Possible JT improvements.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@33733 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 43 insertions, 3 deletions

diff --git a/lib/Target/ARM/README-Thumb.txt b/lib/Target/ARM/README-Thumb.txt
index b0473da6a7..dd0d265af3 100644
--- a/lib/Target/ARM/README-Thumb.txt
+++ b/lib/Target/ARM/README-Thumb.txt
@@ -16,10 +16,50 @@
   and cmp instructions can use high registers. Also, we can use them as
   temporaries to spill values into.
 
+* In thumb mode, short, byte, and bool preferred alignments are currently set
+  to 4 to accommodate ISA restriction (i.e. add sp, #imm, imm must be multiple
+  of 4).
+
+//===---------------------------------------------------------------------===//
+
+Potential jumptable improvements:
+
 * If we know function size is less than (1 << 16) * 2 bytes, we can use 16-bit
   jumptable entries (e.g. (L1 - L2) >> 1). Or even smaller entries if the
   function is even smaller. This also applies to ARM.
 
-* In thumb mode, short, byte, and bool preferred alignments are currently set
-  to 4 to accommodate ISA restriction (i.e. add sp, #imm, imm must be multiple
-  of 4).
+* Thumb jumptable codegen can improve given some help from the assembler. This
+  is what we generate right now:
+
+	.set PCRELV0, (LJTI1_0_0-(LPCRELL0+4))
+LPCRELL0:
+	mov r1, #PCRELV0
+	add r1, pc
+	ldr r0, [r0, r1]
+	cpy pc, r0 
+	.align	2
+LJTI1_0_0:
+	.long	 LBB1_3
+        ...
+
+Note there is another pc relative add that we can take advantage of.
+     add r1, pc, #imm_8 * 4
+
+We should be able to generate:
+
+LPCRELL0:
+	add r1, LJTI1_0_0
+	ldr r0, [r0, r1]
+	cpy pc, r0 
+	.align	2
+LJTI1_0_0:
+	.long	 LBB1_3
+
+if the assembler can translate the add to:
+       add r1, pc, #((LJTI1_0_0-(LPCRELL0+4))&0xfffffffc)
+
+Note the assembler also does something similar to constpool load:
+LPCRELL0:
+     ldr r0, LCPI1_0
+=>
+     ldr r0, pc, #((LCPI1_0-(LPCRELL0+4))&0xfffffffc)