[RISCV] Improve constant materialisation for stores of i8 negative constants (#92131)

asb · web-flow · commit 90109d444839 · 2024-05-14T19:08:04.000+01:00
This follows the same pattern as 20e6265. Although we can't reduce the number of instructions used, if we are able to use a sign-extended 6-bit immediate then the 16-bit c.li instruction can be selected (thus saving code size). Although this _could_ be gated so it only happens if C is enabled, I've opted not to because at worst it's neutral and it doesn't seem helpful to add unnecessary divergence between the RVC and non-RVC paths.
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -902,6 +902,11 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
       return;
     }
     int64_t Imm = ConstNode->getSExtValue();
+    // If only the lower 8 bits are used, try to convert this to a simm6 by
+    // sign-extending bit 7. This is neutral without the C extension, and
+    // allows C.LI to be used if C is present.
+    if (isUInt<8>(Imm) && isInt<6>(SignExtend64<8>(Imm)) && hasAllBUsers(Node))
+      Imm = SignExtend64<8>(Imm);
     // If the upper XLen-16 bits are not used, try to convert this to a simm12
     // by sign extending bit 15.
     if (isUInt<16>(Imm) && isInt<12>(SignExtend64<16>(Imm)) &&
diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.h b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.h
@@ -121,6 +121,7 @@ class RISCVDAGToDAGISel : public SelectionDAGISel {
 
   bool hasAllNBitUsers(SDNode *Node, unsigned Bits,
                        const unsigned Depth = 0) const;
+  bool hasAllBUsers(SDNode *Node) const { return hasAllNBitUsers(Node, 8); }
   bool hasAllHUsers(SDNode *Node) const { return hasAllNBitUsers(Node, 16); }
   bool hasAllWUsers(SDNode *Node) const { return hasAllNBitUsers(Node, 32); }
 
diff --git a/llvm/test/CodeGen/RISCV/imm.ll b/llvm/test/CodeGen/RISCV/imm.ll
@@ -1558,54 +1558,52 @@ define i64 @imm_2reg_1() nounwind {
   ret i64 -1152921504301427080 ; 0xF000_0000_1234_5678
 }
 
