Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[RISCV] Check for COPY_TO_REGCLASS in usesAllOnesMask #67037

Merged
merged 2 commits into from
Sep 22, 2023
Merged

[RISCV] Check for COPY_TO_REGCLASS in usesAllOnesMask #67037

merged 2 commits into from
Sep 22, 2023

Conversation

lukel97
Copy link
Contributor

@lukel97 lukel97 commented Sep 21, 2023 •
edited

Sometimes with mask vectors that have been widened, there is a
CopyToRegClass node in between the VMSET and the CopyToReg.

This is a resurrection of https://reviews.llvm.org/D148524, and is needed to
remove the mask operand when it's extracted from a subvector as planned in
#66267 (comment)

@lukel97
[RISCV] Check for COPY_TO_REGCLASS in usesAllOnesMask
94a0fdc 
Sometimes with mask vectors that have been widened, there is a
CopyToRegClass node in between the VMSET and the CopyToReg.

This is a resurrection https://reviews.llvm.org/D148524, and is needed to
remove the mask operand when it's extracted from a subregister as planned in
llvm#66267 (comment)
@llvmbot
Copy link
Collaborator

llvmbot commented Sep 21, 2023

@llvm/pr-subscribers-backend-risc-v

Changes

Sometimes with mask vectors that have been widened, there is a
CopyToRegClass node in between the VMSET and the CopyToReg.

This is a resurrection https://reviews.llvm.org/D148524, and is needed to
remove the mask operand when it's extracted from a subregister as planned in
#66267 (comment)

Patch is 38.82 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/67037.diff

16 Files Affected:

  • (modified) llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp (+6)
  • (modified) llvm/test/CodeGen/RISCV/rvv/nearbyint-vp.ll (+13-17)
  • (modified) llvm/test/CodeGen/RISCV/rvv/rint-vp.ll (+13-18)
  • (modified) llvm/test/CodeGen/RISCV/rvv/round-vp.ll (+13-18)
  • (modified) llvm/test/CodeGen/RISCV/rvv/roundeven-vp.ll (+13-18)
  • (modified) llvm/test/CodeGen/RISCV/rvv/roundtozero-vp.ll (+13-18)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vcopysign-vp.ll (+3-4)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfabs-vp.ll (+5-6)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfadd-vp.ll (+11-13)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfdiv-vp.ll (+11-13)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfmax-vp.ll (+3-4)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfmin-vp.ll (+3-4)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfmul-vp.ll (+11-13)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfneg-vp.ll (+5-6)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfsqrt-vp.ll (+5-6)
  • (modified) llvm/test/CodeGen/RISCV/rvv/vfsub-vp.ll (+11-13)
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
index e6d0346c45e8d54..0917e95dd1f2eee 100644
--- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -3188,6 +3188,12 @@ static bool usesAllOnesMask(SDValue MaskOp, SDValue GlueOp) {
   // Check the instruction defining V0; it needs to be a VMSET pseudo.
   SDValue MaskSetter = Glued->getOperand(2);
 
