This is the patch to provide clean intrinsic function overloading support in LLVM. It cleans up the intrinsic definitions and generally smooths the process for more complicated intrinsic writing. It will be used by the upcoming atomic intrinsics as well as vector and float intrinsics in the future.

This also changes the syntax for llvm.bswap, llvm.part.set, llvm.part.select, and llvm.ct* intrinsics. They are automatically upgraded by both the LLVM ASM reader and the bitcode reader. The test cases have been updated, with special tests added to ensure the automatic upgrading is supported.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@40807 91177308-0d34-0410-b5e6-96231b3b80d8

8 files changed, 103 insertions, 73 deletions

diff --git a/test/Assembler/AutoUpgradeIntrinsics.ll b/test/Assembler/AutoUpgradeIntrinsics.ll
new file mode 100644
index 0000000..fad3d4e
--- /dev/null
+++ b/test/Assembler/AutoUpgradeIntrinsics.ll
@@ -0,0 +1,52 @@
+; Tests to make sure intrinsics are automatically upgraded.
+; RUN: llvm-as < %s | llvm-dis | not grep {i32 @llvm\\.ct}
+; RUN: llvm-as < %s | llvm-dis | \
+; RUN:   not grep {llvm\\.part\\.set\\.i\[0-9\]*\\.i\[0-9\]*\\.i\[0-9\]*}
+; RUN: llvm-as < %s | llvm-dis | \
+; RUN:   not grep {llvm\\.part\\.select\\.i\[0-9\]*\\.i\[0-9\]*}
+; RUN: llvm-as < %s | llvm-dis | \
+; RUN:   not grep {llvm\\.bswap\\.i\[0-9\]*\\.i\[0-9\]*}
+
+declare i32 @llvm.ctpop.i28(i28 %val)
+declare i32 @llvm.cttz.i29(i29 %val)
+declare i32 @llvm.ctlz.i30(i30 %val)
+
+define i32 @test_ct(i32 %A) {
+  %c1 = call i32 @llvm.ctpop.i28(i28 1234)
+  %c2 = call i32 @llvm.cttz.i29(i29 2345)
+  %c3 = call i32 @llvm.ctlz.i30(i30 3456)
+  %r1 = add i32 %c1, %c2
+  %r2 = add i32 %r1, %c3
+  ret i32 %r2
+}
+
+declare i32 @llvm.part.set.i32.i32.i32(i32 %x, i32 %rep, i32 %hi, i32 %lo)
+declare i16 @llvm.part.set.i16.i16.i16(i16 %x, i16 %rep, i32 %hi, i32 %lo)
+define i32 @test_part_set(i32 %A, i16 %B) {
+  %a = call i32 @llvm.part.set.i32.i32.i32(i32 %A, i32 27, i32 8, i32 0)
+  %b = call i16 @llvm.part.set.i16.i16.i16(i16 %B, i16 27, i32 8, i32 0)
+  %c = zext i16 %b to i32
+  %d = add i32 %a, %c
+  ret i32 %d
+}
+
+declare i32 @llvm.part.select.i32.i32(i32 %x, i32 %hi, i32 %lo)
+declare i16 @llvm.part.select.i16.i16(i16 %x, i32 %hi, i32 %lo)
+define i32 @test_part_select(i32 %A, i16 %B) {
+  %a = call i32 @llvm.part.select.i32.i32(i32 %A, i32 8, i32 0)
+  %b = call i16 @llvm.part.select.i16.i16(i16 %B, i32 8, i32 0)
+  %c = zext i16 %b to i32
+  %d = add i32 %a, %c
+  ret i32 %d
+}
+
+declare i32 @llvm.bswap.i32.i32(i32 %x)
+declare i16 @llvm.bswap.i16.i16(i16 %x)
+define i32 @test_bswap(i32 %A, i16 %B) {
+  %a = call i32 @llvm.bswap.i32.i32(i32 %A)
+  %b = call i16 @llvm.bswap.i16.i16(i16 %B)
+  %c = zext i16 %b to i32
+  %d = add i32 %a, %c
+  ret i32 %d
+}
+
diff --git a/test/Bitcode/AutoUpgradeIntrinsics.ll b/test/Bitcode/AutoUpgradeIntrinsics.ll
new file mode 100644
index 0000000..5f9bcd5
--- /dev/null
+++ b/test/Bitcode/AutoUpgradeIntrinsics.ll
@@ -0,0 +1,10 @@
+; This isn't really an assembly file. It just runs test on bitcode to ensure
+; it is auto-upgraded.
+; RUN: llvm-dis < %s.bc | not grep {i32 @llvm\\.ct}
+; RUN: llvm-dis < %s.bc | \
+; RUN:   not grep {llvm\\.part\\.set\\.i\[0-9\]*\\.i\[0-9\]*\\.i\[0-9\]*}
+; RUN: llvm-dis < %s.bc | \
+; RUN:   not grep {llvm\\.part\\.select\\.i\[0-9\]*\\.i\[0-9\]*}
+; RUN: llvm-dis < %s.bc | \
+; RUN:   not grep {llvm\\.bswap\\.i\[0-9\]*\\.i\[0-9\]*}
+
diff --git a/test/Bitcode/AutoUpgradeIntrinsics.ll.bc b/test/Bitcode/AutoUpgradeIntrinsics.ll.bc
new file mode 100644
index 0000000..9de756b
--- /dev/null
+++ b/test/Bitcode/AutoUpgradeIntrinsics.ll.bc
diff --git a/test/CodeGen/Alpha/ctlz.ll b/test/CodeGen/Alpha/ctlz.ll
index 0ad014d..fba6022 100644
--- a/test/CodeGen/Alpha/ctlz.ll
+++ b/test/CodeGen/Alpha/ctlz.ll
@@ -5,10 +5,11 @@
 ; RUN: llvm-as < %s | llc -march=alpha -mcpu=ev56 | not grep -i ctlz
 ; RUN: llvm-as < %s | llc -march=alpha -mattr=-CIX | not grep -i ctlz
 
