PTX: Set PTX 2.0 as the minimum supported version

- Remove PTX 1.4 code generation
- Change type of intrinsics to .v4.i32 instead of .v4.i16
- Add and/or/xor integer instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127677 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IntrinsicsPTX.td b/include/llvm/IntrinsicsPTX.td
index cbcd56e5f2ce6018b0df93f6319a9cb47c66a821..01241fe4d48e52b4111b6a9ca59a7c04205da595 100644 (file)
--- a/include/llvm/IntrinsicsPTX.td
+++ b/include/llvm/IntrinsicsPTX.td
@@ -12,14 +12,17 @@
 //===----------------------------------------------------------------------===//
 
 let TargetPrefix = "ptx" in {
-  // FIXME Since PTX 2.0, special registers are redefined as v4i32 type
-  multiclass PTXReadSpecialRegisterIntrinsic_v4i16 {
-    def _r64   : Intrinsic<[llvm_i64_ty],   [], [IntrNoMem]>;
-    def _v4i16 : Intrinsic<[llvm_v4i16_ty], [], [IntrNoMem]>;
-    def _x     : Intrinsic<[llvm_i16_ty], [], [IntrNoMem]>;
-    def _y     : Intrinsic<[llvm_i16_ty], [], [IntrNoMem]>;
-    def _z     : Intrinsic<[llvm_i16_ty], [], [IntrNoMem]>;
-    def _w     : Intrinsic<[llvm_i16_ty], [], [IntrNoMem]>;
+  multiclass PTXReadSpecialRegisterIntrinsic_v4i32 {
+// FIXME: Do we need the 128-bit integer type version?
+//    def _r64   : Intrinsic<[llvm_i128_ty],   [], [IntrNoMem]>;
+
+// FIXME: Enable this once v4i32 support is enabled in back-end.
+//    def _v4i16 : Intrinsic<[llvm_v4i32_ty], [], [IntrNoMem]>;
+
+    def _x     : Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
+    def _y     : Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
+    def _z     : Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
+    def _w     : Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
   }
 
   class PTXReadSpecialRegisterIntrinsic_r32
@@ -29,15 +32,15 @@ let TargetPrefix = "ptx" in {
     : Intrinsic<[llvm_i64_ty], [], [IntrNoMem]>;
 }
 
-defm int_ptx_read_tid        : PTXReadSpecialRegisterIntrinsic_v4i16;
-defm int_ptx_read_ntid       : PTXReadSpecialRegisterIntrinsic_v4i16;
+defm int_ptx_read_tid        : PTXReadSpecialRegisterIntrinsic_v4i32;
+defm int_ptx_read_ntid       : PTXReadSpecialRegisterIntrinsic_v4i32;
 
 def int_ptx_read_laneid      : PTXReadSpecialRegisterIntrinsic_r32;
 def int_ptx_read_warpid      : PTXReadSpecialRegisterIntrinsic_r32;
 def int_ptx_read_nwarpid     : PTXReadSpecialRegisterIntrinsic_r32;
 
-defm int_ptx_read_ctaid      : PTXReadSpecialRegisterIntrinsic_v4i16;
-defm int_ptx_read_nctaid     : PTXReadSpecialRegisterIntrinsic_v4i16;
+defm int_ptx_read_ctaid      : PTXReadSpecialRegisterIntrinsic_v4i32;
+defm int_ptx_read_nctaid     : PTXReadSpecialRegisterIntrinsic_v4i32;
 
 def int_ptx_read_smid        : PTXReadSpecialRegisterIntrinsic_r32;
 def int_ptx_read_nsmid       : PTXReadSpecialRegisterIntrinsic_r32;

diff --git a/lib/Target/PTX/PTX.td b/lib/Target/PTX/PTX.td
index 12febcb13decf79793082bd97148128129aa6831..dbc6f579a2983a7c60972a05c374be33fa6ec213 100644 (file)
--- a/lib/Target/PTX/PTX.td
+++ b/lib/Target/PTX/PTX.td
@@ -29,17 +29,18 @@ def Feature64Bit : SubtargetFeature<"64bit", "Use64BitAddresses", "true",
 