+  // Sometimes the VMSET is wrapped in a COPY_TO_REGCLASS node, e.g. if the mask
+  // was extracted from a larger register.
+  if (MaskSetter->isMachineOpcode() &&
+      MaskSetter->getMachineOpcode() == RISCV::COPY_TO_REGCLASS)
+    MaskSetter = MaskSetter->getOperand(0);
+
   const auto IsVMSet = [](unsigned Opc) {
     return Opc == RISCV::PseudoVMSET_M_B1 || Opc == RISCV::PseudoVMSET_M_B16 ||
            Opc == RISCV::PseudoVMSET_M_B2 || Opc == RISCV::PseudoVMSET_M_B32 ||
diff --git a/llvm/test/CodeGen/RISCV/rvv/nearbyint-vp.ll b/llvm/test/CodeGen/RISCV/rvv/nearbyint-vp.ll
index cf984ff149a5a12..7c354c3714c6f44 100644
--- a/llvm/test/CodeGen/RISCV/rvv/nearbyint-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/nearbyint-vp.ll
@@ -555,7 +555,7 @@ define <vscale x 32 x half> @vp_nearbyint_nxv32f16_unmasked(<vscale x 32 x half>
 ; ZVFHMIN-NEXT:    sub sp, sp, a1
 ; ZVFHMIN-NEXT:    .cfi_escape 0x0f, 0x0d, 0x72, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0xa2, 0x38, 0x00, 0x1e, 0x22 # sp + 16 + 8 * vlenb
 ; ZVFHMIN-NEXT:    vsetvli a1, zero, e8, m4, ta, ma
-; ZVFHMIN-NEXT:    vmset.m v1
+; ZVFHMIN-NEXT:    vmset.m v16
 ; ZVFHMIN-NEXT:    csrr a2, vlenb
 ; ZVFHMIN-NEXT:    slli a1, a2, 1
 ; ZVFHMIN-NEXT:    sub a3, a0, a1
@@ -564,21 +564,21 @@ define <vscale x 32 x half> @vp_nearbyint_nxv32f16_unmasked(<vscale x 32 x half>
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v17, v1, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v16, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    lui a2, 307200
 ; ZVFHMIN-NEXT:    fmv.w.x fa5, a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v17, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
 ; ZVFHMIN-NEXT:    frflags a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
 ; ZVFHMIN-NEXT:    fsflags a2
@@ -591,23 +591,19 @@ define <vscale x 32 x half> @vp_nearbyint_nxv32f16_unmasked(<vscale x 32 x half>
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
 ; ZVFHMIN-NEXT:    addi a1, sp, 16
-; ZVFHMIN-NEXT:    vl8r.v v16, (a1) # Unknown-size Folded Reload
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v16
+; ZVFHMIN-NEXT:    vl8r.v v24, (a1) # Unknown-size Folded Reload
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v1
-; ZVFHMIN-NEXT:    vfabs.v v16, v24, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v1, v16, fa5, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v24, v16
+; ZVFHMIN-NEXT:    vmflt.vf v0, v24, fa5
 ; ZVFHMIN-NEXT:    frflags a0
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v1
-; ZVFHMIN-NEXT:    vfcvt.x.f.v v16, v24, v0.t
-; ZVFHMIN-NEXT:    vfcvt.f.x.v v16, v16, v0.t
+; ZVFHMIN-NEXT:    vfcvt.x.f.v v24, v16, v0.t
+; ZVFHMIN-NEXT:    vfcvt.f.x.v v24, v24, v0.t
 ; ZVFHMIN-NEXT:    fsflags a0
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vfsgnj.vv v24, v16, v24, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
 ; ZVFHMIN-NEXT:    slli a0, a0, 3
 ; ZVFHMIN-NEXT:    add sp, sp, a0
diff --git a/llvm/test/CodeGen/RISCV/rvv/rint-vp.ll b/llvm/test/CodeGen/RISCV/rvv/rint-vp.ll
index ebb186b197b4133..04761d4e7bfc4af 100644
--- a/llvm/test/CodeGen/RISCV/rvv/rint-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/rint-vp.ll
@@ -517,47 +517,42 @@ define <vscale x 32 x half> @vp_rint_nxv32f16_unmasked(<vscale x 32 x half> %va,
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v17, v16, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v16, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    lui a2, 307200
 ; ZVFHMIN-NEXT:    fmv.w.x fa5, a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v17, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
 ; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v20, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v24
 ; ZVFHMIN-NEXT:    bltu a0, a1, .LBB11_2
 ; ZVFHMIN-NEXT:  # %bb.1:
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
 ; ZVFHMIN-NEXT:    addi a1, sp, 16
-; ZVFHMIN-NEXT:    vl8r.v v8, (a1) # Unknown-size Folded Reload
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
+; ZVFHMIN-NEXT:    vl8r.v v24, (a1) # Unknown-size Folded Reload
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
-; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v24, v16
+; ZVFHMIN-NEXT:    vmflt.vf v0, v24, fa5
+; ZVFHMIN-NEXT:    vfcvt.x.f.v v24, v16, v0.t
+; ZVFHMIN-NEXT:    vfcvt.f.x.v v24, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v16, v24
-; ZVFHMIN-NEXT:    vmv8r.v v8, v16
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
 ; ZVFHMIN-NEXT:    slli a0, a0, 3
 ; ZVFHMIN-NEXT:    add sp, sp, a0
diff --git a/llvm/test/CodeGen/RISCV/rvv/round-vp.ll b/llvm/test/CodeGen/RISCV/rvv/round-vp.ll
index ce644797738ec74..dbc81035c169e5e 100644
--- a/llvm/test/CodeGen/RISCV/rvv/round-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/round-vp.ll
