[RISCV] Match widening fp instructions with same fpext used in multiple operands #125803
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
…le operands Because the fpext has a single use constraint on it we can't match cases where it's used for both operands. Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns. (For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these)
|
@llvm/pr-subscribers-backend-risc-v Author: Luke Lau (lukel97) ChangesBecause the fpext has a single use constraint on it we can't match cases where it's used for both operands. Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns. (For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these) Full diff: https://github.com/llvm/llvm-project/pull/125803.diff 6 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
index 8f77b2ce34d1f1..72f91f71e78537 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td
@@ -536,19 +536,19 @@ multiclass VPatWidenBinaryFPSDNode_VV_VF<SDNode op, string instruction_name> {
defvar wti = vtiToWti.Wti;
let Predicates = !listconcat(GetVTypePredicates<vti>.Predicates,
GetVTypePredicates<wti>.Predicates) in {
- def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs2),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (wti.Vector (riscv_fpextend_vl_oneuse
+ (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs1),
(vti.Mask true_mask), (XLenVT srcvalue)))),
(!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX)
(wti.Vector (IMPLICIT_DEF)), vti.RegClass:$rs2,
vti.RegClass:$rs1, vti.AVL, vti.Log2SEW, TA_MA)>;
- def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs2),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (wti.Vector (riscv_fpextend_vl_oneuse
+ (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
(vti.Mask true_mask), (XLenVT srcvalue)))),
(!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
@@ -571,10 +571,10 @@ multiclass VPatWidenBinaryFPSDNode_VV_VF_RM<SDNode op, string instruction_name>
defvar wti = vtiToWti.Wti;
let Predicates = !listconcat(GetVTypePredicates<vti>.Predicates,
GetVTypePredicates<wti>.Predicates) in {
- def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs2),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (wti.Vector (riscv_fpextend_vl_oneuse
+ (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs1),
(vti.Mask true_mask), (XLenVT srcvalue)))),
(!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX#"_E"#vti.SEW)
@@ -584,10 +584,10 @@ multiclass VPatWidenBinaryFPSDNode_VV_VF_RM<SDNode op, string instruction_name>
// RISCVInsertReadWriteCSR
FRM_DYN,
vti.AVL, vti.Log2SEW, TA_MA)>;
- def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(op (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs2),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (wti.Vector (riscv_fpextend_vl_oneuse
+ (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector (SplatFPOp (vti.Scalar vti.ScalarRegClass:$rs1))),
(vti.Mask true_mask), (XLenVT srcvalue)))),
(!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX#"_E"#vti.SEW)
@@ -669,10 +669,10 @@ multiclass VPatWidenFPMulAccSDNode_VV_VF_RM<string instruction_name,
!if(!eq(vti.Scalar, bf16),
[HasStdExtZvfbfwma],
[])) in {
- def : Pat<(fma (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(fma (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs1),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (wti.Vector (riscv_fpextend_vl_oneuse
+ (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs2),
(vti.Mask true_mask), (XLenVT srcvalue))),
(wti.Vector wti.RegClass:$rd)),
@@ -749,11 +749,11 @@ multiclass VPatWidenFPMulSacSDNode_VV_VF_RM<string instruction_name> {
defvar suffix = vti.LMul.MX # "_E" # vti.SEW;
let Predicates = !listconcat(GetVTypePredicates<vti>.Predicates,
GetVTypePredicates<wti>.Predicates) in {
- def : Pat<(fma (wti.Vector (riscv_fpextend_vl_oneuse
+ def : Pat<(fma (wti.Vector (riscv_fpextend_vl_oneuser
(vti.Vector vti.RegClass:$rs1),
(vti.Mask true_mask), (XLenVT srcvalue))),
- (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
- (vti.Mask true_mask), (XLenVT srcvalue)),
+ (riscv_fpextend_vl_oneuser (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), (XLenVT srcvalue)),
(fneg wti.RegClass:$rd)),
(!cast<Instruction>(instruction_name#"_VV_"#suffix)
wti.RegClass:$rd, vti.RegClass:$rs1, vti.RegClass:$rs2,
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
index 333ae52534681a..0e6f807ae26af5 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
@@ -554,6 +554,11 @@ def riscv_fpextend_vl_oneuse : PatFrag<(ops node:$A, node:$B, node:$C),
return N->hasOneUse();
}]>;
+def riscv_fpextend_vl_oneuser : PatFrag<(ops node:$A, node:$B, node:$C),
+ (riscv_fpextend_vl node:$A, node:$B, node:$C), [{
+ return !N->use_empty() && all_equal(N->users());
+}]>;
+
def riscv_vfmadd_vl_oneuse : PatFrag<(ops node:$A, node:$B, node:$C, node:$D,
node:$E),
(riscv_vfmadd_vl node:$A, node:$B,
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfwadd-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfwadd-sdnode.ll
index 68014ff4206f8a..f7d287a088cc3e 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfwadd-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfwadd-sdnode.ll
@@ -323,3 +323,15 @@ define <vscale x 8 x double> @vfwadd_wf_nxv8f64_2(<vscale x 8 x double> %va, flo