 //===- PTX Version --------------------------------------------------------===//
 
-def FeaturePTX14 : SubtargetFeature<"ptx14", "PTXVersion", "PTX_VERSION_1_4",
-                                    "Use PTX Language Version 1.4">;
-
 def FeaturePTX20 : SubtargetFeature<"ptx20", "PTXVersion", "PTX_VERSION_2_0",
                                     "Use PTX Language Version 2.0",
-                                    [FeaturePTX14]>;
+                                    []>;
 
 def FeaturePTX21 : SubtargetFeature<"ptx21", "PTXVersion", "PTX_VERSION_2_1",
                                     "Use PTX Language Version 2.1",
                                     [FeaturePTX20]>;
 
+def FeaturePTX22 : SubtargetFeature<"ptx22", "PTXVersion", "PTX_VERSION_2_2",
+                                    "Use PTX Language Version 2.2",
+                                    [FeaturePTX21]>;
+
 //===- PTX Shader Model ---------------------------------------------------===//
 
 def FeatureSM10 : SubtargetFeature<"sm10", "PTXShaderModel", "PTX_SM_1_0",

diff --git a/lib/Target/PTX/PTXInstrInfo.td b/lib/Target/PTX/PTXInstrInfo.td
index 5c198ca4826c6360e10f17afc7cc02b8412ffb48..dc96914ba6ef18ebcf4b7604f41e54b236aabc46 100644 (file)
--- a/lib/Target/PTX/PTXInstrInfo.td
+++ b/lib/Target/PTX/PTXInstrInfo.td
@@ -32,10 +32,11 @@ def SupportsSM20       : Predicate<"getSubtarget().supportsSM20()">;
 def DoesNotSupportSM20 : Predicate<"!getSubtarget().supportsSM20()">;
 
 // PTX Version Support
-def SupportsPTX20       : Predicate<"getSubtarget().supportsPTX20()">;
-def DoesNotSupportPTX20 : Predicate<"!getSubtarget().supportsPTX20()">;
 def SupportsPTX21       : Predicate<"getSubtarget().supportsPTX21()">;
 def DoesNotSupportPTX21 : Predicate<"!getSubtarget().supportsPTX21()">;
+def SupportsPTX22       : Predicate<"getSubtarget().supportsPTX22()">;
+def DoesNotSupportPTX22 : Predicate<"!getSubtarget().supportsPTX22()">;
+
 
 //===----------------------------------------------------------------------===//
 // Instruction Pattern Stuff
@@ -253,6 +254,33 @@ multiclass INT3<string opcstr, SDNode opnode> {
                      [(set RRegu64:$d, (opnode RRegu64:$a, imm:$b))]>;
 }
 
+multiclass PTX_LOGIC<string opcstr, SDNode opnode> {
+  def rr16 : InstPTX<(outs RRegu16:$d),
+                     (ins RRegu16:$a, RRegu16:$b),
+                     !strconcat(opcstr, ".b16\t$d, $a, $b"),
+                     [(set RRegu16:$d, (opnode RRegu16:$a, RRegu16:$b))]>;
+  def ri16 : InstPTX<(outs RRegu16:$d),
+                     (ins RRegu16:$a, i16imm:$b),
+                     !strconcat(opcstr, ".b16\t$d, $a, $b"),
+                     [(set RRegu16:$d, (opnode RRegu16:$a, imm:$b))]>;
+  def rr32 : InstPTX<(outs RRegu32:$d),
+                     (ins RRegu32:$a, RRegu32:$b),
+                     !strconcat(opcstr, ".b32\t$d, $a, $b"),
+                     [(set RRegu32:$d, (opnode RRegu32:$a, RRegu32:$b))]>;
+  def ri32 : InstPTX<(outs RRegu32:$d),
+                     (ins RRegu32:$a, i32imm:$b),
+                     !strconcat(opcstr, ".b32\t$d, $a, $b"),
+                     [(set RRegu32:$d, (opnode RRegu32:$a, imm:$b))]>;
+  def rr64 : InstPTX<(outs RRegu64:$d),
+                     (ins RRegu64:$a, RRegu64:$b),
+                     !strconcat(opcstr, ".b64\t$d, $a, $b"),
+                     [(set RRegu64:$d, (opnode RRegu64:$a, RRegu64:$b))]>;
+  def ri64 : InstPTX<(outs RRegu64:$d),
+                     (ins RRegu64:$a, i64imm:$b),
+                     !strconcat(opcstr, ".b64\t$d, $a, $b"),
+                     [(set RRegu64:$d, (opnode RRegu64:$a, imm:$b))]>;
+}
+
 // no %type directive, non-communtable
 multiclass INT3ntnc<string opcstr, SDNode opnode> {
   def rr : InstPTX<(outs RRegu32:$d),
@@ -359,6 +387,7 @@ multiclass PTX_ST_ALL<string opstr, PatFrag pat_store> {
 
