Add thumb-2 store word, halfword, and byte.

llvm-svn: 74555

4 files changed, 207 insertions, 0 deletions

diff --git a/llvm/lib/Target/ARM/ARMInstrThumb2.td b/llvm/lib/Target/ARM/ARMInstrThumb2.td
index 8479154319f..810cb067f27 100644
--- a/llvm/lib/Target/ARM/ARMInstrThumb2.td
+++ b/llvm/lib/Target/ARM/ARMInstrThumb2.td
@@ -399,6 +399,19 @@ multiclass T2I_ld<string opc, PatFrag opnode> {
                    [(set GPR:$dst, (opnode (ARMWrapper tconstpool:$addr)))]>;
 }
 
+/// T2I_st - Defines a set of (op r, {imm12|imm8|so_reg}) store patterns.
+multiclass T2I_st<string opc, PatFrag opnode> {
+  def i12 : T2Ii12<(outs), (ins GPR:$src, t2addrmode_imm12:$addr),
+                   opc, " $src, $addr",
+                   [(opnode GPR:$src, t2addrmode_imm12:$addr)]>;
+  def i8  : T2Ii8 <(outs), (ins GPR:$src, t2addrmode_imm8:$addr),
+                   opc, " $src, $addr",
+                   [(opnode GPR:$src, t2addrmode_imm8:$addr)]>;
+  def s   : T2Iso <(outs), (ins GPR:$src, t2addrmode_so_reg:$addr),
+                   opc, " $src, $addr",
+                   [(opnode GPR:$src, t2addrmode_so_reg:$addr)]>;
+}
+
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
@@ -509,6 +522,11 @@ def : T2Pat<(extloadi16 t2addrmode_so_reg:$addr),
 def : T2Pat<(extloadi16 (ARMWrapper tconstpool:$addr)),
             (t2LDRHpci  tconstpool:$addr)>;
 
