[llvm] 658ff0b - [RISCV][VLOPT] Add support for integer widening multiply instructions (#112204)

via llvm-commits llvm-commits at lists.llvm.org
Wed Oct 16 06:37:33 PDT 2024


Author: Michael Maitland
Date: 2024-10-16T09:37:27-04:00
New Revision: 658ff0b84c9dd5f33f4d769ba7378cc2c64315a1

URL: https://github.com/llvm/llvm-project/commit/658ff0b84c9dd5f33f4d769ba7378cc2c64315a1
DIFF: https://github.com/llvm/llvm-project/commit/658ff0b84c9dd5f33f4d769ba7378cc2c64315a1.diff

LOG: [RISCV][VLOPT] Add support for integer widening multiply instructions (#112204)

This adds support for these instructions and also tests getOperandInfo
for these instructions as well.

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
    llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
index 53373b7a0f1575..6053899987db9b 100644
--- a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
@@ -563,7 +563,12 @@ static bool isSupportedInstr(const MachineInstr &MI) {
   case RISCV::VREM_VV:
   case RISCV::VREM_VX:
   // Vector Widening Integer Multiply Instructions
-  // FIXME: Add support
+  case RISCV::VWMUL_VV:
+  case RISCV::VWMUL_VX:
+  case RISCV::VWMULSU_VV:
+  case RISCV::VWMULSU_VX:
+  case RISCV::VWMULU_VV:
+  case RISCV::VWMULU_VX:
   // Vector Single-Width Integer Multiply-Add Instructions
   // FIXME: Add support
   // Vector Widening Integer Multiply-Add Instructions

diff  --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
index a360ae1998f77a..11f603b56b6e56 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
@@ -1122,6 +1122,132 @@ define <vscale x 4 x i32> @vrem_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
   ret <vscale x 4 x i32> %2
 }
 
+define <vscale x 4 x i64> @vwmul_vv(<vscale x 4 x i16> %a, <vscale x 4 x i16> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vwmul_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e16, m1, ta, ma
+; NOVLOPT-NEXT:    vwmul.vv v12, v8, v9
+; NOVLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmul.vv v8, v12, v12
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmul_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e16, m1, ta, ma
+; VLOPT-NEXT:    vwmul.vv v12, v8, v9
+; VLOPT-NEXT:    vsetvli zero, zero, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmul.vv v8, v12, v12
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i32> @llvm.riscv.vwmul.nxv4i64.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i16> %a, <vscale x 4 x i16> %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vwmul.nxv4i64.nxv4i32.nxv4i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %1, <vscale x 4 x i32> %1, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
+define <vscale x 4 x i64> @vwmul_vx(<vscale x 4 x i16> %a, i16 %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vwmul_vx:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a3, zero, e16, m1, ta, ma
+; NOVLOPT-NEXT:    vwmul.vx v12, v8, a0
+; NOVLOPT-NEXT:    vsetvli zero, a2, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmul.vx v8, v12, a1
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmul_vx:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a2, e16, m1, ta, ma
+; VLOPT-NEXT:    vwmul.vx v12, v8, a0
+; VLOPT-NEXT:    vsetvli zero, zero, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmul.vx v8, v12, a1
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i32> @llvm.riscv.vwmul.nxv4i32.nxv4i16.i16(<vscale x 4 x i32> poison, <vscale x 4 x i16> %a, i16 %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vwmul.nxv4i64.nxv4i64.i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %1, i32 %c, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
+define <vscale x 4 x i64> @vwmulsu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vwmulsu_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmulsu.vv v12, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e64, m4, ta, ma
+; NOVLOPT-NEXT:    vadd.vv v8, v12, v12
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmulsu_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmulsu.vv v12, v8, v10
+; VLOPT-NEXT:    vsetvli zero, zero, e64, m4, ta, ma
+; VLOPT-NEXT:    vadd.vv v8, v12, v12
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i64> @llvm.riscv.vwmulsu.nxv4i64.nxv4i32.nxv4i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vadd.nxv4i64.nxv4i64(<vscale x 4 x i64> poison, <vscale x 4 x i64> %1, <vscale x 4 x i64> %1, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
+define <vscale x 4 x i64> @vwmulsu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
+; NOVLOPT-LABEL: vwmulsu_vx:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmulsu.vx v12, v8, a0
+; NOVLOPT-NEXT:    vsetvli zero, a1, e64, m4, ta, ma
+; NOVLOPT-NEXT:    vadd.vv v8, v12, v12
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmulsu_vx:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a1, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmulsu.vx v12, v8, a0
+; VLOPT-NEXT:    vsetvli zero, zero, e64, m4, ta, ma
+; VLOPT-NEXT:    vadd.vv v8, v12, v12
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i64> @llvm.riscv.vwmulsu.nxv4i64.nxv4i32.i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vadd.nxv4i64.nxv4i64(<vscale x 4 x i64> poison, <vscale x 4 x i64> %1, <vscale x 4 x i64> %1, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
+define <vscale x 4 x i64> @vwmulu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vwmulu_vv:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmulu.vv v12, v8, v10
+; NOVLOPT-NEXT:    vsetvli zero, a0, e64, m4, ta, ma
+; NOVLOPT-NEXT:    vadd.vv v8, v12, v12
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmulu_vv:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a0, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmulu.vv v12, v8, v10
+; VLOPT-NEXT:    vsetvli zero, zero, e64, m4, ta, ma
+; VLOPT-NEXT:    vadd.vv v8, v12, v12
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i64> @llvm.riscv.vwmulu.nxv4i64.nxv4i32.nxv4i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vadd.nxv4i64.nxv4i64(<vscale x 4 x i64> poison, <vscale x 4 x i64> %1, <vscale x 4 x i64> %1, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
+define <vscale x 4 x i64> @vwmulu_vx(<vscale x 4 x i32> %a, i32 %b, iXLen %vl) {
+; NOVLOPT-LABEL: vwmulu_vx:
+; NOVLOPT:       # %bb.0:
+; NOVLOPT-NEXT:    vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT:    vwmulu.vx v12, v8, a0
+; NOVLOPT-NEXT:    vsetvli zero, a1, e64, m4, ta, ma
+; NOVLOPT-NEXT:    vadd.vv v8, v12, v12
+; NOVLOPT-NEXT:    ret
+;
+; VLOPT-LABEL: vwmulu_vx:
+; VLOPT:       # %bb.0:
+; VLOPT-NEXT:    vsetvli zero, a1, e32, m2, ta, ma
+; VLOPT-NEXT:    vwmulu.vx v12, v8, a0
+; VLOPT-NEXT:    vsetvli zero, zero, e64, m4, ta, ma
+; VLOPT-NEXT:    vadd.vv v8, v12, v12
+; VLOPT-NEXT:    ret
+  %1 = call <vscale x 4 x i64> @llvm.riscv.vwmulu.nxv4i64.nxv4i32.i32(<vscale x 4 x i64> poison, <vscale x 4 x i32> %a, i32 %b, iXLen -1)
+  %2 = call <vscale x 4 x i64> @llvm.riscv.vadd.nxv4i64.nxv4i64(<vscale x 4 x i64> poison, <vscale x 4 x i64> %1, <vscale x 4 x i64> %1, iXLen %vl)
+  ret <vscale x 4 x i64> %2
+}
+
 define <vscale x 4 x i32> @vwmacc_vx(<vscale x 4 x i16> %a, i16 %b, iXLen %vl) {
 ; NOVLOPT-LABEL: vwmacc_vx:
 ; NOVLOPT:       # %bb.0:


        


More information about the llvm-commits mailing list