 defm ADD : INT3<"add", add>;
 defm SUB : INT3<"sub", sub>;
+defm MUL : INT3<"mul.lo", mul>; // FIXME: Allow 32x32 -> 64 multiplies
 
 ///===- Floating-Point Arithmetic Instructions ----------------------------===//
 
@@ -462,6 +491,10 @@ defm SHL : INT3ntnc<"shl.b32", PTXshl>;
 defm SRL : INT3ntnc<"shr.u32", PTXsrl>;
 defm SRA : INT3ntnc<"shr.s32", PTXsra>;
 
+defm AND : PTX_LOGIC<"and", and>;
+defm OR  : PTX_LOGIC<"or",  or>;
+defm XOR : PTX_LOGIC<"xor", xor>;
+
 ///===- Data Movement and Conversion Instructions -------------------------===//
 
 let neverHasSideEffects = 1 in {

diff --git a/lib/Target/PTX/PTXIntrinsicInstrInfo.td b/lib/Target/PTX/PTXIntrinsicInstrInfo.td
index 2f2578cb5c95a868ca87ce43d5f7bb7e9eaf9b99..320934a2228cb5241f40b8df3c2f89216641ef72 100644 (file)
--- a/lib/Target/PTX/PTXIntrinsicInstrInfo.td
+++ b/lib/Target/PTX/PTXIntrinsicInstrInfo.td
@@ -23,40 +23,35 @@ class PTX_READ_SPECIAL_REGISTER_R32<string regname, Intrinsic intop>
             !strconcat("mov.u32\t$d, %", regname),
             [(set RRegu32:$d, (intop))]>;
 
-class PTX_READ_SPECIAL_SUB_REGISTER<string regname, Intrinsic intop>
-  : InstPTX<(outs RRegu16:$d), (ins),
-            !strconcat("mov.u16\t$d, %", regname),
-            [(set RRegu16:$d, (intop))]>;
-
 // TODO Add read vector-version of special registers
 
-def PTX_READ_TID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"tid", int_ptx_read_tid_r64>;
-def PTX_READ_TID_X   : PTX_READ_SPECIAL_SUB_REGISTER<"tid.x", int_ptx_read_tid_x>;
-def PTX_READ_TID_Y   : PTX_READ_SPECIAL_SUB_REGISTER<"tid.y", int_ptx_read_tid_y>;
-def PTX_READ_TID_Z   : PTX_READ_SPECIAL_SUB_REGISTER<"tid.z", int_ptx_read_tid_z>;
-def PTX_READ_TID_W   : PTX_READ_SPECIAL_SUB_REGISTER<"tid.w", int_ptx_read_tid_w>;
+//def PTX_READ_TID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"tid", int_ptx_read_tid_r64>;
+def PTX_READ_TID_X   : PTX_READ_SPECIAL_REGISTER_R32<"tid.x", int_ptx_read_tid_x>;
+def PTX_READ_TID_Y   : PTX_READ_SPECIAL_REGISTER_R32<"tid.y", int_ptx_read_tid_y>;
+def PTX_READ_TID_Z   : PTX_READ_SPECIAL_REGISTER_R32<"tid.z", int_ptx_read_tid_z>;
+def PTX_READ_TID_W   : PTX_READ_SPECIAL_REGISTER_R32<"tid.w", int_ptx_read_tid_w>;
 
