public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r10-6374] Fix TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL handling.
@ 2020-01-31 15:45 Uros Bizjak
0 siblings, 0 replies; only message in thread
From: Uros Bizjak @ 2020-01-31 15:45 UTC (permalink / raw)
To: gcc-cvs
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="us-ascii", Size: 18620 bytes --]
https://gcc.gnu.org/g:828573a573553e9c38f0431f0998de8c905f8460
commit r10-6374-g828573a573553e9c38f0431f0998de8c905f8460
Author: Uros Bizjak <ubizjak@gmail.com>
Date: Fri Jan 31 16:44:36 2020 +0100
Fix TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL handling.
The reason for TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL on AMD target is
only insn size, as advised in e.g. Software Optimization Guide for the
AMD Family 15h Processors [1], section 7.1.2, where it is said:
--quote--
7.1.2 Reduce Instruction SizeOptimization
Reduce the size of instructions when possible.
Rationale
Using smaller instruction sizes improves instruction fetch throughput.
Specific examples include the following:
*In SIMD code, use the single-precision (PS) form of instructions
instead of the double-precision (PD) form. For example, for register
to register moves, MOVAPS achieves the same result as MOVAPD, but uses
one less byte to encode the instruction and has no prefix byte. Other
examples in which single-precision forms can be substituted for
double-precision forms include MOVUPS, MOVNTPS, XORPS, ORPS, ANDPS,
and SHUFPS.
...
--/quote--
Please note that this optimization applies only to non-AVX forms, as
demonstrated by:
0: 0f 28 c8 movaps %xmm0,%xmm1
3: 66 0f 28 c8 movapd %xmm0,%xmm1
7: c5 f8 28 d1 vmovaps %xmm1,%xmm2
b: c5 f9 28 d1 vmovapd %xmm1,%xmm2
Also note that MOVDQA is missing in the above optimization. It is
harmful to substitute MOVDQA with MOVAPS, as it can (and does)
introduce +1 cycle forwarding penalty between FLT (FPA/FPM) and INT
(VALU) FP clusters.
[1] https://www.amd.com/system/files/TechDocs/47414_15h_sw_opt_guide.pdf
Diff:
---
gcc/ChangeLog | 29 +++++++++++++++
gcc/config/i386/i386.md | 65 +++++++++++++-------------------
gcc/config/i386/mmx.md | 6 +--
gcc/config/i386/sse.md | 88 +++++++++++++++++++-------------------------
gcc/config/i386/x86-tune.def | 8 ++--
5 files changed, 98 insertions(+), 98 deletions(-)
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 78a8310..0e99b72 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,32 @@
+2020-01-31 Uroš Bizjak <ubizjak@gmail.com>
+
+ * config/i386/i386.md (*movoi_internal_avx): Do not check for
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL. Remove MODE_V8SF handling.
+ (*movti_internal): Do not check for
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL.
+ (*movtf_internal): Move check for TARGET_SSE2 and size optimization
+ just after check for TARGET_AVX.
+ (*movdf_internal): Ditto.
+ * config/i386/mmx.md (*mov<mode>_internal): Do not check for
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL.
+ * config/i386/sse.md (mov<mode>_internal): Only check
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL with V2DFmode. Move check
+ for TARGET_SSE2 and size optimization just after check for TARGET_AVX.
+ (<sse>_andnot<mode>3<mask_name>): Move check for
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL after check for TARGET_AVX.
+ (<code><mode>3<mask_name>): Ditto.
+ (*andnot<mode>3): Ditto.
+ (*andnottf3): Ditto.
+ (*<code><mode>3): Ditto.
+ (*<code>tf3): Ditto.
+ (*andnot<VI:mode>3): Remove
+ TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL handling.
+ (<mask_codefor><code><VI48_AVX_AVX512F:mode>3<mask_name>): Ditto.
+ (*<code><VI12_AVX_AVX512F:mode>3): Ditto.
+ (sse4_1_blendv<ssemodesuffix>): Ditto.
+ * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL):
+ Explain that tune applies to 128bit instructions only.