-; TODO: Selecting -1 would be better in this case as it can be loaded with a
-; 16 bit instruction when the compressed extension is enabled.
 define void @imm_store_i8_neg1(ptr %p) nounwind {
 ; RV32I-LABEL: imm_store_i8_neg1:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    li a1, 255
+; RV32I-NEXT:    li a1, -1
 ; RV32I-NEXT:    sb a1, 0(a0)
 ; RV32I-NEXT:    ret
 ;
 ; RV64I-LABEL: imm_store_i8_neg1:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    li a1, 255
+; RV64I-NEXT:    li a1, -1
 ; RV64I-NEXT:    sb a1, 0(a0)
 ; RV64I-NEXT:    ret
 ;
 ; RV64IZBA-LABEL: imm_store_i8_neg1:
 ; RV64IZBA:       # %bb.0:
-; RV64IZBA-NEXT:    li a1, 255
+; RV64IZBA-NEXT:    li a1, -1
 ; RV64IZBA-NEXT:    sb a1, 0(a0)
 ; RV64IZBA-NEXT:    ret
 ;
 ; RV64IZBB-LABEL: imm_store_i8_neg1:
 ; RV64IZBB:       # %bb.0:
-; RV64IZBB-NEXT:    li a1, 255
+; RV64IZBB-NEXT:    li a1, -1
 ; RV64IZBB-NEXT:    sb a1, 0(a0)
 ; RV64IZBB-NEXT:    ret
 ;
 ; RV64IZBS-LABEL: imm_store_i8_neg1:
 ; RV64IZBS:       # %bb.0:
-; RV64IZBS-NEXT:    li a1, 255
+; RV64IZBS-NEXT:    li a1, -1
 ; RV64IZBS-NEXT:    sb a1, 0(a0)
 ; RV64IZBS-NEXT:    ret
 ;
 ; RV64IXTHEADBB-LABEL: imm_store_i8_neg1:
 ; RV64IXTHEADBB:       # %bb.0:
-; RV64IXTHEADBB-NEXT:    li a1, 255
+; RV64IXTHEADBB-NEXT:    li a1, -1
 ; RV64IXTHEADBB-NEXT:    sb a1, 0(a0)
 ; RV64IXTHEADBB-NEXT:    ret
 ;
 ; RV32-REMAT-LABEL: imm_store_i8_neg1:
 ; RV32-REMAT:       # %bb.0:
-; RV32-REMAT-NEXT:    li a1, 255
+; RV32-REMAT-NEXT:    li a1, -1
 ; RV32-REMAT-NEXT:    sb a1, 0(a0)
 ; RV32-REMAT-NEXT:    ret
 ;
 ; RV64-REMAT-LABEL: imm_store_i8_neg1:
 ; RV64-REMAT:       # %bb.0:
-; RV64-REMAT-NEXT:    li a1, 255
+; RV64-REMAT-NEXT:    li a1, -1
 ; RV64-REMAT-NEXT:    sb a1, 0(a0)
 ; RV64-REMAT-NEXT:    ret
   store i8 -1, ptr %p
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll
@@ -17,7 +17,7 @@ define <4 x half> @shuffle_v4f16(<4 x half> %x, <4 x half> %y) {
 define <8 x float> @shuffle_v8f32(<8 x float> %x, <8 x float> %y) {
 ; CHECK-LABEL: shuffle_v8f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a0, 236
+; CHECK-NEXT:    li a0, -20
 ; CHECK-NEXT:    vsetivli zero, 8, e32, m2, ta, ma
 ; CHECK-NEXT:    vmv.s.x v0, a0
 ; CHECK-NEXT:    vmerge.vvm v8, v10, v8, v0
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-buildvec.ll
@@ -260,7 +260,7 @@ define <4 x i8> @buildvec_vid_stepn3_add3_v4i8() {
 ; CHECK-NEXT:    vsetivli zero, 4, e8, mf4, ta, ma
 ; CHECK-NEXT:    vmv.v.i v9, 3
 ; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    li a0, 253
+; CHECK-NEXT:    li a0, -3
 ; CHECK-NEXT:    vmadd.vx v8, a0, v9
 ; CHECK-NEXT:    ret
   ret <4 x i8> <i8 3, i8 0, i8 -3, i8 -6>
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
@@ -611,7 +611,7 @@ define <8 x i8> @concat_4xi8_start_undef(<8 x i8> %v, <8 x i8> %w) {
 define <8 x i8> @concat_4xi8_start_undef_at_start(<8 x i8> %v, <8 x i8> %w) {
 ; CHECK-LABEL: concat_4xi8_start_undef_at_start:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a0, 224
+; CHECK-NEXT:    li a0, -32
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, mu
 ; CHECK-NEXT:    vmv.s.x v0, a0
 ; CHECK-NEXT:    vslideup.vi v8, v9, 4, v0.t
@@ -682,7 +682,7 @@ define <8 x i8> @merge_non_contiguous_slideup_slidedown(<8 x i8> %v, <8 x i8> %w
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, mu
 ; CHECK-NEXT:    vslidedown.vi v8, v8, 2
-; CHECK-NEXT:    li a0, 234
+; CHECK-NEXT:    li a0, -22
 ; CHECK-NEXT:    vmv.s.x v0, a0
 ; CHECK-NEXT:    vslideup.vi v8, v9, 1, v0.t
 ; CHECK-NEXT:    ret
@@ -699,7 +699,7 @@ define <8 x i8> @unmergable(<8 x i8> %v, <8 x i8> %w) {
 ; CHECK-NEXT:    lui a0, %hi(.LCPI46_0)
 ; CHECK-NEXT:    addi a0, a0, %lo(.LCPI46_0)
 ; CHECK-NEXT:    vle8.v v10, (a0)
-; CHECK-NEXT:    li a0, 234
+; CHECK-NEXT:    li a0, -22
 ; CHECK-NEXT:    vmv.s.x v0, a0
 ; CHECK-NEXT:    vrgather.vv v8, v9, v10, v0.t
 ; CHECK-NEXT:    ret
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-reduction-formation.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-reduction-formation.ll
@@ -160,7 +160,7 @@ define i32 @reduce_sum_16xi32_prefix4(ptr %p) {
 define i32 @reduce_sum_16xi32_prefix5(ptr %p) {
 ; CHECK-LABEL: reduce_sum_16xi32_prefix5:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 224
+; CHECK-NEXT:    li a1, -32
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.s.x v0, a1
 ; CHECK-NEXT:    vmv.v.i v8, -1
@@ -532,7 +532,7 @@ define i32 @reduce_xor_16xi32_prefix2(ptr %p) {
 define i32 @reduce_xor_16xi32_prefix5(ptr %p) {
 ; CHECK-LABEL: reduce_xor_16xi32_prefix5:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 224
+; CHECK-NEXT:    li a1, -32
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.s.x v0, a1
 ; CHECK-NEXT:    vmv.v.i v8, -1
@@ -620,7 +620,7 @@ define i32 @reduce_or_16xi32_prefix2(ptr %p) {
 define i32 @reduce_or_16xi32_prefix5(ptr %p) {
 ; CHECK-LABEL: reduce_or_16xi32_prefix5:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 224
+; CHECK-NEXT:    li a1, -32
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.s.x v0, a1
 ; CHECK-NEXT:    vmv.v.i v8, -1
@@ -757,7 +757,7 @@ define i32 @reduce_umax_16xi32_prefix2(ptr %p) {
 define i32 @reduce_umax_16xi32_prefix5(ptr %p) {
 ; CHECK-LABEL: reduce_umax_16xi32_prefix5:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    li a1, 224
+; CHECK-NEXT:    li a1, -32
 ; CHECK-NEXT:    vsetivli zero, 8, e8, mf2, ta, ma
 ; CHECK-NEXT:    vmv.s.x v0, a1
 ; CHECK-NEXT:    vmv.v.i v8, -1
diff --git a/llvm/test/CodeGen/RISCV/unaligned-load-store.ll b/llvm/test/CodeGen/RISCV/unaligned-load-store.ll
@@ -419,7 +419,7 @@ define void @merge_stores_i32_i64(ptr %p) {
 define void @store_large_constant(ptr %x) {
 ; SLOW-LABEL: store_large_constant:
 ; SLOW:       # %bb.0:
-; SLOW-NEXT:    li a1, 254
+; SLOW-NEXT:    li a1, -2
 ; SLOW-NEXT:    sb a1, 7(a0)
 ; SLOW-NEXT:    li a1, 220
 ; SLOW-NEXT:    sb a1, 6(a0)
