From 8a85bce215eda0fa56bf67186d0fd487954185f2 Mon Sep 17 00:00:00 2001 From: Ben Shi Date: Fri, 18 May 2018 07:19:17 +0000 Subject: [PATCH] test/codegen: improve test cases for arm64 1. Some incorrect test cases are disabled. 2. Some wrong test cases are corrected. 3. Some new test cases are added. Change-Id: Ib5d0473d55159f233ddab79f96967eaec7b08597 Reviewed-on: https://go-review.googlesource.com/113736 Reviewed-by: Cherry Zhang Run-TryBot: Cherry Zhang TryBot-Result: Gobot Gobot --- test/codegen/memcombine.go | 116 ++++++++++++++++++++++++++++++++----- 1 file changed, 103 insertions(+), 13 deletions(-) diff --git a/test/codegen/memcombine.go b/test/codegen/memcombine.go index 908da16fd8..4008fdb2cb 100644 --- a/test/codegen/memcombine.go +++ b/test/codegen/memcombine.go @@ -99,38 +99,108 @@ func load_be32_idx(b []byte, idx int) { func load_be16(b []byte) { // amd64:`ROLW\s\$8` - // arm64: `REV16W`,`MOVHU\s\(R[0-9]+\),`,-`MOVB` + // arm64:`REV16W`,`MOVHU\s\(R[0-9]+\),`,-`MOVB` // ppc64le:`MOVHBR` sink16 = binary.BigEndian.Uint16(b) } func load_be16_idx(b []byte, idx int) { // amd64:`ROLW\s\$8` - // arm64: `REV16W`,`MOVHU\s\(R[0-9]+\)\(R[0-9]+\),`,-`MOVB` + // arm64:`REV16W`,`MOVHU\s\(R[0-9]+\)\(R[0-9]+\),`,-`MOVB` // ppc64le:`MOVHBR` sink16 = binary.BigEndian.Uint16(b[idx:]) } -func load_byte2_uint16(s []byte) uint16 { - // arm64:`MOVHU\t\(R[0-9]+\)`,-`ORR\tR[0-9]+<<8` +func load_le_byte2_uint16(s []byte) uint16 { + // arm64:`MOVHU\t\(R[0-9]+\)`,-`ORR`,-`MOVB` return uint16(s[0]) | uint16(s[1])<<8 } -func load_byte2_uint16_idx(s []byte, idx int) uint16 { +func load_le_byte4_uint32(s []byte) uint32 { + // arm64:`MOVWU\t\(R[0-9]+\)`,-`ORR`,-`MOV[BH]` + return uint32(s[0]) | uint32(s[1])<<8 | uint32(s[2])<<16 | uint32(s[3])<<24 +} + +func load_le_byte8_uint64(s []byte) uint64 { + // arm64:`MOVD\t\(R[0-9]+\)`,-`ORR`,-`MOV[BHW]` + return uint64(s[0]) | uint64(s[1])<<8 | uint64(s[2])<<16 | uint64(s[3])<<24 | uint64(s[4])<<32 | uint64(s[5])<<40 | uint64(s[6])<<48 | uint64(s[7])<<56 +} + +func load_be_byte2_uint16(s []byte) uint16 { + // arm64:`MOVHU\t\(R[0-9]+\)`,`REV16W`,-`ORR`,-`MOVB` + return uint16(s[0])<<8 | uint16(s[1]) +} + +func load_be_byte4_uint32(s []byte) uint32 { + // arm64:`MOVWU\t\(R[0-9]+\)`,`REVW`,-`ORR`,-`REV16W`,-`MOV[BH]` + return uint32(s[0])<<24 | uint32(s[1])<<16 | uint32(s[2])<<8 | uint32(s[3]) +} + +func load_be_byte8_uint64(s []byte) uint64 { + // arm64:`MOVD\t\(R[0-9]+\)`,`REV`,-`ORR`,-`REVW`,-`REV16W`,-`MOV[BHW]` + return uint64(s[0])<<56 | uint64(s[1])<<48 | uint64(s[2])<<40 | uint64(s[3])<<32 | uint64(s[4])<<24 | uint64(s[5])<<16 | uint64(s[6])<<8 | uint64(s[7]) +} + +func load_le_byte2_uint16_idx(s []byte, idx int) uint16 { + // arm64:`MOVHU\s\(R[0-9]+\)\(R[0-9]+\)`,-`ORR`,-`MOVB` + return uint16(s[idx]) | uint16(s[idx+1])<<8 +} + +func load_le_byte4_uint32_idx(s []byte, idx int) uint32 { + // arm64:`MOVWU\s\(R[0-9]+\)\(R[0-9]+\)`,-`ORR`,-`MOV[BH]` + return uint32(s[idx]) | uint32(s[idx+1])<<8 | uint32(s[idx+2])<<16 | uint32(s[idx+3])<<24 +} + +func load_le_byte8_uint64_idx(s []byte, idx int) uint64 { + // arm64:`MOVD\s\(R[0-9]+\)\(R[0-9]+\)`,-`ORR`,-`MOV[BHW]` + return uint64(s[idx]) | uint64(s[idx+1])<<8 | uint64(s[idx+2])<<16 | uint64(s[idx+3])<<24 | uint64(s[idx+4])<<32 | uint64(s[idx+5])<<40 | uint64(s[idx+6])<<48 | uint64(s[idx+7])<<56 +} + +func load_be_byte2_uint16_idx(s []byte, idx int) uint16 { + // arm64:`MOVHU\s\(R[0-9]+\)\(R[0-9]+\)`,`REV16W`,-`ORR`,-`MOVB` + return uint16(s[idx])<<8 | uint16(s[idx+1]) +} + +func load_be_byte4_uint32_idx(s []byte, idx int) uint32 { + // arm64:`MOVWU\s\(R[0-9]+\)\(R[0-9]+\)`,`REVW`,-`ORR`,-`MOV[BH]`,-`REV16W` + return uint32(s[idx])<<24 | uint32(s[idx+1])<<16 | uint32(s[idx+2])<<8 | uint32(s[idx+3]) +} + +func load_be_byte8_uint64_idx(s []byte, idx int) uint64 { + // arm64:`MOVD\s\(R[0-9]+\)\(R[0-9]+\)`,`REV`,-`ORR`,-`MOV[BHW]`,-`REVW`,-`REV16W` + return uint64(s[idx])<<56 | uint64(s[idx+1])<<48 | uint64(s[idx+2])<<40 | uint64(s[idx+3])<<32 | uint64(s[idx+4])<<24 | uint64(s[idx+5])<<16 | uint64(s[idx+6])<<8 | uint64(s[idx+7]) +} + +func load_le_byte2_uint16_idx2(s []byte, idx int) uint16 { // arm64:`MOVHU\s\(R[0-9]+\)\(R[0-9]+<<1\)`,-`ORR`,-`MOVB` return uint16(s[idx<<1]) | uint16(s[(idx<<1)+1])<<8 } -func load_byte4_uint32_idx(s []byte, idx int) uint32 { +func load_le_byte4_uint32_idx4(s []byte, idx int) uint32 { // arm64:`MOVWU\s\(R[0-9]+\)\(R[0-9]+<<2\)`,-`ORR`,-`MOV[BH]` return uint32(s[idx<<2]) | uint32(s[(idx<<2)+1])<<8 | uint32(s[(idx<<2)+2])<<16 | uint32(s[(idx<<2)+3])<<24 } -func load_byte8_uint64_idx(s []byte, idx int) uint64 { +func load_le_byte8_uint64_idx8(s []byte, idx int) uint64 { // arm64:`MOVD\s\(R[0-9]+\)\(R[0-9]+<<3\)`,-`ORR`,-`MOV[BHW]` return uint64(s[idx<<3]) | uint64(s[(idx<<3)+1])<<8 | uint64(s[(idx<<3)+2])<<16 | uint64(s[(idx<<3)+3])<<24 | uint64(s[(idx<<3)+4])<<32 | uint64(s[(idx<<3)+5])<<40 | uint64(s[(idx<<3)+6])<<48 | uint64(s[(idx<<3)+7])<<56 } +func load_be_byte2_uint16_idx2(s []byte, idx int) uint16 { + // arm64:`MOVHU\s\(R[0-9]+\)\(R[0-9]+<<1\)`,`REV16W`,-`ORR`,-`MOVB` + return uint16(s[idx<<1])<<8 | uint16(s[(idx<<1)+1]) +} + +func