-def PTX_READ_NTID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"ntid", int_ptx_read_ntid_r64>;
-def PTX_READ_NTID_X   : PTX_READ_SPECIAL_SUB_REGISTER<"ntid.x", int_ptx_read_ntid_x>;
-def PTX_READ_NTID_Y   : PTX_READ_SPECIAL_SUB_REGISTER<"ntid.y", int_ptx_read_ntid_y>;
-def PTX_READ_NTID_Z   : PTX_READ_SPECIAL_SUB_REGISTER<"ntid.z", int_ptx_read_ntid_z>;
-def PTX_READ_NTID_W   : PTX_READ_SPECIAL_SUB_REGISTER<"ntid.w", int_ptx_read_ntid_w>;
+//def PTX_READ_NTID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"ntid", int_ptx_read_ntid_r64>;
+def PTX_READ_NTID_X   : PTX_READ_SPECIAL_REGISTER_R32<"ntid.x", int_ptx_read_ntid_x>;
+def PTX_READ_NTID_Y   : PTX_READ_SPECIAL_REGISTER_R32<"ntid.y", int_ptx_read_ntid_y>;
+def PTX_READ_NTID_Z   : PTX_READ_SPECIAL_REGISTER_R32<"ntid.z", int_ptx_read_ntid_z>;
+def PTX_READ_NTID_W   : PTX_READ_SPECIAL_REGISTER_R32<"ntid.w", int_ptx_read_ntid_w>;
 
 def PTX_READ_LANEID  : PTX_READ_SPECIAL_REGISTER_R32<"laneid", int_ptx_read_laneid>;
 def PTX_READ_WARPID  : PTX_READ_SPECIAL_REGISTER_R32<"warpid", int_ptx_read_warpid>;
 def PTX_READ_NWARPID : PTX_READ_SPECIAL_REGISTER_R32<"nwarpid", int_ptx_read_nwarpid>;
 
-def PTX_READ_CTAID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"ctaid", int_ptx_read_ctaid_r64>;
-def PTX_READ_CTAID_X   : PTX_READ_SPECIAL_SUB_REGISTER<"ctaid.x", int_ptx_read_ctaid_x>;
-def PTX_READ_CTAID_Y   : PTX_READ_SPECIAL_SUB_REGISTER<"ctaid.y", int_ptx_read_ctaid_y>;
-def PTX_READ_CTAID_Z   : PTX_READ_SPECIAL_SUB_REGISTER<"ctaid.z", int_ptx_read_ctaid_z>;
-def PTX_READ_CTAID_W   : PTX_READ_SPECIAL_SUB_REGISTER<"ctaid.w", int_ptx_read_ctaid_w>;
-
-def PTX_READ_NCTAID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"nctaid", int_ptx_read_nctaid_r64>;
-def PTX_READ_NCTAID_X   : PTX_READ_SPECIAL_SUB_REGISTER<"nctaid.x", int_ptx_read_nctaid_x>;
-def PTX_READ_NCTAID_Y   : PTX_READ_SPECIAL_SUB_REGISTER<"nctaid.y", int_ptx_read_nctaid_y>;
-def PTX_READ_NCTAID_Z   : PTX_READ_SPECIAL_SUB_REGISTER<"nctaid.z", int_ptx_read_nctaid_z>;
-def PTX_READ_NCTAID_W   : PTX_READ_SPECIAL_SUB_REGISTER<"nctaid.w", int_ptx_read_nctaid_w>;
+//def PTX_READ_CTAID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"ctaid", int_ptx_read_ctaid_r64>;
+def PTX_READ_CTAID_X   : PTX_READ_SPECIAL_REGISTER_R32<"ctaid.x", int_ptx_read_ctaid_x>;
+def PTX_READ_CTAID_Y   : PTX_READ_SPECIAL_REGISTER_R32<"ctaid.y", int_ptx_read_ctaid_y>;
+def PTX_READ_CTAID_Z   : PTX_READ_SPECIAL_REGISTER_R32<"ctaid.z", int_ptx_read_ctaid_z>;
+def PTX_READ_CTAID_W   : PTX_READ_SPECIAL_REGISTER_R32<"ctaid.w", int_ptx_read_ctaid_w>;
+
+//def PTX_READ_NCTAID_R64 : PTX_READ_SPECIAL_REGISTER_R64<"nctaid", int_ptx_read_nctaid_r64>;
+def PTX_READ_NCTAID_X   : PTX_READ_SPECIAL_REGISTER_R32<"nctaid.x", int_ptx_read_nctaid_x>;
+def PTX_READ_NCTAID_Y   : PTX_READ_SPECIAL_REGISTER_R32<"nctaid.y", int_ptx_read_nctaid_y>;
+def PTX_READ_NCTAID_Z   : PTX_READ_SPECIAL_REGISTER_R32<"nctaid.z", int_ptx_read_nctaid_z>;
+def PTX_READ_NCTAID_W   : PTX_READ_SPECIAL_REGISTER_R32<"nctaid.w", int_ptx_read_nctaid_w>;
 