-declare i32 @llvm.ctlz.i8(i8)
+declare i8 @llvm.ctlz.i8(i8)
 
 define i32 @bar(i8 %x) {
 entry:
-	%tmp.1 = call i32 @llvm.ctlz.i8( i8 %x ) 
-	ret i32 %tmp.1
+	%tmp.1 = call i8 @llvm.ctlz.i8( i8 %x ) 
+	%tmp.2 = sext i8 %tmp.1 to i32
+	ret i32 %tmp.2
 }
diff --git a/test/CodeGen/Generic/bit-intrinsics.ll b/test/CodeGen/Generic/bit-intrinsics.ll
index 427387f..9541011 100644
--- a/test/CodeGen/Generic/bit-intrinsics.ll
+++ b/test/CodeGen/Generic/bit-intrinsics.ll
@@ -3,21 +3,21 @@
 ; RUN: llvm-as < %s > %t.bc
 ; RUN: lli --force-interpreter=true %t.bc
 
-declare i32 @llvm.part.set.i32.i32.i32(i32 %x, i32 %rep, i32 %hi, i32 %lo)
-declare i16 @llvm.part.set.i16.i16.i16(i16 %x, i16 %rep, i32 %hi, i32 %lo)
+declare i32 @llvm.part.set.i32.i32(i32 %x, i32 %rep, i32 %hi, i32 %lo)
+declare i16 @llvm.part.set.i16.i16(i16 %x, i16 %rep, i32 %hi, i32 %lo)
 define i32 @test_part_set(i32 %A, i16 %B) {
-  %a = call i32 @llvm.part.set.i32.i32.i32(i32 %A, i32 27, i32 8, i32 0)
-  %b = call i16 @llvm.part.set.i16.i16.i16(i16 %B, i16 27, i32 8, i32 0)
+  %a = call i32 @llvm.part.set.i32.i32(i32 %A, i32 27, i32 8, i32 0)
+  %b = call i16 @llvm.part.set.i16.i16(i16 %B, i16 27, i32 8, i32 0)
   %c = zext i16 %b to i32
   %d = add i32 %a, %c
   ret i32 %d
 }
 
-declare i32 @llvm.part.select.i32.i32(i32 %x, i32 %hi, i32 %lo)
-declare i16 @llvm.part.select.i16.i16(i16 %x, i32 %hi, i32 %lo)
+declare i32 @llvm.part.select.i32(i32 %x, i32 %hi, i32 %lo)
+declare i16 @llvm.part.select.i16(i16 %x, i32 %hi, i32 %lo)
 define i32 @test_part_select(i32 %A, i16 %B) {
-  %a = call i32 @llvm.part.select.i32.i32(i32 %A, i32 8, i32 0)
-  %b = call i16 @llvm.part.select.i16.i16(i16 %B, i32 8, i32 0)
+  %a = call i32 @llvm.part.select.i32(i32 %A, i32 8, i32 0)
+  %b = call i16 @llvm.part.select.i16(i16 %B, i32 8, i32 0)
   %c = zext i16 %b to i32
   %d = add i32 %a, %c
   ret i32 %d
diff --git a/test/CodeGen/PowerPC/2007-03-24-cntlzd.ll b/test/CodeGen/PowerPC/2007-03-24-cntlzd.ll
index 1ea6174..098e748 100644
--- a/test/CodeGen/PowerPC/2007-03-24-cntlzd.ll
+++ b/test/CodeGen/PowerPC/2007-03-24-cntlzd.ll
@@ -2,10 +2,11 @@
 