@@ -565,51 +565,46 @@ define <vscale x 32 x half> @vp_round_nxv32f16_unmasked(<vscale x 32 x half> %va
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v17, v16, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v16, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    lui a2, 307200
 ; ZVFHMIN-NEXT:    fmv.w.x fa5, a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v17, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
 ; ZVFHMIN-NEXT:    fsrmi a2, 4
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    fsrm a2
 ; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
 ; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v20, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v24
 ; ZVFHMIN-NEXT:    bltu a0, a1, .LBB11_2
 ; ZVFHMIN-NEXT:  # %bb.1:
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
 ; ZVFHMIN-NEXT:    addi a1, sp, 16
-; ZVFHMIN-NEXT:    vl8r.v v8, (a1) # Unknown-size Folded Reload
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
+; ZVFHMIN-NEXT:    vl8r.v v24, (a1) # Unknown-size Folded Reload
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v24, v16
+; ZVFHMIN-NEXT:    vmflt.vf v0, v24, fa5
 ; ZVFHMIN-NEXT:    fsrmi a0, 4
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfcvt.x.f.v v24, v16, v0.t
 ; ZVFHMIN-NEXT:    fsrm a0
-; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
+; ZVFHMIN-NEXT:    vfcvt.f.x.v v24, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v16, v24
-; ZVFHMIN-NEXT:    vmv8r.v v8, v16
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
 ; ZVFHMIN-NEXT:    slli a0, a0, 3
 ; ZVFHMIN-NEXT:    add sp, sp, a0
diff --git a/llvm/test/CodeGen/RISCV/rvv/roundeven-vp.ll b/llvm/test/CodeGen/RISCV/rvv/roundeven-vp.ll
index 6f6721c19b36f53..6c93ae68344cb6d 100644
--- a/llvm/test/CodeGen/RISCV/rvv/roundeven-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/roundeven-vp.ll
@@ -565,51 +565,46 @@ define <vscale x 32 x half> @vp_roundeven_nxv32f16_unmasked(<vscale x 32 x half>
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v17, v16, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v16, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    lui a2, 307200
 ; ZVFHMIN-NEXT:    fmv.w.x fa5, a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v17, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
 ; ZVFHMIN-NEXT:    fsrmi a2, 0
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    fsrm a2
 ; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
 ; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v20, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v24
 ; ZVFHMIN-NEXT:    bltu a0, a1, .LBB11_2
 ; ZVFHMIN-NEXT:  # %bb.1:
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
 ; ZVFHMIN-NEXT:    addi a1, sp, 16
-; ZVFHMIN-NEXT:    vl8r.v v8, (a1) # Unknown-size Folded Reload
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
+; ZVFHMIN-NEXT:    vl8r.v v24, (a1) # Unknown-size Folded Reload
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v24, v16
+; ZVFHMIN-NEXT:    vmflt.vf v0, v24, fa5
 ; ZVFHMIN-NEXT:    fsrmi a0, 0
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfcvt.x.f.v v24, v16, v0.t
 ; ZVFHMIN-NEXT:    fsrm a0
-; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
+; ZVFHMIN-NEXT:    vfcvt.f.x.v v24, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v16, v24
-; ZVFHMIN-NEXT:    vmv8r.v v8, v16
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
 ; ZVFHMIN-NEXT:    slli a0, a0, 3
 ; ZVFHMIN-NEXT:    add sp, sp, a0
diff --git a/llvm/test/CodeGen/RISCV/rvv/roundtozero-vp.ll b/llvm/test/CodeGen/RISCV/rvv/roundtozero-vp.ll
index 798c7e05bd47b2c..f35d70d6d470b37 100644
--- a/llvm/test/CodeGen/RISCV/rvv/roundtozero-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/roundtozero-vp.ll
@@ -565,51 +565,46 @@ define <vscale x 32 x half> @vp_roundtozero_nxv32f16_unmasked(<vscale x 32 x hal
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v17, v16, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v16, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    lui a2, 307200
 ; ZVFHMIN-NEXT:    fmv.w.x fa5, a2
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v17, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
 ; ZVFHMIN-NEXT:    fsrmi a2, 1
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v17
+; ZVFHMIN-NEXT:    vmv1r.v v0, v16
 ; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
 ; ZVFHMIN-NEXT:    fsrm a2
 ; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
 ; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v20, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v24
 ; ZVFHMIN-NEXT:    bltu a0, a1, .LBB11_2
 ; ZVFHMIN-NEXT:  # %bb.1:
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
 ; ZVFHMIN-NEXT:    addi a1, sp, 16
