instruction, | |||
| list<VTypeInfo> vtilist> { | |||
| foreach vti = vtilist in { | |||
| let Predicates = GetVTypePredicates<vti>.Predicates in | |||
| let Predicates = GetVTypeMinimalPredicates<vti>.Predicates in | |||
There was a problem hiding this comment.
Why weren't any tests failing if we using the wrong predicate?
There was a problem hiding this comment.
Why weren't any tests failing if we using the wrong predicate?
Currently, only int_riscv_vcompress uses the multiclass VPatUnaryV_V_AnyMask.
Originally, the predicate for the fp16 vector type was written as:
let Predicates = [HasVInstructionsF16Minimal] in
defm : VPatUnaryV_V_AnyMask<"int_riscv_vcompress", "PseudoVCOMPRESS", AllFP16Vectors>;
I simply moved this predicate into the multiclass VPatUnaryV_V_AnyMask by replacing GetVTypePredicates with GetVTypeMinimalPredicates.
I hope I didn’t misunderstand your concern.
lukel97
approved these changes
Jun 13, 2025
Comment on lines
+786
to
+794
| class GetVTypeMinimalPredicates<VTypeInfo vti> { | ||
| list<Predicate> Predicates = !cond(!eq(vti.Scalar, f16) : [HasVInstructionsF16Minimal], | ||
| !eq(vti.Scalar, bf16) : [HasVInstructionsBF16Minimal], | ||
| !eq(vti.Scalar, f32) : [HasVInstructionsAnyF], | ||
| !eq(vti.Scalar, f64) : [HasVInstructionsF64], | ||
| !eq(vti.SEW, 64) : [HasVInstructionsI64], | ||
| true : [HasVInstructions]); | ||
| } | ||
|
|
There was a problem hiding this comment.
Maybe for a separate PR, I think for anything that's not a widening/converting instruction e.g. vmerge/vrgather/vcompress/vle/vse, we can just use the integer predicate, maybe something like:
class GetVTypeIntPredicates<VTypeInfo vti> {
defvar ivti = GetIntVTypeInfo<vti>;
list<Predicate> Predicates = GetVTypePredicates<ivti>;
}
Since vmerge.vvm with a fp16 vector isn't really a zvfhmin instruction, it only requires zve32x.
There was a problem hiding this comment.
I think this is somewhat related to #143975
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch adds a new
GetVTypeMinimalPredicatesforf16operation supported byZvfhmin. Split the type predicates for minimal support and full compute support. This is a refactor patch for implementing vector compute support for bf16 (Zvfbfa), that we can checkbf16type whether withZvfbfaextension inGetVTypePredicates.