 define i32 @_ZNK4llvm5APInt17countLeadingZerosEv(i64 *%t) {
         %tmp19 = load i64* %t
-        %tmp23 = tail call i32 @llvm.ctlz.i64( i64 %tmp19 )             ; <i64> [#uses=1]
+        %tmp22 = tail call i64 @llvm.ctlz.i64( i64 %tmp19 )             ; <i64> [#uses=1]
+        %tmp23 = trunc i64 %tmp22 to i32
         %tmp89 = add i32 %tmp23, -64          ; <i32> [#uses=1]
         %tmp90 = add i32 %tmp89, 0            ; <i32> [#uses=1]
         ret i32 %tmp90
 }
 
-declare i32 @llvm.ctlz.i64(i64)
+declare i64 @llvm.ctlz.i64(i64)
diff --git a/test/Feature/llvm2cpp.ll b/test/Feature/llvm2cpp.ll
index 88cfa34..5c381c7 100644
--- a/test/Feature/llvm2cpp.ll
+++ b/test/Feature/llvm2cpp.ll
@@ -403,30 +403,6 @@ declare i1 @llvm.isunordered.f64(double, double)
 
 declare void @llvm.prefetch(i8*, i32, i32)
 
-declare i32 @upgrd.rm.llvm.ctpop.i8(i8)
-
-declare i32 @upgrd.rm.llvm.ctpop.i16(i16)
-
-declare i32 @upgrd.rm.llvm.ctpop.i32(i32)
-
-declare i32 @upgrd.rm.llvm.ctpop.i64(i64)
-
-declare i32 @upgrd.rm.llvm.cttz.i8(i8)
-
-declare i32 @upgrd.rm.llvm.cttz.i16(i16)
-
-declare i32 @upgrd.rm.llvm.cttz.i32(i32)
-
-declare i32 @upgrd.rm.llvm.cttz.i64(i64)
-
-declare i32 @upgrd.rm.llvm.ctlz.i8(i8)
-
-declare i32 @upgrd.rm.llvm.ctlz.i16(i16)
-
-declare i32 @upgrd.rm.llvm.ctlz.i32(i32)
-
-declare i32 @upgrd.rm.llvm.ctlz.i64(i64)
-
 declare float @llvm.sqrt.f32(float)
 
 declare double @llvm.sqrt.f64(double)
@@ -437,56 +413,44 @@ define void @libm() {
 	call void @llvm.prefetch( i8* null, i32 1, i32 3 )
 	call float @llvm.sqrt.f32( float 5.000000e+00 )		; <float>:3 [#uses=0]
 	call double @llvm.sqrt.f64( double 6.000000e+00 )		; <double>:4 [#uses=0]
-	call i32 @llvm.ctpop.i8( i8 10 )		; <i32>:5 [#uses=1]
-	bitcast i32 %5 to i32		; <i32>:6 [#uses=0]
-	call i32 @llvm.ctpop.i16( i16 11 )		; <i32>:7 [#uses=1]
-	bitcast i32 %7 to i32		; <i32>:8 [#uses=0]
+	call i8 @llvm.ctpop.i8( i8 10 )		; <i32>:5 [#uses=1]
+	call i16 @llvm.ctpop.i16( i16 11 )		; <i32>:7 [#uses=1]
 	call i32 @llvm.ctpop.i32( i32 12 )		; <i32>:9 [#uses=1]
-	bitcast i32 %9 to i32		; <i32>:10 [#uses=0]
-	call i32 @llvm.ctpop.i64( i64 13 )		; <i32>:11 [#uses=1]
-	bitcast i32 %11 to i32		; <i32>:12 [#uses=0]
-	call i32 @llvm.ctlz.i8( i8 14 )		; <i32>:13 [#uses=1]
-	bitcast i32 %13 to i32		; <i32>:14 [#uses=0]
-	call i32 @llvm.ctlz.i16( i16 15 )		; <i32>:15 [#uses=1]
-	bitcast i32 %15 to i32		; <i32>:16 [#uses=0]
+	call i64 @llvm.ctpop.i64( i64 13 )		; <i32>:11 [#uses=1]
+	call i8 @llvm.ctlz.i8( i8 14 )		; <i32>:13 [#uses=1]
+	call i16 @llvm.ctlz.i16( i16 15 )		; <i32>:15 [#uses=1]
 	call i32 @llvm.ctlz.i32( i32 16 )		; <i32>:17 [#uses=1]
-	bitcast i32 %17 to i32		; <i32>:18 [#uses=0]
-	call i32 @llvm.ctlz.i64( i64 17 )		; <i32>:19 [#uses=1]
-	bitcast i32 %19 to i32		; <i32>:20 [#uses=0]
-	call i32 @llvm.cttz.i8( i8 18 )		; <i32>:21 [#uses=1]
-	bitcast i32 %21 to i32		; <i32>:22 [#uses=0]
-	call i32 @llvm.cttz.i16( i16 19 )		; <i32>:23 [#uses=1]
-	bitcast i32 %23 to i32		; <i32>:24 [#uses=0]
+	call i64 @llvm.ctlz.i64( i64 17 )		; <i32>:19 [#uses=1]
+	call i8 @llvm.cttz.i8( i8 18 )		; <i32>:21 [#uses=1]
+	call i16 @llvm.cttz.i16( i16 19 )		; <i32>:23 [#uses=1]
 	call i32 @llvm.cttz.i32( i32 20 )		; <i32>:25 [#uses=1]
-	bitcast i32 %25 to i32		; <i32>:26 [#uses=0]
-	call i32 @llvm.cttz.i64( i64 21 )		; <i32>:27 [#uses=1]
-	bitcast i32 %27 to i32		; <i32>:28 [#uses=0]
+	call i64 @llvm.cttz.i64( i64 21 )		; <i32>:27 [#uses=1]
 	ret void
 }
 