+// Store
+defm t2STR   : T2I_st<"str", BinOpFrag<(store node:$LHS, node:$RHS)>>;
+defm t2STRB   : T2I_st<"strb", BinOpFrag<(truncstorei8 node:$LHS, node:$RHS)>>;
+defm t2STRH   : T2I_st<"strh", BinOpFrag<(truncstorei16 node:$LHS, node:$RHS)>>;
+
 //===----------------------------------------------------------------------===//
 //  Move Instructions.
 //
diff --git a/llvm/test/CodeGen/Thumb2/thumb2-str.ll b/llvm/test/CodeGen/Thumb2/thumb2-str.ll
new file mode 100644
index 00000000000..4097a6c1579
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/thumb2-str.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4092\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#-128\\\]$} | count 2
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | not grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4096\\\]$}
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*\\\]$} | count 3
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {str\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*,\\Wlsl #2\\\]$} | count 1
+
+define i32 @f1(i32 %a, i32* %v) {
+        store i32 %a, i32* %v
+        ret i32 %a
+}
+
+define i32 @f2(i32 %a, i32* %v) {
+        %tmp2 = getelementptr i32* %v, i32 1023
+        store i32 %a, i32* %tmp2
+        ret i32 %a
+}
+
+define i32 @f2a(i32 %a, i32* %v) {
+        %tmp2 = getelementptr i32* %v, i32 -32
+        store i32 %a, i32* %tmp2
+        ret i32 %a
+}
+
+define i32 @f3(i32 %a, i32* %v) {
+        %tmp2 = getelementptr i32* %v, i32 1024
+        store i32 %a, i32* %tmp2
+        ret i32 %a
+}
+
+define i32 @f4(i32 %a, i32 %base) {
+entry:
+        %tmp1 = sub i32 %base, 128
+        %tmp2 = inttoptr i32 %tmp1 to i32*
+        store i32 %a, i32* %tmp2
+        ret i32 %a
+}
+
+define i32 @f5(i32 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = add i32 %base, %offset
+        %tmp2 = inttoptr i32 %tmp1 to i32*
+        store i32 %a, i32* %tmp2
+        ret i32 %a
+}
+
+define i32 @f6(i32 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = shl i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i32*
+        store i32 %a, i32* %tmp3
+        ret i32 %a
+}
+
+define i32 @f7(i32 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = lshr i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i32*
+        store i32 %a, i32* %tmp3
+        ret i32 %a
+}
diff --git a/llvm/test/CodeGen/Thumb2/thumb2-strb.ll b/llvm/test/CodeGen/Thumb2/thumb2-strb.ll
new file mode 100644
index 00000000000..d8401cd6847
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/thumb2-strb.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4092\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#-128\\\]$} | count 2
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | not grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4096\\\]$}
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*\\\]$} | count 3
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strb\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*,\\Wlsl #2\\\]$} | count 1
+
+define i8 @f1(i8 %a, i8* %v) {
+        store i8 %a, i8* %v
+        ret i8 %a
+}
+
+define i8 @f2(i8 %a, i8* %v) {
+        %tmp2 = getelementptr i8* %v, i32 4092
+        store i8 %a, i8* %tmp2
+        ret i8 %a
+}
+
+define i8 @f2a(i8 %a, i8* %v) {
+        %tmp2 = getelementptr i8* %v, i32 -128
+        store i8 %a, i8* %tmp2
+        ret i8 %a
+}
+
+define i8 @f3(i8 %a, i8* %v) {
+        %tmp2 = getelementptr i8* %v, i32 4096
+        store i8 %a, i8* %tmp2
+        ret i8 %a
+}
+
+define i8 @f4(i8 %a, i32 %base) {
+entry:
+        %tmp1 = sub i32 %base, 128
+        %tmp2 = inttoptr i32 %tmp1 to i8*
+        store i8 %a, i8* %tmp2
+        ret i8 %a
+}
+
+define i8 @f5(i8 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = add i32 %base, %offset
+        %tmp2 = inttoptr i32 %tmp1 to i8*
+        store i8 %a, i8* %tmp2
+        ret i8 %a
+}
+
+define i8 @f6(i8 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = shl i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i8*
+        store i8 %a, i8* %tmp3
+        ret i8 %a
+}
+
+define i8 @f7(i8 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = lshr i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i8*
+        store i8 %a, i8* %tmp3
+        ret i8 %a
+}
diff --git a/llvm/test/CodeGen/Thumb2/thumb2-strh.ll b/llvm/test/CodeGen/Thumb2/thumb2-strh.ll
new file mode 100644
index 00000000000..80dedf0c28d
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/thumb2-strh.ll
@@ -0,0 +1,63 @@
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4092\\\]$} | count 1
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#-128\\\]$} | count 2
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | not grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*#+4096\\\]$}
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*\\\]$} | count 3
+; RUN: llvm-as < %s | llc -march=thumb -mattr=+thumb2 | grep {strh\\W*r\[0-9\],\\W*\\\[r\[0-9\]*,\\W*+r\[0-9\]*,\\Wlsl #2\\\]$} | count 1
+
+define i16 @f1(i16 %a, i16* %v) {
+        store i16 %a, i16* %v
+        ret i16 %a
+}
+
+define i16 @f2(i16 %a, i16* %v) {
+        %tmp2 = getelementptr i16* %v, i32 2046
+        store i16 %a, i16* %tmp2
+        ret i16 %a
+}
+
+define i16 @f2a(i16 %a, i16* %v) {
+        %tmp2 = getelementptr i16* %v, i32 -64
+        store i16 %a, i16* %tmp2
+        ret i16 %a
+}
+
+define i16 @f3(i16 %a, i16* %v) {
+        %tmp2 = getelementptr i16* %v, i32 2048
+        store i16 %a, i16* %tmp2
+        ret i16 %a
+}
+
+define i16 @f4(i16 %a, i32 %base) {
+entry:
+        %tmp1 = sub i32 %base, 128
+        %tmp2 = inttoptr i32 %tmp1 to i16*
+        store i16 %a, i16* %tmp2
+        ret i16 %a
+}
+
+define i16 @f5(i16 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = add i32 %base, %offset
+        %tmp2 = inttoptr i32 %tmp1 to i16*
+        store i16 %a, i16* %tmp2
+        ret i16 %a
+}
+
+define i16 @f6(i16 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = shl i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i16*
+        store i16 %a, i16* %tmp3
+        ret i16 %a
+}
+
+define i16 @f7(i16 %a, i32 %base, i32 %offset) {
+entry:
+        %tmp1 = lshr i32 %offset, 2
+        %tmp2 = add i32 %base, %tmp1
+        %tmp3 = inttoptr i32 %tmp2 to i16*
+        store i16 %a, i16* %tmp3
+        ret i16 %a
+}