%vd = fadd <vscale x 8 x double> %va, %splat
ret <vscale x 8 x double> %vd
}
+
+define <vscale x 1 x double> @vfwadd_vv_nxv1f64_same_op(<vscale x 1 x float> %va) {
+; CHECK-LABEL: vfwadd_vv_nxv1f64_same_op:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vfwadd.vv v9, v8, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %vb = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
+ %vc = fadd <vscale x 1 x double> %vb, %vb
+ ret <vscale x 1 x double> %vc
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfwmacc-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfwmacc-sdnode.ll
index f69b2346226ee9..63113b87809894 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfwmacc-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfwmacc-sdnode.ll
@@ -1764,3 +1764,28 @@ define <vscale x 8 x double> @vfwnmsac_fv_nxv8f64(<vscale x 8 x double> %va, <vs
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %vf, <vscale x 8 x double> %va)
ret <vscale x 8 x double> %vg
}
+
+define <vscale x 1 x double> @vfwma_vv_nxv1f64_same_op(<vscale x 1 x float> %va, <vscale x 1 x double> %vb) {
+; CHECK-LABEL: vfwma_vv_nxv1f64_same_op:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vfwmacc.vv v9, v8, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %vc = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
+ %vd = call <vscale x 1 x double> @llvm.fma(<vscale x 1 x double> %vc, <vscale x 1 x double> %vc, <vscale x 1 x double> %vb)
+ ret <vscale x 1 x double> %vd
+}
+
+define <vscale x 1 x double> @vfwmsac_vv_nxv1f64_same_op(<vscale x 1 x float> %va, <vscale x 1 x double> %vb) {
+; CHECK-LABEL: vfwmsac_vv_nxv1f64_same_op:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vfwmsac.vv v9, v8, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %vc = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
+ %vd = fneg <vscale x 1 x double> %vb
+ %ve = call <vscale x 1 x double> @llvm.fma(<vscale x 1 x double> %vc, <vscale x 1 x double> %vc, <vscale x 1 x double> %vd)
+ ret <vscale x 1 x double> %ve
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfwmul-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfwmul-sdnode.ll
index f00ff4b6d2cec2..8cc8c5cffca6b2 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfwmul-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfwmul-sdnode.ll
@@ -175,3 +175,15 @@ define <vscale x 8 x double> @vfwmul_vf_nxv8f64_2(<vscale x 8 x float> %va, floa
%ve = fmul <vscale x 8 x double> %vc, %splat
ret <vscale x 8 x double> %ve
}
+
+define <vscale x 1 x double> @vfwmul_vv_nxv1f64_same_op(<vscale x 1 x float> %va) {
+; CHECK-LABEL: vfwmul_vv_nxv1f64_same_op:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vfwmul.vv v9, v8, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %vb = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
+ %vc = fmul <vscale x 1 x double> %vb, %vb
+ ret <vscale x 1 x double> %vc
+}
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfwsub-sdnode.ll b/llvm/test/CodeGen/RISCV/rvv/vfwsub-sdnode.ll
index b9f66d5d30825d..d0cb64d9866618 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfwsub-sdnode.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfwsub-sdnode.ll
@@ -323,3 +323,15 @@ define <vscale x 8 x double> @vfwsub_wf_nxv8f64_2(<vscale x 8 x double> %va, flo
%vd = fsub <vscale x 8 x double> %va, %splat
ret <vscale x 8 x double> %vd
}
+
+define <vscale x 1 x double> @vfwsub_vv_nxv1f64_same_op(<vscale x 1 x float> %va) {
+; CHECK-LABEL: vfwsub_vv_nxv1f64_same_op:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
+; CHECK-NEXT: vfwsub.vv v9, v8, v8
+; CHECK-NEXT: vmv1r.v v8, v9
+; CHECK-NEXT: ret
+ %vb = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
+ %vc = fsub <vscale x 1 x double> %vb, %vb
+ ret <vscale x 1 x double> %vc
+}
|
wangpc-pp
reviewed
Feb 5, 2025
Icohedron
pushed a commit
to Icohedron/llvm-project
that referenced
this pull request
Feb 11, 2025
…le operands (llvm#125803) Because the fpext has a single use constraint on it we can't match cases where it's used for both operands. Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns. (For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these)
joaosaffran
pushed a commit
to joaosaffran/llvm-project
that referenced
this pull request
Feb 14, 2025
…le operands (llvm#125803) Because the fpext has a single use constraint on it we can't match cases where it's used for both operands. Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns. (For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these)
sivan-shani
pushed a commit
to sivan-shani/llvm-project
that referenced
this pull request
Feb 24, 2025
…le operands (llvm#125803) Because the fpext has a single use constraint on it we can't match cases where it's used for both operands. Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns. (For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these)
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.
Because the fpext has a single use constraint on it we can't match cases where it's used for both operands.
Introduce a new PatFrag that allows multiple uses on a single user and use it for the binary patterns, and some ternary patterns.
(For some of the ternary patterns there is a fneg that counts as a separate user, we still need to handle these)