From 24ec87debb76a59016e854a7b28bd8e99fa74dc5 Mon Sep 17 00:00:00 2001
From: Tim Northover <tnorthover@apple.com>
Date: Tue, 2 Dec 2014 17:15:22 +0000
Subject: [PATCH] AArch64: make register block rules apply to vector types too.

The blocking code originated in ARM, which is more aggressive about casting
types to a canonical representative before doing anything else, so I missed out
most vector HFAs and broke the ABI. This should fix it.

llvm-svn: 223126
---
 llvm/lib/Target/AArch64/AArch64CallingConvention.h |  6 +-
 llvm/test/CodeGen/AArch64/argument-blocks.ll       | 98 ++++++++++++++++++++++
 2 files changed, 101 insertions(+), 3 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64CallingConvention.h b/llvm/lib/Target/AArch64/AArch64CallingConvention.h
index 6355164..8c798b7 100644
--- a/llvm/lib/Target/AArch64/AArch64CallingConvention.h
+++ b/llvm/lib/Target/AArch64/AArch64CallingConvention.h
@@ -88,11 +88,11 @@ static bool CC_AArch64_Custom_Block(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
   ArrayRef<uint16_t> RegList;
   if (LocVT.SimpleTy == MVT::i64)
     RegList = XRegList;
-  else if (LocVT.SimpleTy == MVT::f32)
+  else if (LocVT.SimpleTy == MVT::f32 || LocVT.is32BitVector())
     RegList = SRegList;
-  else if (LocVT.SimpleTy == MVT::f64)
+  else if (LocVT.SimpleTy == MVT::f64 || LocVT.is64BitVector())
     RegList = DRegList;
-  else if (LocVT.SimpleTy == MVT::v2f64)
+  else if (LocVT.SimpleTy == MVT::f128 || LocVT.is128BitVector())
     RegList = QRegList;
   else {
     // Not an array we want to split up after all.
diff --git a/llvm/test/CodeGen/AArch64/argument-blocks.ll b/llvm/test/CodeGen/AArch64/argument-blocks.ll
index cc65541..9204da8 100644
--- a/llvm/test/CodeGen/AArch64/argument-blocks.ll
+++ b/llvm/test/CodeGen/AArch64/argument-blocks.ll
@@ -90,3 +90,101 @@ define i64 @test_smallstruct_block_consume([7 x i64], [2 x i64] %in, i64 %rhs) {
   %sum = add i64 %lhs, %rhs
   ret i64 %sum
 }
+
+define <1 x i64> @test_v1i64_blocked([7 x double], [2 x <1 x i64>] %in) {
+; CHECK-LABEL: test_v1i64_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <1 x i64>] %in, 0
+  ret <1 x i64> %val
+}
+
+define <1 x double> @test_v1f64_blocked([7 x double], [2 x <1 x double>] %in) {
+; CHECK-LABEL: test_v1f64_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <1 x double>] %in, 0
+  ret <1 x double> %val
+}
+
+define <2 x i32> @test_v2i32_blocked([7 x double], [2 x <2 x i32>] %in) {
+; CHECK-LABEL: test_v2i32_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <2 x i32>] %in, 0
+  ret <2 x i32> %val
+}
+
+define <2 x float> @test_v2f32_blocked([7 x double], [2 x <2 x float>] %in) {
+; CHECK-LABEL: test_v2f32_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <2 x float>] %in, 0
+  ret <2 x float> %val
+}
+
+define <4 x i16> @test_v4i16_blocked([7 x double], [2 x <4 x i16>] %in) {
+; CHECK-LABEL: test_v4i16_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <4 x i16>] %in, 0
+  ret <4 x i16> %val
+}
+
+define <4 x half> @test_v4f16_blocked([7 x double], [2 x <4 x half>] %in) {
+; CHECK-LABEL: test_v4f16_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <4 x half>] %in, 0
+  ret <4 x half> %val
+}
+
+define <8 x i8> @test_v8i8_blocked([7 x double], [2 x <8 x i8>] %in) {
+; CHECK-LABEL: test_v8i8_blocked:
+; CHECK: ldr d0, [sp]
+  %val = extractvalue [2 x <8 x i8>] %in, 0
+  ret <8 x i8> %val
+}
+
+define <2 x i64> @test_v2i64_blocked([7 x double], [2 x <2 x i64>] %in) {
+; CHECK-LABEL: test_v2i64_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <2 x i64>] %in, 0
+  ret <2 x i64> %val
+}
+
+define <2 x double> @test_v2f64_blocked([7 x double], [2 x <2 x double>] %in) {
+; CHECK-LABEL: test_v2f64_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <2 x double>] %in, 0
+  ret <2 x double> %val
+}
+
+define <4 x i32> @test_v4i32_blocked([7 x double], [2 x <4 x i32>] %in) {
+; CHECK-LABEL: test_v4i32_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <4 x i32>] %in, 0
+  ret <4 x i32> %val
+}
+
+define <4 x float> @test_v4f32_blocked([7 x double], [2 x <4 x float>] %in) {
+; CHECK-LABEL: test_v4f32_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <4 x float>] %in, 0
+  ret <4 x float> %val
+}
+
+define <8 x i16> @test_v8i16_blocked([7 x double], [2 x <8 x i16>] %in) {
+; CHECK-LABEL: test_v8i16_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <8 x i16>] %in, 0
+  ret <8 x i16> %val
+}
+
+define <8 x half> @test_v8f16_blocked([7 x double], [2 x <8 x half>] %in) {
+; CHECK-LABEL: test_v8f16_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <8 x half>] %in, 0
+  ret <8 x half> %val
+}
+
+define <16 x i8> @test_v16i8_blocked([7 x double], [2 x <16 x i8>] %in) {
+; CHECK-LABEL: test_v16i8_blocked:
+; CHECK: ldr q0, [sp]
+  %val = extractvalue [2 x <16 x i8>] %in, 0
+  ret <16 x i8> %val
+}
-- 
2.7.4