-; ZVFHMIN-NEXT:    vl8r.v v8, (a1) # Unknown-size Folded Reload
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
+; ZVFHMIN-NEXT:    vl8r.v v24, (a1) # Unknown-size Folded Reload
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfabs.v v8, v24, v0.t
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vmflt.vf v16, v8, fa5, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v24, v16
+; ZVFHMIN-NEXT:    vmflt.vf v0, v24, fa5
 ; ZVFHMIN-NEXT:    fsrmi a0, 1
-; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfcvt.x.f.v v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfcvt.x.f.v v24, v16, v0.t
 ; ZVFHMIN-NEXT:    fsrm a0
-; ZVFHMIN-NEXT:    vfcvt.f.x.v v8, v8, v0.t
+; ZVFHMIN-NEXT:    vfcvt.f.x.v v24, v24, v0.t
 ; ZVFHMIN-NEXT:    vsetvli zero, zero, e32, m8, ta, mu
-; ZVFHMIN-NEXT:    vfsgnj.vv v24, v8, v24, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v16, v24
-; ZVFHMIN-NEXT:    vmv8r.v v8, v16
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
 ; ZVFHMIN-NEXT:    slli a0, a0, 3
 ; ZVFHMIN-NEXT:    add sp, sp, a0
diff --git a/llvm/test/CodeGen/RISCV/rvv/vcopysign-vp.ll b/llvm/test/CodeGen/RISCV/rvv/vcopysign-vp.ll
index 58d172068f39422..f3574200054fd44 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vcopysign-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vcopysign-vp.ll
@@ -312,7 +312,7 @@ define <vscale x 32 x half> @vfsgnj_vv_nxv32f16_unmasked(<vscale x 32 x half> %v
 ; ZVFHMIN-NEXT:    sub sp, sp, a1
 ; ZVFHMIN-NEXT:    .cfi_escape 0x0f, 0x0d, 0x72, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0xa2, 0x38, 0x00, 0x1e, 0x22 # sp + 16 + 8 * vlenb
 ; ZVFHMIN-NEXT:    vsetvli a1, zero, e8, m4, ta, ma
-; ZVFHMIN-NEXT:    vmset.m v1
+; ZVFHMIN-NEXT:    vmset.m v24
 ; ZVFHMIN-NEXT:    csrr a2, vlenb
 ; ZVFHMIN-NEXT:    slli a1, a2, 1
 ; ZVFHMIN-NEXT:    sub a3, a0, a1
@@ -321,7 +321,7 @@ define <vscale x 32 x half> @vfsgnj_vv_nxv32f16_unmasked(<vscale x 32 x half> %v
 ; ZVFHMIN-NEXT:    and a3, a4, a3
 ; ZVFHMIN-NEXT:    srli a2, a2, 2
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
-; ZVFHMIN-NEXT:    vslidedown.vx v0, v1, a2
+; ZVFHMIN-NEXT:    vslidedown.vx v0, v24, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    addi a2, sp, 16
 ; ZVFHMIN-NEXT:    vs8r.v v16, (a2) # Unknown-size Folded Spill
@@ -340,8 +340,7 @@ define <vscale x 32 x half> @vfsgnj_vv_nxv32f16_unmasked(<vscale x 32 x half> %v
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v24
 ; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v1
-; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16, v0.t
+; ZVFHMIN-NEXT:    vfsgnj.vv v16, v24, v16
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    csrr a0, vlenb
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfabs-vp.ll b/llvm/test/CodeGen/RISCV/rvv/vfabs-vp.ll
index 6761e4801eb58a5..22f0d8bba2e629c 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfabs-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfabs-vp.ll
@@ -291,19 +291,18 @@ define <vscale x 32 x half> @vfabs_vv_nxv32f16_unmasked(<vscale x 32 x half> %va
 ; ZVFHMIN-NEXT:    vsetvli a4, zero, e8, mf2, ta, ma
 ; ZVFHMIN-NEXT:    vslidedown.vx v0, v16, a2
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v12
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v12
 ; ZVFHMIN-NEXT:    vsetvli zero, a3, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vfabs.v v24, v24, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v16, v16, v0.t
 ; ZVFHMIN-NEXT:    vsetvli a2, zero, e16, m4, ta, ma
-; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v24
+; ZVFHMIN-NEXT:    vfncvt.f.f.w v12, v16
 ; ZVFHMIN-NEXT:    bltu a0, a1, .LBB11_2
 ; ZVFHMIN-NEXT:  # %bb.1:
 ; ZVFHMIN-NEXT:    mv a0, a1
 ; ZVFHMIN-NEXT:  .LBB11_2:
-; ZVFHMIN-NEXT:    vfwcvt.f.f.v v24, v8
+; ZVFHMIN-NEXT:    vfwcvt.f.f.v v16, v8
 ; ZVFHMIN-NEXT:    vsetvli zero, a0, e32, m8, ta, ma
-; ZVFHMIN-NEXT:    vmv1r.v v0, v16
-; ZVFHMIN-NEXT:    vfabs.v v16, v24, v0.t
+; ZVFHMIN-NEXT:    vfabs.v v16, v16
 ; ZVFHMIN-NEXT:    vsetvli a0, zero, e16, m4, ta, ma
 ; ZVFHMIN-NEXT:    vfncvt.f.f.w v8, v16
 ; ZVFHMIN-NEXT:    ret
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfadd-vp.ll b/llvm/test/CodeGen/RISCV/rvv/vfadd-vp.ll
index 1ca70dbd2d2513e..00cb54c61a7a94c 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfadd-vp.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfadd-vp.ll
@@ -648,7 +648,7 @@ define <vscale x 32 x half> @vfadd_vv_nxv32f16_unmasked(<vscale x 32 x half> %va
 ; ZVFHMIN-NEXT:    sub sp, sp, a1
 ; ZVFHMIN-NEXT:    .cfi_escape 0x0f, 0x0d, 0x72, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0xa2, 0x38, 0x00, 0x1e, 0x22 # sp + 16 + 8 * vlenb
 ; ZVFHMIN-NEXT:    vsetvli a1, zero, e8, m4, ta, ma
-; ZVFHMIN-NEXT:    vmset.m v1
+; ZVFHMIN-NEXT:    vmset.m v24
 ; ZVFHMIN-NEXT:    csrr a2,...
[truncated]