+
2020-01-31 Kwok Cheung Yeung <kcy@codesourcery.com>
* config/gcn/mkoffload.c (process_asm): Add sgpr_count and vgpr_count
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index f5c8d55..46b442d 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -1949,18 +1949,14 @@
if (misaligned_operand (operands[0], OImode)
|| misaligned_operand (operands[1], OImode))
{
- if (get_attr_mode (insn) == MODE_V8SF)
- return "vmovups\t{%1, %0|%0, %1}";
- else if (get_attr_mode (insn) == MODE_XI)
+ if (get_attr_mode (insn) == MODE_XI)
return "vmovdqu32\t{%1, %0|%0, %1}";
else
return "vmovdqu\t{%1, %0|%0, %1}";
}
else
{
- if (get_attr_mode (insn) == MODE_V8SF)
- return "vmovaps\t{%1, %0|%0, %1}";
- else if (get_attr_mode (insn) == MODE_XI)
+ if (get_attr_mode (insn) == MODE_XI)
return "vmovdqa32\t{%1, %0|%0, %1}";
else
return "vmovdqa\t{%1, %0|%0, %1}";
@@ -1980,8 +1976,6 @@
(and (eq_attr "alternative" "1")
(match_test "TARGET_AVX512VL"))
(const_string "XI")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (const_string "V8SF")
]
(const_string "OI")))])
@@ -2060,11 +2054,10 @@
(match_test "TARGET_AVX")
(const_string "TI")
(ior (not (match_test "TARGET_SSE2"))
- (ior (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (and (eq_attr "alternative" "5")
- (match_test "TARGET_SSE_TYPELESS_STORES"))))
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
- (match_test "optimize_function_for_size_p (cfun)")
+ (and (eq_attr "alternative" "5")
+ (match_test "TARGET_SSE_TYPELESS_STORES"))
(const_string "V4SF")
]
(const_string "TI")))
@@ -2243,12 +2236,10 @@
(cond [(ior (match_operand 0 "ext_sse_reg_operand")
(match_operand 1 "ext_sse_reg_operand"))
(const_string "TI")
- (ior (not (match_test "TARGET_SSE2"))
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "TI")
- (match_test "optimize_function_for_size_p (cfun)")
+ (ior (not (match_test "TARGET_SSE2"))
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
]
(const_string "TI"))
@@ -2453,12 +2444,10 @@
(cond [(ior (match_operand 0 "ext_sse_reg_operand")
(match_operand 1 "ext_sse_reg_operand"))
(const_string "XI")
- (ior (not (match_test "TARGET_SSE2"))
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "TI")
- (match_test "optimize_function_for_size_p (cfun)")
+ (ior (not (match_test "TARGET_SSE2"))
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
]
(const_string "TI"))
@@ -3324,14 +3313,14 @@
(const_string "DI")
(match_test "TARGET_AVX")
(const_string "TI")
+ (ior (not (match_test "TARGET_SSE2"))
+ (match_test "optimize_function_for_size_p (cfun)"))
+ (const_string "V4SF")
(match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
(const_string "V4SF")
(and (eq_attr "alternative" "2")
(match_test "TARGET_SSE_TYPELESS_STORES"))
(const_string "V4SF")
- (ior (not (match_test "TARGET_SSE2"))
- (match_test "optimize_function_for_size_p (cfun)"))
- (const_string "V4SF")
]
(const_string "TI")))])
@@ -3541,14 +3530,13 @@
/* xorps is one byte shorter for non-AVX targets. */
(eq_attr "alternative" "12,16")
- (cond [(not (match_test "TARGET_SSE2"))
- (const_string "V4SF")
- (and (match_test "TARGET_AVX512F")
- (not (match_test "TARGET_PREFER_AVX256")))
+ (cond [(and (match_test "TARGET_AVX512F")
+ (not (match_test "TARGET_PREFER_AVX256")))
(const_string "XI")
(match_test "TARGET_AVX")
(const_string "V2DF")
- (match_test "optimize_function_for_size_p (cfun)")
+ (ior (not (match_test "TARGET_SSE2"))
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
(match_test "TARGET_SSE_LOAD0_BY_PXOR")
(const_string "TI")
@@ -3566,15 +3554,15 @@
(ior (match_operand 0 "ext_sse_reg_operand")
(match_operand 1 "ext_sse_reg_operand")))
(const_string "V8DF")
+ (match_test "TARGET_AVX")
+ (const_string "DF")
(ior (not (match_test "TARGET_SSE2"))
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
+ (match_test "optimize_function_for_size_p (cfun)"))
+ (const_string "V4SF")
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
(const_string "V4SF")
(match_test "TARGET_SSE_PARTIAL_REG_DEPENDENCY")
(const_string "V2DF")
- (match_test "TARGET_AVX")
- (const_string "DF")
- (match_test "optimize_function_for_size_p (cfun)")