 def PTX_READ_SMID  : PTX_READ_SPECIAL_REGISTER_R32<"smid", int_ptx_read_smid>;
 def PTX_READ_NSMID  : PTX_READ_SPECIAL_REGISTER_R32<"nsmid", int_ptx_read_nsmid>;

diff --git a/lib/Target/PTX/PTXSubtarget.cpp b/lib/Target/PTX/PTXSubtarget.cpp
index ef4060d6f015b75bec6da0277b45cfa4b61ce465..527622d0c78821cda46ac233be1a5257874aa74f 100644 (file)
--- a/lib/Target/PTX/PTXSubtarget.cpp
+++ b/lib/Target/PTX/PTXSubtarget.cpp
@@ -18,7 +18,7 @@ using namespace llvm;
 
 PTXSubtarget::PTXSubtarget(const std::string &TT, const std::string &FS)
   : PTXShaderModel(PTX_SM_1_0),
-    PTXVersion(PTX_VERSION_1_4),
+    PTXVersion(PTX_VERSION_2_0),
     SupportsDouble(false),
     Use64BitAddresses(false) {
   std::string TARGET = "generic";
@@ -37,9 +37,9 @@ std::string PTXSubtarget::getTargetString() const {
 std::string PTXSubtarget::getPTXVersionString() const {
   switch(PTXVersion) {
     default: llvm_unreachable("Unknown PTX version");
-    case PTX_VERSION_1_4: return "1.4";
     case PTX_VERSION_2_0: return "2.0";
     case PTX_VERSION_2_1: return "2.1";
+    case PTX_VERSION_2_2: return "2.2";
   }
 }
 

diff --git a/lib/Target/PTX/PTXSubtarget.h b/lib/Target/PTX/PTXSubtarget.h
index 19a870d4ce8610d054a479a52d2e5a39f361e99b..57cd43da4763190b4943f1a5fc7b538d16ba8af3 100644 (file)
--- a/lib/Target/PTX/PTXSubtarget.h
+++ b/lib/Target/PTX/PTXSubtarget.h
@@ -19,16 +19,25 @@
 namespace llvm {
   class PTXSubtarget : public TargetSubtarget {
     private:
+
+      /**
+       * Enumeration of Shader Models supported by the back-end.
+       */
       enum PTXShaderModelEnum {
-        PTX_SM_1_0,
-        PTX_SM_1_3,
-        PTX_SM_2_0
+        PTX_SM_1_0, /*< Shader Model 1.0 */
+        PTX_SM_1_3, /*< Shader Model 1.3 */
+        PTX_SM_2_0  /*< Shader Model 2.0 */
       };
 
+      /**
+       * Enumeration of PTX versions supported by the back-end.
+       *
+       * Currently, PTX 2.0 is the minimum supported version.
+       */
       enum PTXVersionEnum {
-        PTX_VERSION_1_4,
-        PTX_VERSION_2_0,
-        PTX_VERSION_2_1
+        PTX_VERSION_2_0,  /*< PTX Version 2.0 */
+        PTX_VERSION_2_1,  /*< PTX Version 2.1 */
+        PTX_VERSION_2_2   /*< PTX Version 2.2 */
       };
 
       /// Shader Model supported on the target GPU.
@@ -58,10 +67,10 @@ namespace llvm {
 
       bool supportsSM20() const { return PTXShaderModel >= PTX_SM_2_0; }
 
-      bool supportsPTX20() const { return PTXVersion >= PTX_VERSION_2_0; }
-
       bool supportsPTX21() const { return PTXVersion >= PTX_VERSION_2_1; }
 
