OSDN Git Service

2010-05-14 Harsha Jagasia <harsha.jagasia@amd.com>
authorhjagasia <hjagasia@138bc75d-0d04-0410-961f-82ee72b054a4>
Fri, 14 May 2010 17:35:11 +0000 (17:35 +0000)
committerhjagasia <hjagasia@138bc75d-0d04-0410-961f-82ee72b054a4>
Fri, 14 May 2010 17:35:11 +0000 (17:35 +0000)
commit6fc76bb02fac69d05787ea749a9cf8230b59afed
treea6abb15b279401560c616e9b447f8a29e3c9704d
parent3a4137294458e201f42049e4696ee86d1fd7cb4a
2010-05-14  Harsha Jagasia  <harsha.jagasia@amd.com>

* config.gcc: Add support for --with-cpu option for bdver1.
* config/i386/i386.h (TARGET_BDVER1): New macro.
(ix86_tune_indices): Change SSE_UNALIGNED_MOVE_OPTIMAL
to SSE_UNALIGNED_LOAD_OPTIMAL. Add SSE_UNALIGNED_STORE_OPTIMAL.
(ix86_tune_features) :Change SSE_UNALIGNED_MOVE_OPTIMAL
to SSE_UNALIGNED_LOAD_OPTIMAL. Add SSE_UNALIGNED_STORE_OPTIMAL.
Add SSE_PACKED_SINGLE_INSN_OPTIMAL.
(TARGET_CPU_DEFAULT_NAMES): Add bdver1.
(processor_type): Add PROCESSOR_BDVER1.
* config/i386/i386.md: Add bdver1 as a new cpu attribute to match
processor_type in config/i386/i386.h.
Add check for TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL to emit
movaps <reg, reg> instead of movapd <reg, reg> when replacing
movsd <reg, reg> or movss <reg, reg> for SSE and AVX.
Add check for  TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL
to emit packed xor instead of packed double/packed integer
xor for SSE and AVX when moving a zero value.
* config/i386/sse.md: Add check for TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL
 to emit movaps instead of movapd/movdqa for SSE and AVX.
Add check for TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL to emit packed single
logical operations i.e and, or and xor instead of packed double logical
operations for SSE and AVX.
* config/i386/i386-c.c:
(ix86_target_macros_internal): Add PROCESSOR_BDVER1.
* config/i386/driver-i386.c: Turn on -mtune=native for BDVER1.
(has_fma4, has_xop): New.
* config/i386/i386.c (bdver1_cost): New variable.
(m_BDVER1): New macro.
(m_AMD_MULTIPLE): Add m_BDVER1.
(x86_tune_use_leave, x86_tune_push_memory, x86_tune_unroll_strlen,
 x86_tune_deep_branch_prediction, x86_tune_use_sahf, x86_tune_movx,
 x86_tune_use_simode_fiop, x86_tune_promote_qimode,
 x86_tune_add_esp_8, x86_tune_tune_sub_esp_4, x86_tune_sub_esp_8,
 x86_tune_integer_dfmode_moves, x86_tune_partial_reg_dependency,
 x86_tune_sse_partial_reg_dependency, x86_tune_sse_unaligned_load_optimal,
 x86_tune_sse_unaligned_store_optimal, x86_tune_sse_typeless_stores,
 x86_tune_memory_mismatch_stall, x86_tune_use_ffreep,
 x86_tune_inter_unit_moves, x86_tune_inter_unit_conversions,
 x86_tune_use_bt, x86_tune_pad_returns, x86_tune_slow_imul_imm32_mem,
 x86_tune_slow_imul_imm8, x86_tune_fuse_cmp_and_branch):
Enable/disable for bdver1.
(processor_target_table): Add bdver1_cost.
(cpu_names): Add bdver1.
(override_options): Set up PROCESSOR_BDVER1 for bdver1 entry in
 processor_alias_table.
(ix86_expand_vector_move_misalign): Change
 TARGET_SSE_UNALIGNED_MOVE_OPTIMAL to TARGET_SSE_UNALIGNED_LOAD_OPTIMAL.
 Check for TARGET_SSE_UNALIGNED_STORE_OPTIMAL.
 Check for TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL to emit movups instead
 of movupd/movdqu for SSE and AVX.
(ix86_tune_issue_rate): Add PROCESSOR_BDVER1.
(ix86_tune_adjust_cost): Add code for bdver1.
(standard_sse_constant_opcode): Add check for
TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL to emit packed single xor instead
of packed double xor for SSE and AVX.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@159399 138bc75d-0d04-0410-961f-82ee72b054a4
gcc/ChangeLog
gcc/config.gcc
gcc/config/i386/driver-i386.c
gcc/config/i386/i386-c.c
gcc/config/i386/i386.c
gcc/config/i386/i386.h
gcc/config/i386/i386.md
gcc/config/i386/sse.md