- (const_string "V4SF")
]
(const_string "DF"))
@@ -3723,16 +3711,15 @@
(eq_attr "alternative" "11")
(const_string "DI")
(eq_attr "alternative" "5")
- (cond [(not (match_test "TARGET_SSE2"))
- (const_string "V4SF")
- (and (match_test "TARGET_AVX512F")
- (not (match_test "TARGET_PREFER_AVX256")))
+ (cond [(and (match_test "TARGET_AVX512F")
+ (not (match_test "TARGET_PREFER_AVX256")))
(const_string "V16SF")
(match_test "TARGET_AVX")
(const_string "V4SF")
- (match_test "optimize_function_for_size_p (cfun)")
+ (ior (not (match_test "TARGET_SSE2"))
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
- (match_test "TARGET_SSE_LOAD0_BY_PXOR")
+ (match_test "TARGET_SSE_LOAD0_BY_PXOR")
(const_string "TI")
]
(const_string "V4SF"))
diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index d06e3b1..f695831 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -195,11 +195,7 @@
(match_test "<MODE>mode == V2SFmode")
(const_string "V4SF")
(ior (not (match_test "TARGET_SSE2"))
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "V4SF")
- (match_test "TARGET_AVX")
- (const_string "TI")
- (match_test "optimize_function_for_size_p (cfun)")
+ (match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
]
(const_string "TI"))
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index f5ff2e9..46f00e3 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -1118,13 +1118,15 @@
(const_string "<sseinsnmode>")
(match_test "TARGET_AVX")
(const_string "<sseinsnmode>")
- (ior (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (and (eq_attr "alternative" "3")
- (match_test "TARGET_SSE_TYPELESS_STORES")))
- (const_string "<ssePSmode>")
(ior (not (match_test "TARGET_SSE2"))
(match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
+ (and (match_test "<MODE>mode == V2DFmode")
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
+ (const_string "V4SF")
+ (and (eq_attr "alternative" "3")
+ (match_test "TARGET_SSE_TYPELESS_STORES"))
+ (const_string "V4SF")
(and (eq_attr "alternative" "0")
(match_test "TARGET_SSE_LOAD0_BY_PXOR"))
(const_string "TI")
@@ -3555,16 +3557,14 @@
(const_string "<sseintvecmode2>")
(eq_attr "alternative" "3")
(const_string "<sseintvecmode2>")
- (and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "<ssePSmode>")
(match_test "TARGET_AVX")
(const_string "<MODE>")
(match_test "optimize_function_for_size_p (cfun)")
(const_string "V4SF")
- ]
- (const_string "<MODE>")))])
-
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "<MODE>")))])
(define_insn "<sse>_andnot<mode>3<mask_name>"
[(set (match_operand:VF_512 0 "register_operand" "=v")
@@ -3673,15 +3673,14 @@
(const_string "<sseintvecmode2>")
(eq_attr "alternative" "3")
(const_string "<sseintvecmode2>")
- (and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "<ssePSmode>")
(match_test "TARGET_AVX")
(const_string "<MODE>")
(match_test "optimize_function_for_size_p (cfun)")
(const_string "V4SF")
- ]
- (const_string "<MODE>")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "<MODE>")))])
(define_insn "*<code><mode>3<mask_name>"
[(set (match_operand:VF_512 0 "register_operand" "=v")
@@ -3822,15 +3821,14 @@
(if_then_else (match_test "TARGET_AVX512DQ")
(const_string "<avx512fvecmode>")
(const_string "XI"))
- (and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "<ssevecmode>")
(match_test "optimize_function_for_size_p (cfun)")
(const_string "V4SF")
- ]
- (const_string "<ssevecmode>")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "<ssevecmode>")))])
(define_insn "*andnottf3"
[(set (match_operand:TF 0 "register_operand" "=x,x,v,v")
@@ -3879,15 +3877,15 @@
(const_string "TI")
(eq_attr "alternative" "3")
(const_string "XI")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "TI")
(ior (not (match_test "TARGET_SSE2"))
(match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
- ]
- (const_string "TI")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "TI")))])
(define_insn "*<code><mode>3"
[(set (match_operand:MODEF 0 "register_operand" "=x,x,v,v")
@@ -3946,15 +3944,14 @@
(if_then_else (match_test "TARGET_AVX512DQ")
(const_string "<avx512fvecmode>")
(const_string "XI"))