preames
preames reviewed Sep 21, 2023
View reviewed changes
llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp Outdated
@@ -3188,6 +3188,12 @@ static bool usesAllOnesMask(SDValue MaskOp, SDValue GlueOp) {
// Check the instruction defining V0; it needs to be a VMSET pseudo.
SDValue MaskSetter = Glued->getOperand(2);

// Sometimes the VMSET is wrapped in a COPY_TO_REGCLASS node, e.g. if the mask
// was extracted from a larger register.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There's something off here. The docs and lowering code for COPY_TO_REGCLASS say that it is "like a copy", but fixes the destination register class. Your comment seems to say that it's more like a EXTRACT_SUBREG. I can't reconcile these two. Are you really sure the two register classes here are different sizes? Can you give an example of the before and after register class?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I think this should say "if the mask was extracted from a larger type". The register type is indeed the same (VRRegClassID), here's the DAG from the test insertelt_nxv32i8_imm:

t28: nxv8i8 = PseudoVSLIDEUP_VI_M1_MASK t41, t30, TargetConstant:i64<3>, Register:nxv8i1 $v0, TargetConstant:i64<4>, TargetConstant:i64<3>, TargetConstant:i64<0>, t37:1
  t37: ch,glue = CopyToReg t0, Register:nxv8i1 $v0, t40
    t40: nxv8i1 = COPY_TO_REGCLASS t17, TargetConstant:i64<20>
      t17: nxv32i1 = PseudoVMSET_M_B32 TargetConstant:i64<-1>, TargetConstant:i64<0>

