Implement vmul for type ShortVector of size 128

Signed-off-by: midronij <jackie.midroni@ibm.com>
eclipse · Nov 26, 2021 · 5f7c29a · 5f7c29a
1 parent 08ac8e9
commit 5f7c29a
Show file tree

Hide file tree

Showing 5 changed files with 50 additions and 14 deletions.
diff --git a/compiler/p/codegen/OMRCodeGenerator.cpp b/compiler/p/codegen/OMRCodeGenerator.cpp
@@ -1779,6 +1779,8 @@ bool OMR::Power::CodeGenerator::getSupportsOpCodeForAutoSIMD(TR::ILOpCode opcode
          else
             return false;
       case TR::vmul:
+         if (dt == TR::Int16)
+            return true;
       case TR::vdiv:
       case TR::vneg:
          if (dt == TR::Int32 || dt == TR::Float || dt == TR::Double)

diff --git a/compiler/p/codegen/OMRInstOpCode.enum b/compiler/p/codegen/OMRInstOpCode.enum
@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2019, 2020 IBM Corp. and others
+ * Copyright (c) 2019, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -711,7 +711,7 @@
    vsubuws,          // vector subtract unsigned word saturate
 // vmhaddshs,        // Vector Multiply-High-Add Signed Hword Saturate
 // vmhraddshs,       // Vector Multiply-High-Round-Add Signed Hword Saturate
-// vmladduhm,        // Vector Multiply-Low-Add Unsigned Hword Modulo
+   vmladduhm,        // Vector Multiply-Low-Add Unsigned Hword Modulo
 // vmsummbm,         // Vector Multiply-Sum Mixed Byte Modulo
 // vmsumshm,         // Vector Multiply-Sum Signed Hword Modulo
 // vmsumshs,         // Vector Multiply-Sum Signed Hword Saturate

diff --git a/compiler/p/codegen/OMRInstOpCodeProperties.hpp b/compiler/p/codegen/OMRInstOpCodeProperties.hpp
@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2019, 2020 IBM Corp. and others
+ * Copyright (c) 2019, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -8353,17 +8353,17 @@
    /*                   PPCOpProp_SyncSideEffectFree, */
    /* }, */
 
-   /* { */
-   /* .mnemonic    =    OMR::InstOpCode::vmladduhm, */
-   /* .name        =    "vmladduhm", */
-   /* .description =    "Vector Multiply-Low-Add Unsigned Hword Modulo", */
-   /* .prefix      =    0x00000000, */
-   /* .opcode      =    0x10000022, */
-   /* .format      =    FORMAT_UNKNOWN, */
-   /* .minimumALS  =    OMR_PROCESSOR_PPC_P6, */
-   /* .properties  =    PPCOpProp_IsVMX | */
-   /*                   PPCOpProp_SyncSideEffectFree, */
-   /* }, */
+   {
+   /* .mnemonic    = */   OMR::InstOpCode::vmladduhm,
+   /* .name        = */   "vmladduhm",
+   /* .description =      "Vector Multiply-Low-Add Unsigned Hword Modulo", */
+   /* .prefix      = */   0x00000000,
+   /* .opcode      = */   0x10000022,
+   /* .format      = */   FORMAT_VRT_VRA_VRB_VRC,
+   /* .minimumALS  = */   OMR_PROCESSOR_PPC_P6,
+   /* .properties  = */   PPCOpProp_IsVMX | 
+                          PPCOpProp_SyncSideEffectFree,
+   },
 
    /* { */
    /* .mnemonic    =    OMR::InstOpCode::vmsummbm, */

diff --git a/compiler/p/codegen/OMRTreeEvaluator.cpp b/compiler/p/codegen/OMRTreeEvaluator.cpp
@@ -3466,11 +3466,44 @@ TR::Register *OMR::Power::TreeEvaluator::vmulEvaluator(TR::Node *node, TR::CodeG
        return TR::TreeEvaluator::vmulFloatHelper(node,cg);
      case TR::VectorDouble:
        return TR::TreeEvaluator::vmulDoubleHelper(node,cg);
+     case TR::VectorInt16:
+       return TR::TreeEvaluator::vmulInt16Helper(node,cg);
      default:
        TR_ASSERT(false, "unrecognized vector type %s\n", node->getDataType().toString()); return NULL;
      }
    }
 
+TR::Register *OMR::Power::TreeEvaluator::vmulInt16Helper(TR::Node *node, TR::CodeGenerator *cg)
+   {
+   TR::Node *firstChild;
+   TR::Node *secondChild;
+   TR::Register *lhsReg, *rhsReg;
+   TR::Register *productReg;
+   TR::Register *temp;
+
+   firstChild = node->getFirstChild();
+   secondChild = node->getSecondChild();
+   lhsReg = NULL;
+   rhsReg = NULL;
+
+   lhsReg = cg->evaluate(firstChild);
+   rhsReg = cg->evaluate(secondChild);
+
+   productReg = cg->allocateRegister(TR_VRF);
+   temp = cg->allocateRegister(TR_VRF);
+
+   node->setRegister(productReg);
+
+   generateTrg1ImmInstruction(cg, TR::InstOpCode::vspltish, node, temp, 0);
+   generateTrg1Src3Instruction(cg, TR::InstOpCode::vmladduhm, node, productReg, lhsReg, rhsReg, temp);
+
+   cg->stopUsingRegister(temp);
+   cg->decReferenceCount(firstChild);
+   cg->decReferenceCount(secondChild);
+
+   return productReg;
+   }
+
 TR::Register *OMR::Power::TreeEvaluator::vmulInt32Helper(TR::Node *node, TR::CodeGenerator *cg)
    {
    return TR::TreeEvaluator::inlineVectorBinaryOp(node, cg, TR::InstOpCode::vmuluwm);

diff --git a/compiler/p/codegen/OMRTreeEvaluator.hpp b/compiler/p/codegen/OMRTreeEvaluator.hpp
@@ -638,6 +638,7 @@ class OMR_EXTENSIBLE TreeEvaluator: public OMR::TreeEvaluator
    static TR::Register *vaddEvaluator(TR::Node *node, TR::CodeGenerator *cg);
    static TR::Register *vsubEvaluator(TR::Node *node, TR::CodeGenerator *cg);
    static TR::Register *vmulEvaluator(TR::Node *node, TR::CodeGenerator *cg);
+   static TR::Register *vmulInt16Helper(TR::Node *node, TR::CodeGenerator *cg);
    static TR::Register *vmulInt32Helper(TR::Node *node, TR::CodeGenerator *cg);
    static TR::Register *vmulFloatHelper(TR::Node *node, TR::CodeGenerator *cg);
    static TR::Register *vmulDoubleHelper(TR::Node *node, TR::CodeGenerator *cg);