+      bool supportsPTX22() const { return PTXVersion >= PTX_VERSION_2_2; }
+
       std::string ParseSubtargetFeatures(const std::string &FS,
                                          const std::string &CPU);
   }; // class PTXSubtarget

diff --git a/test/CodeGen/PTX/intrinsic.ll b/test/CodeGen/PTX/intrinsic.ll
index 139e29ee05a49e72afc557418a6b39f15ae70eb8..7405dd6f5e5a237ba316208e43aae9ebbbc3c4d9 100644 (file)
--- a/test/CodeGen/PTX/intrinsic.ll
+++ b/test/CodeGen/PTX/intrinsic.ll
@@ -1,59 +1,59 @@
 ; RUN: llc < %s -march=ptx -mattr=+ptx20,+sm20 | FileCheck %s
 
-define ptx_device i16 @test_tid_x() {
-; CHECK: mov.u16 rh0, %tid.x;
+define ptx_device i32 @test_tid_x() {
+; CHECK: mov.u32 r0, %tid.x;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.tid.x()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.tid.x()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_tid_y() {
-; CHECK: mov.u16 rh0, %tid.y;
+define ptx_device i32 @test_tid_y() {
+; CHECK: mov.u32 r0, %tid.y;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.tid.y()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.tid.y()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_tid_z() {
-; CHECK: mov.u16 rh0, %tid.z;
+define ptx_device i32 @test_tid_z() {
+; CHECK: mov.u32 r0, %tid.z;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.tid.z()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.tid.z()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_tid_w() {
-; CHECK: mov.u16 rh0, %tid.w;
+define ptx_device i32 @test_tid_w() {
+; CHECK: mov.u32 r0, %tid.w;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.tid.w()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.tid.w()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ntid_x() {
-; CHECK: mov.u16 rh0, %ntid.x;
+define ptx_device i32 @test_ntid_x() {
+; CHECK: mov.u32 r0, %ntid.x;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ntid.x()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ntid.x()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ntid_y() {
-; CHECK: mov.u16 rh0, %ntid.y;
+define ptx_device i32 @test_ntid_y() {
+; CHECK: mov.u32 r0, %ntid.y;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ntid.y()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ntid.y()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ntid_z() {
-; CHECK: mov.u16 rh0, %ntid.z;
+define ptx_device i32 @test_ntid_z() {
+; CHECK: mov.u32 r0, %ntid.z;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ntid.z()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ntid.z()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ntid_w() {
-; CHECK: mov.u16 rh0, %ntid.w;
+define ptx_device i32 @test_ntid_w() {
+; CHECK: mov.u32 r0, %ntid.w;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ntid.w()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ntid.w()
+       ret i32 %x
 }
 
 define ptx_device i32 @test_laneid() {
@@ -77,60 +77,60 @@ define ptx_device i32 @test_nwarpid() {
        ret i32 %x
 }
 
-define ptx_device i16 @test_ctaid_x() {
-; CHECK: mov.u16 rh0, %ctaid.x;
+define ptx_device i32 @test_ctaid_x() {
+; CHECK: mov.u32 r0, %ctaid.x;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ctaid.x()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ctaid.x()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ctaid_y() {
-; CHECK: mov.u16 rh0, %ctaid.y;
+define ptx_device i32 @test_ctaid_y() {
+; CHECK: mov.u32 r0, %ctaid.y;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ctaid.y()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ctaid.y()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ctaid_z() {
-; CHECK: mov.u16 rh0, %ctaid.z;
+define ptx_device i32 @test_ctaid_z() {
+; CHECK: mov.u32 r0, %ctaid.z;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ctaid.z()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ctaid.z()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_ctaid_w() {
-; CHECK: mov.u16 rh0, %ctaid.w;
+define ptx_device i32 @test_ctaid_w() {
+; CHECK: mov.u32 r0, %ctaid.w;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.ctaid.w()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.ctaid.w()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_nctaid_x() {
-; CHECK: mov.u16 rh0, %nctaid.x;
+define ptx_device i32 @test_nctaid_x() {
+; CHECK: mov.u32 r0, %nctaid.x;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.nctaid.x()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.nctaid.x()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_nctaid_y() {
-; CHECK: mov.u16 rh0, %nctaid.y;
+define ptx_device i32 @test_nctaid_y() {
+; CHECK: mov.u32 r0, %nctaid.y;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.nctaid.y()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.nctaid.y()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_nctaid_z() {
-; CHECK: mov.u16 rh0, %nctaid.z;
+define ptx_device i32 @test_nctaid_z() {
+; CHECK: mov.u32 r0, %nctaid.z;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.nctaid.z()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.nctaid.z()
+       ret i32 %x
 }
 