-declare i32 @llvm.ctpop.i8(i8)
+declare i8 @llvm.ctpop.i8(i8)
 
-declare i32 @llvm.ctpop.i16(i16)
+declare i16 @llvm.ctpop.i16(i16)
 
 declare i32 @llvm.ctpop.i32(i32)
 
-declare i32 @llvm.ctpop.i64(i64)
+declare i64 @llvm.ctpop.i64(i64)
 
-declare i32 @llvm.ctlz.i8(i8)
+declare i8 @llvm.ctlz.i8(i8)
 
-declare i32 @llvm.ctlz.i16(i16)
+declare i16 @llvm.ctlz.i16(i16)
 
 declare i32 @llvm.ctlz.i32(i32)
 
-declare i32 @llvm.ctlz.i64(i64)
+declare i64 @llvm.ctlz.i64(i64)
 
-declare i32 @llvm.cttz.i8(i8)
+declare i8 @llvm.cttz.i8(i8)
 
-declare i32 @llvm.cttz.i16(i16)
+declare i16 @llvm.cttz.i16(i16)
 
 declare i32 @llvm.cttz.i32(i32)
 
-declare i32 @llvm.cttz.i64(i64)
+declare i64 @llvm.cttz.i64(i64)
 
 ; ModuleID = 'packed.ll'
 @foo1 = external global <4 x float>		; <<4 x float>*> [#uses=2]
diff --git a/test/Transforms/InstCombine/bitcount.ll b/test/Transforms/InstCombine/bitcount.ll
index da53920..8ebf289 100644
--- a/test/Transforms/InstCombine/bitcount.ll
+++ b/test/Transforms/InstCombine/bitcount.ll
@@ -3,15 +3,17 @@
 ; RUN: llvm-as < %s | opt -instcombine | llvm-dis | \
 ; RUN:   grep -v declare | not grep llvm.ct
 
-declare i32 @llvm.ctpop.i31(i31 %val) 
+declare i31 @llvm.ctpop.i31(i31 %val) 
 declare i32 @llvm.cttz.i32(i32 %val) 
-declare i32 @llvm.ctlz.i33(i33 %val) 
+declare i33 @llvm.ctlz.i33(i33 %val) 
 
 define i32 @test(i32 %A) {
-  %c1 = call i32 @llvm.ctpop.i31(i31 12415124)
+  %c1 = call i31 @llvm.ctpop.i31(i31 12415124)
   %c2 = call i32 @llvm.cttz.i32(i32 87359874)
-  %c3 = call i32 @llvm.ctlz.i33(i33 87359874)
-  %r1 = add i32 %c1, %c2
-  %r2 = add i32 %r1, %c3
+  %c3 = call i33 @llvm.ctlz.i33(i33 87359874)
+  %t1 = zext i31 %c1 to i32
+  %t3 = trunc i33 %c3 to i32
+  %r1 = add i32 %t1, %c2
+  %r2 = add i32 %r1, %t3
   ret i32 %r2
 }