[llvm] 09606d6 - [RISCV] Update the computeKnownBitsForTargetNode for RISCVISD::READ_VLENB to consider Zve/Zvl.

Craig Topper via llvm-commits llvm-commits at lists.llvm.org
Mon Jan 31 09:13:36 PST 2022


Author: Craig Topper
Date: 2022-01-31T09:13:14-08:00
New Revision: 09606d6a635b15ae08d596bedaa5bd88d7f8ea1a

URL: https://github.com/llvm/llvm-project/commit/09606d6a635b15ae08d596bedaa5bd88d7f8ea1a
DIFF: https://github.com/llvm/llvm-project/commit/09606d6a635b15ae08d596bedaa5bd88d7f8ea1a.diff

LOG: [RISCV] Update the computeKnownBitsForTargetNode for RISCVISD::READ_VLENB to consider Zve/Zvl.

We had previously hardcoded this to assume that vector registers
are 128 bits. This was true when only V existed, but after Zve
extensions were added this became incorrect.

This patch adjusts it to support 128, 64, or 32 bit vectors depending
on Zvl. The 128-bit limit is artificial, but we don't have any test
coverage showing that we larger values so I was being conservative.

None of our lit tests depend on this code today due to the custom
lowering of ISD::VSCALE that inserts the appropriate left or right
shift to convert from VLENB to VSCALE. That code was added after
this code in computeKnownBitsForTargetNode.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D118582

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVISelLowering.cpp
    llvm/lib/Target/RISCV/RISCVSubtarget.h

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 4225ae42ed39..de240583f949 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -8274,12 +8274,17 @@ void RISCVTargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
     }
     break;
   }
-  case RISCVISD::READ_VLENB:
-    // We assume VLENB is at least 16 bytes.
-    Known.Zero.setLowBits(4);
+  case RISCVISD::READ_VLENB: {
+    // If we know the minimum VLen from Zvl extensions, we can use that to
+    // determine the trailing zeros of VLENB.
+    // FIXME: Limit to 128 bit vectors until we have more testing.
+    unsigned MinVLenB = std::min(128U, Subtarget.getMinVLen()) / 8;
+    if (MinVLenB > 0)
+      Known.Zero.setLowBits(Log2_32(MinVLenB));
     // We assume VLENB is no more than 65536 / 8 bytes.
     Known.Zero.setBitsFrom(14);
     break;
+  }
   case ISD::INTRINSIC_W_CHAIN:
   case ISD::INTRINSIC_WO_CHAIN: {
     unsigned IntNo =

diff  --git a/llvm/lib/Target/RISCV/RISCVSubtarget.h b/llvm/lib/Target/RISCV/RISCVSubtarget.h
index 044dda0a1ccc..34c6e8e684ac 100644
--- a/llvm/lib/Target/RISCV/RISCVSubtarget.h
+++ b/llvm/lib/Target/RISCV/RISCVSubtarget.h
@@ -195,6 +195,7 @@ class RISCVSubtarget : public RISCVGenSubtargetInfo {
 
     return 0;
   }
+  unsigned getMinVLen() const { return ZvlLen; }
   RISCVABI::ABI getTargetABI() const { return TargetABI; }
   bool isRegisterReservedByUser(Register i) const {
     assert(i < RISCV::NUM_TARGET_REGS && "Register out of range");


        


More information about the llvm-commits mailing list