-define ptx_device i16 @test_nctaid_w() {
-; CHECK: mov.u16 rh0, %nctaid.w;
+define ptx_device i32 @test_nctaid_w() {
+; CHECK: mov.u32 r0, %nctaid.w;
 ; CHECK-NEXT: ret;
-       %x = call i16 @llvm.ptx.read.nctaid.w()
-       ret i16 %x
+       %x = call i32 @llvm.ptx.read.nctaid.w()
+       ret i32 %x
 }
 
 define ptx_device i32 @test_smid() {
@@ -238,27 +238,27 @@ define ptx_device void @test_bar_sync() {
        ret void
 }
 
-declare i16 @llvm.ptx.read.tid.x()
-declare i16 @llvm.ptx.read.tid.y()
-declare i16 @llvm.ptx.read.tid.z()
-declare i16 @llvm.ptx.read.tid.w()
-declare i16 @llvm.ptx.read.ntid.x()
-declare i16 @llvm.ptx.read.ntid.y()
-declare i16 @llvm.ptx.read.ntid.z()
-declare i16 @llvm.ptx.read.ntid.w()
+declare i32 @llvm.ptx.read.tid.x()
+declare i32 @llvm.ptx.read.tid.y()
+declare i32 @llvm.ptx.read.tid.z()
+declare i32 @llvm.ptx.read.tid.w()
+declare i32 @llvm.ptx.read.ntid.x()
+declare i32 @llvm.ptx.read.ntid.y()
+declare i32 @llvm.ptx.read.ntid.z()
+declare i32 @llvm.ptx.read.ntid.w()
 
 declare i32 @llvm.ptx.read.laneid()
 declare i32 @llvm.ptx.read.warpid()
 declare i32 @llvm.ptx.read.nwarpid()
 
-declare i16 @llvm.ptx.read.ctaid.x()
-declare i16 @llvm.ptx.read.ctaid.y()
-declare i16 @llvm.ptx.read.ctaid.z()
-declare i16 @llvm.ptx.read.ctaid.w()
-declare i16 @llvm.ptx.read.nctaid.x()
-declare i16 @llvm.ptx.read.nctaid.y()
-declare i16 @llvm.ptx.read.nctaid.z()
-declare i16 @llvm.ptx.read.nctaid.w()
+declare i32 @llvm.ptx.read.ctaid.x()
+declare i32 @llvm.ptx.read.ctaid.y()
+declare i32 @llvm.ptx.read.ctaid.z()
+declare i32 @llvm.ptx.read.ctaid.w()
+declare i32 @llvm.ptx.read.nctaid.x()
+declare i32 @llvm.ptx.read.nctaid.y()
+declare i32 @llvm.ptx.read.nctaid.z()
+declare i32 @llvm.ptx.read.nctaid.w()
 
 declare i32 @llvm.ptx.read.smid()
 declare i32 @llvm.ptx.read.nsmid()

diff --git a/test/CodeGen/PTX/options.ll b/test/CodeGen/PTX/options.ll
index 1435537e007869d897d6b1a59ebbbc991eb295e1..6576a6d8bbb874dfd53af179a21c01799ea4c4be 100644 (file)
--- a/test/CodeGen/PTX/options.ll
+++ b/test/CodeGen/PTX/options.ll
@@ -1,8 +1,9 @@
-; RUN: llc < %s -march=ptx -mattr=ptx14 | grep ".version 1.4"
 ; RUN: llc < %s -march=ptx -mattr=ptx20 | grep ".version 2.0"
 ; RUN: llc < %s -march=ptx -mattr=ptx21 | grep ".version 2.1"
-; RUN: llc < %s -march=ptx -mattr=sm20 | grep ".target sm_20"
+; RUN: llc < %s -march=ptx -mattr=ptx22 | grep ".version 2.2"
+; RUN: llc < %s -march=ptx -mattr=sm10 | grep ".target sm_10"
 ; RUN: llc < %s -march=ptx -mattr=sm13 | grep ".target sm_13"
+; RUN: llc < %s -march=ptx -mattr=sm20 | grep ".target sm_20"
 
 define ptx_device void @t1() {
        ret void