- (and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "<ssevecmode>")
(match_test "optimize_function_for_size_p (cfun)")
(const_string "V4SF")
- ]
- (const_string "<ssevecmode>")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "<ssevecmode>")))])
(define_expand "<code>tf3"
[(set (match_operand:TF 0 "register_operand")
@@ -4011,15 +4008,15 @@
(const_string "TI")
(eq_attr "alternative" "3")
(const_string "QI")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (const_string "V4SF")
(match_test "TARGET_AVX")
(const_string "TI")
(ior (not (match_test "TARGET_SSE2"))
(match_test "optimize_function_for_size_p (cfun)"))
(const_string "V4SF")
- ]
- (const_string "TI")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "TI")))])
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;
@@ -13007,10 +13004,7 @@
(const_string "*")))
(set_attr "prefix" "orig,vex,evex")
(set (attr "mode")
- (cond [(and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "<ssePSmode>")
- (match_test "TARGET_AVX2")
+ (cond [(match_test "TARGET_AVX2")
(const_string "<sseinsnmode>")
(match_test "TARGET_AVX")
(if_then_else
@@ -13148,10 +13142,7 @@
(const_string "*")))
(set_attr "prefix" "<mask_prefix3>,evex")
(set (attr "mode")
- (cond [(and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "<ssePSmode>")
- (match_test "TARGET_AVX2")
+ (cond [(match_test "TARGET_AVX2")
(const_string "<sseinsnmode>")
(match_test "TARGET_AVX")
(if_then_else
@@ -13244,10 +13235,7 @@
(const_string "*")))
(set_attr "prefix" "orig,vex,evex")
(set (attr "mode")
- (cond [(and (match_test "<MODE_SIZE> == 16")
- (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL"))
- (const_string "<ssePSmode>")
- (match_test "TARGET_AVX2")
+ (cond [(match_test "TARGET_AVX2")
(const_string "<sseinsnmode>")
(match_test "TARGET_AVX")
(if_then_else
@@ -17190,14 +17178,14 @@
(set_attr "prefix" "orig,orig,vex")
(set_attr "btver2_decode" "vector,vector,vector")
(set (attr "mode")
- (cond [(match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
- (const_string "V4SF")
- (match_test "TARGET_AVX")
+ (cond [(match_test "TARGET_AVX")
(const_string "<ssevecmode>")
(match_test "optimize_function_for_size_p (cfun)")
(const_string "V4SF")
- ]
- (const_string "<ssevecmode>")))])
+ (match_test "TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL")
+ (const_string "V4SF")
+ ]
+ (const_string "<ssevecmode>")))])
(define_insn_and_split "*<sse4_1>_blendv<ssemodesuffix><avxsizesuffix>_lt"
[(set (match_operand:VF_128_256 0 "register_operand" "=Yr,*x,x")
diff --git a/gcc/config/i386/x86-tune.def b/gcc/config/i386/x86-tune.def
index 2125474..41b3d52 100644
--- a/gcc/config/i386/x86-tune.def
+++ b/gcc/config/i386/x86-tune.def
@@ -366,15 +366,15 @@ DEF_TUNE (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL, "sse_unaligned_load_optimal",
| m_INTEL | m_GOLDMONT | m_GOLDMONT_PLUS
| m_TREMONT | m_AMDFAM10 | m_BDVER | m_BTVER | m_ZNVER | m_GENERIC)
-/* X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL: Use movups for misaligned stores instead
- of a sequence loading registers by parts. */
+/* X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL: Use movups for misaligned stores
+ instead of a sequence loading registers by parts. */
DEF_TUNE (X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL, "sse_unaligned_store_optimal",
m_NEHALEM | m_SANDYBRIDGE | m_CORE_AVX2 | m_SILVERMONT | m_KNL | m_KNM
| m_INTEL | m_GOLDMONT | m_GOLDMONT_PLUS
| m_TREMONT | m_BDVER | m_ZNVER | m_GENERIC)
-/* Use packed single precision instructions where posisble. I.e. movups instead
- of movupd. */
+/* X86_TUNE_SSE_PACKED_SINGLE_INSN_OPTIMAL: Use packed single
+ precision 128bit instructions instead of double where possible. */
DEF_TUNE (X86_TUNE_SSE_PACKED_SINGLE_INSN_OPTIMAL, "sse_packed_single_insn_optimal",
m_BDVER | m_ZNVER)
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2020-01-31 15:45 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-01-31 15:45 [gcc r10-6374] Fix TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL handling Uros Bizjak
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).