load_be_byte4_uint32_idx4(s []byte, idx int) uint32 { + // arm64:`MOVWU\s\(R[0-9]+\)\(R[0-9]+<<2\)`,`REVW`,-`ORR`,-`MOV[BH]`,-`REV16W` + return uint32(s[idx<<2])<<24 | uint32(s[(idx<<2)+1])<<16 | uint32(s[(idx<<2)+2])<<8 | uint32(s[(idx<<2)+3]) +} + +func load_be_byte8_uint64_idx8(s []byte, idx int) uint64 { + // arm64:`MOVD\s\(R[0-9]+\)\(R[0-9]+<<3\)`,`REV`,-`ORR`,-`MOV[BHW]`,-`REVW`,-`REV16W` + return uint64(s[idx<<3])<<56 | uint64(s[(idx<<3)+1])<<48 | uint64(s[(idx<<3)+2])<<40 | uint64(s[(idx<<3)+3])<<32 | uint64(s[(idx<<3)+4])<<24 | uint64(s[(idx<<3)+5])<<16 | uint64(s[(idx<<3)+6])<<8 | uint64(s[(idx<<3)+7]) +} + // Check load combining across function calls. func fcall_byte(a, b byte) (byte, byte) { @@ -262,6 +332,18 @@ func store_be16_idx(b []byte, idx int) { binary.BigEndian.PutUint16(b[idx:], sink16) } +func store_be_byte_2_idx(b []byte, idx int, val uint16) { + _, _ = b[idx+0], b[idx+1] + // arm64:`REV16W`,`MOVH\sR[0-9]+,\s\(R[0-9]+\)\(R[0-9]+\)`,-`MOVB` + b[idx+0], b[idx+1] = byte(val>>8), byte(val) +} + +func store_be_byte_2_idx2(b []byte, idx int, val uint16) { + _, _ = b[(idx<<1)+0], b[(idx<<1)+1] + // arm64:`REV16W`,`MOVH\sR[0-9]+,\s\(R[0-9]+\)\(R[0-9]+<<1\)`,-`MOVB` + b[(idx<<1)+0], b[(idx<<1)+1] = byte(val>>8), byte(val) +} + // ------------- // // Zeroing // // ------------- // @@ -295,12 +377,7 @@ func zero_byte_16(b []byte) { b[12], b[13], b[14], b[15] = 0, 0, 0, 0 // arm64:"STP",-"MOVB",-"MOVH",-"MOVW" } -func zero_byte_2_idx(b []byte, idx int) { - // arm64: `MOVH\sZR,\s\(R[0-9]+\)\(R[0-9]+<<1\)`,-`MOVB` - b[(idx<<1)+0] = 0 - b[(idx<<1)+1] = 0 -} - +/* TODO: enable them when corresponding optimization are implemented func zero_byte_4_idx(b []byte, idx int) { // arm64: `MOVW\sZR,\s\(R[0-9]+\)\(R[0-9]+<<2\)`,-`MOV[BH]` b[(idx<<2)+0] = 0 @@ -320,6 +397,7 @@ func zero_byte_8_idx(b []byte, idx int) { b[(idx<<3)+6] = 0 b[(idx<<3)+7] = 0 } +*/ func zero_byte_30(a *[30]byte) { *a = [30]byte{} // arm64:"STP",-"MOVB",-"MOVH",-"MOVW" @@ -329,6 +407,18 @@ func zero_byte_39(a *[39]byte) { *a = [39]byte{} // arm64:"MOVD",-"MOVB",-"MOVH",-"MOVW" } +func zero_byte_2_idx(b []byte, idx int) { + _, _ = b[idx+0], b[idx+1] + // arm64:`MOVH\sZR,\s\(R[0-9]+\)\(R[0-9]+\)`,-`MOVB` + b[idx+0], b[idx+1] = 0, 0 +} + +func zero_byte_2_idx2(b []byte, idx int) { + _, _ = b[(idx<<1)+0], b[(idx<<1)+1] + // arm64:`MOVH\sZR,\s\(R[0-9]+\)\(R[0-9]+<<1\)`,-`MOVB` + b[(idx<<1)+0], b[(idx<<1)+1] = 0, 0 +} + func zero_uint16_2(h1, h2 []uint16) { _, _ = h1[1], h2[1] h1[0], h1[1] = 0, 0 // arm64:"MOVW\tZR",-"MOVB",-"MOVH"