Which is selected to

bb.0 (%ir-block.0):
  %3:vr = PseudoVMSET_M_B32 -1, 0
  %4:vr = COPY %3:vr
  $v0 = COPY %4:vr
  early-clobber %6:vrnov0 = PseudoVSLIDEUP_VI_M1_MASK %2:vrnov0(tied-def 0), killed %5:vr, 3, $v0, 4, 3, 0

The COPY_TO_REGCLASS comes from here:

llvm-project/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

Lines 2045 to 2056 in c6eb9e2

// If we haven't set a SubRegIdx, then we must be going between
// equally-sized LMUL types (e.g. VR -> VR). This can be done as a copy.
if (SubRegIdx == RISCV::NoSubRegister) {
unsigned InRegClassID = RISCVTargetLowering::getRegClassIDForVecVT(InVT);
assert(RISCVTargetLowering::getRegClassIDForVecVT(SubVecContainerVT) ==
InRegClassID &&
"Unexpected subvector extraction");
SDValue RC = CurDAG->getTargetConstant(InRegClassID, DL, XLenVT);
SDNode *NewNode =
CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, DL, VT, V, RC);
ReplaceNode(Node, NewNode);
return;

Swapping it out for a plain COPY seems to have no effect, but we still need to peek through it in usesAllOnesMask

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, yeah, that fits more. We've got two types of different sizes, but they share the same register class. Ok, understood.

preames
preames approved these changes Sep 22, 2023
View reviewed changes
Copy link
Collaborator

@preames preames left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@lukel97 lukel97 merged commit 3510552 into llvm:main Sep 22, 2023
3 checks passed
@kstoimenov kstoimenov mentioned this pull request Sep 22, 2023
Closed
Guzhu-AMD pushed a commit to GPUOpen-Drivers/llvm-project that referenced this pull request Sep 28, 2023
Merged main:3510552df673 into amd-gfx:d975d0fd1040
957cf8f 
Local branch amd-gfx d975d0f Merged main:50d1500447d6 into amd-gfx:aa1d70641799
Remote branch main 3510552 [RISCV] Check for COPY_TO_REGCLASS in usesAllOnesMask (llvm#67037)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@preames preames preames approved these changes

@topperc topperc Awaiting requested review from topperc

@yetingk yetingk Awaiting requested review from yetingk

Assignees
No one assigned
Labels
backend:RISC-V
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@lukel97 @llvmbot @preames