[PATCH] R600: Fix trunc i64 to i32 on SI
Matt Arsenault
Matthew.Arsenault at amd.com
Wed Oct 9 16:23:55 PDT 2013
Attach right patch
http://llvm-reviews.chandlerc.com/D1882
CHANGE SINCE LAST DIFF
http://llvm-reviews.chandlerc.com/D1882?vs=4791&id=4792#toc
Files:
lib/Target/R600/SIInstructions.td
test/CodeGen/R600/trunc.ll
Index: lib/Target/R600/SIInstructions.td
===================================================================
--- lib/Target/R600/SIInstructions.td
+++ lib/Target/R600/SIInstructions.td
@@ -851,6 +851,11 @@
(V_CNDMASK_B32_e64 $src0, $src1, $src2)
>;
+def : Pat <
+ (i32 (trunc i64:$val)),
+ (EXTRACT_SUBREG $val, sub0)
+>;
+
//use two V_CNDMASK_B32_e64 instructions for f64
def : Pat <
(f64 (select i1:$src2, f64:$src1, f64:$src0)),
Index: test/CodeGen/R600/trunc.ll
===================================================================
--- test/CodeGen/R600/trunc.ll
+++ test/CodeGen/R600/trunc.ll
@@ -17,3 +17,15 @@
ret void
}
+; SI-LABEL: @trunc_shl_i64:
+; SI: S_LOAD_DWORDX2
+; SI: S_LOAD_DWORDX2 [[SREG:SGPR[0-9]+_SGPR[0-9]+]]
+; SI: V_LSHL_B64 [[LO_VREG:VGPR[0-9]+]]_VGPR{{[0-9]+}}, [[SREG]], 2
+; SI-NOT: [[LO_VREG]]
+; SI: BUFFER_STORE_DWORD [[LO_VREG]],
+define void @trunc_shl_i64(i32 addrspace(1)* %out, i64 %a) {
+ %b = shl i64 %a, 2
+ %result = trunc i64 %b to i32
+ store i32 %result, i32 addrspace(1)* %out, align 4
+ ret void
+}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D1882.2.patch
Type: text/x-patch
Size: 1066 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131009/075f1a69/attachment.bin>
More information about the llvm-commits
mailing list