[AArch64] The shift right/left and insert immediate builtins expect 3

source operands, a vector, an element to insert, and a shift amount.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194406 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsAArch64.td b/include/llvm/IR/IntrinsicsAArch64.td
index 0bff9876e4f2df7aeb25b4f7e08ac53d5f6d9067..4d54a2303d62affbc063249ebd33011eb7d00f36 100644 (file)
--- a/include/llvm/IR/IntrinsicsAArch64.td
+++ b/include/llvm/IR/IntrinsicsAArch64.td
@@ -255,10 +255,10 @@ def int_aarch64_neon_vqshlu_n : Neon_N2V_Intrinsic;
 def int_aarch64_neon_vqshlus_n : Neon_N2V_Intrinsic;
 
 // Shift Right And Insert (Immediate)
-def int_aarch64_neon_vsrid_n : Neon_2Arg_ShiftImm_Intrinsic;
+def int_aarch64_neon_vsrid_n : Neon_3Arg_ShiftImm_Intrinsic;
 
 // Shift Left And Insert (Immediate)
-def int_aarch64_neon_vslid_n : Neon_2Arg_ShiftImm_Intrinsic;
+def int_aarch64_neon_vslid_n : Neon_3Arg_ShiftImm_Intrinsic;
 
 // Scalar Signed Fixed-point Convert To Floating-Point (Immediate)
 def int_aarch64_neon_vcvtf32_n_s32 :

diff --git a/lib/Target/AArch64/AArch64InstrNEON.td b/lib/Target/AArch64/AArch64InstrNEON.td
index c53909edce64529aed7a293f038a981297537fb0..95e54f2fcb3126b3f149dbe03314542e15b073f9 100644 (file)
--- a/lib/Target/AArch64/AArch64InstrNEON.td
+++ b/lib/Target/AArch64/AArch64InstrNEON.td
@@ -4017,7 +4017,7 @@ multiclass NeonI_ScalarShiftLeftImm_BHSD_size<bit u, bits<5> opcode,
   }
 }
 
-class NeonI_ScalarShiftImm_accum_D_size<bit u, bits<5> opcode, string asmop>
+class NeonI_ScalarShiftRightImm_accum_D_size<bit u, bits<5> opcode, string asmop>
   : NeonI_ScalarShiftImm<u, opcode,
                          (outs FPR64:$Rd), (ins FPR64:$Src, FPR64:$Rn, shr_imm64:$Imm),
                          !strconcat(asmop, "\t$Rd, $Rn, $Imm"),
@@ -4028,6 +4028,17 @@ class NeonI_ScalarShiftImm_accum_D_size<bit u, bits<5> opcode, string asmop>
     let Constraints = "$Src = $Rd";
 }
 
+class NeonI_ScalarShiftLeftImm_accum_D_size<bit u, bits<5> opcode, string asmop>
+  : NeonI_ScalarShiftImm<u, opcode,
+                         (outs FPR64:$Rd), (ins FPR64:$Src, FPR64:$Rn, shl_imm64:$Imm),
+                         !strconcat(asmop, "\t$Rd, $Rn, $Imm"),
+                         [], NoItinerary> {
+    bits<6> Imm;
+    let Inst{22} = 0b1; // immh:immb = 1xxxxxx
+    let Inst{21-16} = Imm;
+    let Constraints = "$Src = $Rd";
+}
+
 class NeonI_ScalarShiftImm_narrow_size<bit u, bits<5> opcode, string asmop,
                                        RegisterClass FPRCD, RegisterClass FPRCS,
                                        Operand ImmTy>
@@ -4092,7 +4103,7 @@ multiclass Neon_ScalarShiftImm_BHSD_size_patterns<SDPatternOperator opnode,
 }
 
 class Neon_ScalarShiftImm_accum_D_size_patterns<SDPatternOperator opnode,
-                                          Instruction INSTD>
+                                                Instruction INSTD>
   : Pat<(v1i64 (opnode (v1i64 FPR64:$Src), (v1i64 FPR64:$Rn), (i32 imm:$Imm))),
         (INSTD FPR64:$Src, FPR64:$Rn, imm:$Imm)>;
 
@@ -4146,19 +4157,19 @@ defm URSHR : NeonI_ScalarShiftRightImm_D_size<0b1, 0b00100, "urshr">;
 defm : Neon_ScalarShiftImm_D_size_patterns<int_aarch64_neon_vrshrdu_n, URSHRddi>;
 
 // Scalar Signed Shift Right and Accumulate (Immediate)
-def SSRA : NeonI_ScalarShiftImm_accum_D_size<0b0, 0b00010, "ssra">;
+def SSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b0, 0b00010, "ssra">;
 def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vsrads_n, SSRA>;
 
 // Scalar Unsigned Shift Right and Accumulate (Immediate)
-def USRA : NeonI_ScalarShiftImm_accum_D_size<0b1, 0b00010, "usra">;
+def USRA : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b00010, "usra">;
 def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vsradu_n, USRA>;
 
 // Scalar Signed Rounding Shift Right and Accumulate (Immediate)
-def SRSRA : NeonI_ScalarShiftImm_accum_D_size<0b0, 0b00110, "srsra">;
+def SRSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b0, 0b00110, "srsra">;
 def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vrsrads_n, SRSRA>;
 
