[NVPTX] Use prmt.f4e to lower pointer alignment fshr idiom (#143407)

Author: AlexMaclean

llvm/lib/Target/NVPTX/NVPTXInstrInfo.td

@@ -1621,6 +1621,10 @@ let hasSideEffects = false in {
 
 }
 
+// PRMT folding patterns
+def : Pat<(fshr i32:$hi, i32:$lo, (shl i32:$amt, (i32 3))),
+          (PRMT_B32rrr $lo, $hi, $amt, PrmtF4E)>;
+
 
 // byte extraction + signed/unsigned extension to i32.
 def : Pat<(i32 (sext_inreg (bfe i32:$s, i32:$o, 8), i8)),

llvm/test/CodeGen/NVPTX/prmt.ll

@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc < %s -verify-machineinstrs | FileCheck %s
-; RUN: %if ptxas %{ llc < %s -verify-machineinstrs | %ptxas-verify %}
+; RUN: llc < %s -verify-machineinstrs -mcpu=sm_50 | FileCheck %s
+; RUN: %if ptxas %{ llc < %s -verify-machineinstrs -mcpu=sm_50 | %ptxas-verify %}
 
 target triple = "nvptx64-nvidia-cuda"
 
@@ -111,3 +111,20 @@ define i32 @test_prmt_rc16(i32 %lo, i32 %selector) {
   %val = call i32 @llvm.nvvm.prmt.rc16(i32 %lo, i32 %selector)
   ret i32 %val
 }
+
+define i32 @test_prmt_f4e_folding(i32 %lo, i32 %hi, i32 %ptr) {
+; CHECK-LABEL: test_prmt_f4e_folding(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b32 %r<5>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    ld.param.b32 %r1, [test_prmt_f4e_folding_param_0];
+; CHECK-NEXT:    ld.param.b32 %r2, [test_prmt_f4e_folding_param_1];
+; CHECK-NEXT:    ld.param.b32 %r3, [test_prmt_f4e_folding_param_2];
+; CHECK-NEXT:    prmt.b32.f4e %r4, %r1, %r2, %r3;
+; CHECK-NEXT:    st.param.b32 [func_retval0], %r4;
+; CHECK-NEXT:    ret;
+  %sh_amt = shl i32 %ptr, 3
+  %val = call i32 @llvm.fshr.i32(i32 %hi, i32 %lo, i32 %sh_amt)
+  ret i32 %val
+}