Split the Add, Sub, and Mul instruction opcodes into separate

integer and floating-point opcodes, introducing
FAdd, FSub, and FMul.

For now, the AsmParser, BitcodeReader, and IRBuilder all preserve
backwards compatability, and the Core LLVM APIs preserve backwards
compatibility for IR producers. Most front-ends won't need to change
immediately.

This implements the first step of the plan outlined here:
http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72897 91177308-0d34-0410-b5e6-96231b3b80d8

test/CodeGen/XCore/2009-01-14-Remat-Crash.ll

@@ -8,11 +8,11 @@ bb113:		; preds = %entry
 	ret double 0.000000e+00
 
 bb129:		; preds = %entry
-	%tmp134 = sub double %b, %a		; <double> [#uses=1]
-	%tmp136 = sub double %tmp134, %c		; <double> [#uses=1]
-	%tmp138 = add double %tmp136, %d		; <double> [#uses=1]
-	%tmp140 = sub double %tmp138, %e		; <double> [#uses=1]
-	%tmp142 = add double %tmp140, %f		; <double> [#uses=1]
-	%tmp.0 = mul double %tmp142, 0.000000e+00		; <double> [#uses=1]
+	%tmp134 = fsub double %b, %a		; <double> [#uses=1]
+	%tmp136 = fsub double %tmp134, %c		; <double> [#uses=1]
+	%tmp138 = fadd double %tmp136, %d		; <double> [#uses=1]
+	%tmp140 = fsub double %tmp138, %e		; <double> [#uses=1]
+	%tmp142 = fadd double %tmp140, %f		; <double> [#uses=1]
+	%tmp.0 = fmul double %tmp142, 0.000000e+00		; <double> [#uses=1]
 	ret double %tmp.0
 }

test/CodeGen/XCore/fneg.ll

@@ -2,7 +2,7 @@
 ; RUN: grep "xor" %t1.s | count 1
 define i1 @test(double %F) nounwind {
 entry:
-	%0 = sub double -0.000000e+00, %F
+	%0 = fsub double -0.000000e+00, %F
 	%1 = fcmp olt double 0.000000e+00, %0
 	ret i1 %1
 }