 // Scalar Unsigned Rounding Shift Right and Accumulate (Immediate)
-def URSRA : NeonI_ScalarShiftImm_accum_D_size<0b1, 0b00110, "ursra">;
+def URSRA : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b00110, "ursra">;
 def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vrsradu_n, URSRA>;
 
 // Scalar Shift Left (Immediate)
@@ -4184,12 +4195,12 @@ defm : Neon_ScalarShiftImm_BHSD_size_patterns<int_aarch64_neon_vqshlus_n,
                                               SQSHLUssi, SQSHLUddi>;
 
 // Shift Right And Insert (Immediate)
-defm SRI : NeonI_ScalarShiftRightImm_D_size<0b1, 0b01000, "sri">;
-defm : Neon_ScalarShiftImm_D_size_patterns<int_aarch64_neon_vsrid_n, SRIddi>;
+def SRI : NeonI_ScalarShiftRightImm_accum_D_size<0b1, 0b01000, "sri">;
+def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vsrid_n, SRI>;
 
 // Shift Left And Insert (Immediate)
-defm SLI : NeonI_ScalarShiftLeftImm_D_size<0b1, 0b01010, "sli">;
-defm : Neon_ScalarShiftImm_D_size_patterns<int_aarch64_neon_vslid_n, SLIddi>;
+def SLI : NeonI_ScalarShiftLeftImm_accum_D_size<0b1, 0b01010, "sli">;
+def : Neon_ScalarShiftImm_accum_D_size_patterns<int_aarch64_neon_vslid_n, SLI>;
 
 // Signed Saturating Shift Right Narrow (Immediate)
 defm SQSHRN : NeonI_ScalarShiftImm_narrow_HSD_size<0b0, 0b10010, "sqshrn">;

diff --git a/test/CodeGen/AArch64/neon-scalar-shift-imm.ll b/test/CodeGen/AArch64/neon-scalar-shift-imm.ll
index b11540f80ae6aff9f8a9edf4b828271de8d03b19..7e099a31a3b9e985d133d736f41eda7182124be3 100644 (file)
--- a/test/CodeGen/AArch64/neon-scalar-shift-imm.ll
+++ b/test/CodeGen/AArch64/neon-scalar-shift-imm.ll
@@ -266,47 +266,51 @@ entry:
 
 declare <1 x i64> @llvm.aarch64.neon.vqshlus.n.v1i64(<1 x i64>, i32)
 
-define i64 @test_vsrid_n_s64(i64 %a) {
+define i64 @test_vsrid_n_s64(i64 %a, i64 %b) {
 ; CHECK: test_vsrid_n_s64
 ; CHECK: sri {{d[0-9]+}}, {{d[0-9]+}}, #63
 entry:
   %vsri = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vsri1 = call <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64> %vsri, i32 63)
-  %0 = extractelement <1 x i64> %vsri1, i32 0
+  %vsri1 = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vsri2 = call <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64> %vsri, <1 x i64> %vsri1, i32 63)
+  %0 = extractelement <1 x i64> %vsri2, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64>, i32)
+declare <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64>, <1 x i64>, i32)
 
-define i64 @test_vsrid_n_u64(i64 %a) {
+define i64 @test_vsrid_n_u64(i64 %a, i64 %b) {
 ; CHECK: test_vsrid_n_u64
 ; CHECK: sri {{d[0-9]+}}, {{d[0-9]+}}, #63
 entry:
   %vsri = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vsri1 = call <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64> %vsri, i32 63)
-  %0 = extractelement <1 x i64> %vsri1, i32 0
+  %vsri1 = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vsri2 = call <1 x i64> @llvm.aarch64.neon.vsrid.n(<1 x i64> %vsri, <1 x i64> %vsri1, i32 63)
+  %0 = extractelement <1 x i64> %vsri2, i32 0
   ret i64 %0
 }
 
-define i64 @test_vslid_n_s64(i64 %a) {
+define i64 @test_vslid_n_s64(i64 %a, i64 %b) {
 ; CHECK: test_vslid_n_s64
 ; CHECK: sli {{d[0-9]+}}, {{d[0-9]+}}, #63
 entry:
   %vsli = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vsli1 = call <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64> %vsli, i32 63)
-  %0 = extractelement <1 x i64> %vsli1, i32 0
+  %vsli1 = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vsli2 = call <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64> %vsli, <1 x i64> %vsli1, i32 63)
+  %0 = extractelement <1 x i64> %vsli2, i32 0
   ret i64 %0
 }
 
-declare <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64>, i32)
+declare <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64>, <1 x i64>, i32)
 
-define i64 @test_vslid_n_u64(i64 %a) {
+define i64 @test_vslid_n_u64(i64 %a, i64 %b) {
 ; CHECK: test_vslid_n_u64
 ; CHECK: sli {{d[0-9]+}}, {{d[0-9]+}}, #63
 entry:
   %vsli = insertelement <1 x i64> undef, i64 %a, i32 0
-  %vsli1 = call <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64> %vsli, i32 63)
-  %0 = extractelement <1 x i64> %vsli1, i32 0
+  %vsli1 = insertelement <1 x i64> undef, i64 %b, i32 0
+  %vsli2 = call <1 x i64> @llvm.aarch64.neon.vslid.n(<1 x i64> %vsli, <1 x i64> %vsli1, i32 63)
+  %0 = extractelement <1 x i64> %vsli2, i32 0
   ret i64 %0
 }