File dxvkavx-fixed.patch of Package mingw64-gcc
From 02840e72b20cea4248485462245d4fd3fbe4f385 Mon Sep 17 00:00:00 2001
From: Kai Tietz <ktietz@anaconda.com>
Date: Wed, 21 Apr 2021 07:54:59 +0200
Subject: [PATCH 15/16] add -m(no-)align-vector-insn option for i386
---
gcc/config/i386/i386-options.cc | 9 +++++--
gcc/config/i386/i386.opt | 8 ++++++
gcc/config/i386/mingw32.h | 2 +-
gcc/config/i386/predicates.md | 2 +-
gcc/config/i386/sse.md | 43 +++++++++++++++++++++++++++++----
5 files changed, 55 insertions(+), 9 deletions(-)
diff --git a/gcc/config/i386/i386-options.cc b/gcc/config/i386/i386-options.cc
index acb2291e70f..2de64ed5cde 100644
--- a/gcc/config/i386/i386-options.cc
+++ b/gcc/config/i386/i386-options.cc
@@ -386,6 +386,7 @@ ix86_target_string (HOST_WIDE_INT isa, HOST_WIDE_INT isa2,
{ "-minline-stringops-dynamically", MASK_INLINE_STRINGOPS_DYNAMICALLY },
{ "-mms-bitfields", MASK_MS_BITFIELD_LAYOUT },
{ "-mno-align-stringops", MASK_NO_ALIGN_STRINGOPS },
+ { "-mno-align-vector-insn", MASK_NO_ALIGN_VECTOR_INSN },
{ "-mno-fancy-math-387", MASK_NO_FANCY_MATH_387 },
{ "-mno-push-args", MASK_NO_PUSH_ARGS },
{ "-mno-red-zone", MASK_NO_RED_ZONE },
@@ -1103,8 +1104,12 @@ ix86_valid_target_attribute_inner_p (tree fndecl, tree args, char *p_strings[],
MASK_INLINE_STRINGOPS_DYNAMICALLY),
IX86_ATTR_NO ("align-stringops",
- OPT_mno_align_stringops,
- MASK_NO_ALIGN_STRINGOPS),
+ OPT_mno_align_stringops,
+ MASK_NO_ALIGN_STRINGOPS),
+
+ IX86_ATTR_NO ("align-vector-insn",
+ OPT_mno_align_vector_insn,
+ MASK_NO_ALIGN_VECTOR_INSN),
IX86_ATTR_YES ("recip",
OPT_mrecip,
diff --git a/gcc/config/i386/i386.opt b/gcc/config/i386/i386.opt
index 0dbaacb57ed..3b1a644e2b6 100644
--- a/gcc/config/i386/i386.opt
+++ b/gcc/config/i386/i386.opt
@@ -241,6 +241,10 @@ malign-stringops
Target RejectNegative InverseMask(NO_ALIGN_STRINGOPS, ALIGN_STRINGOPS) Save
Align destination of the string operations.
+malign-vector-insn
+Target RejectNegative InverseMask(NO_ALIGN_VECTOR_INSN, ALIGN_VECTOR_INSN) Save
+Use aligned vector instruction
+
malign-data=
Target RejectNegative Joined Var(ix86_align_data_type) Enum(ix86_align_data) Init(ix86_align_data_type_compat)
Use the given data alignment.
@@ -423,6 +424,10 @@ mpc80
Target RejectNegative
Set 80387 floating-point precision to 80-bit.
+mno-align-vector-insn
+Target Mask(NO_ALIGN_VECTOR_INSN) Save
+Uses unaligned over aligned vector instruction memonics
+
mdaz-ftz
Target
Set the FTZ and DAZ Flags.
diff --git a/gcc/config/i386/mingw32.h b/gcc/config/i386/mingw32.h
index baf80400862..2ff2a99e31b 100644
--- a/gcc/config/i386/mingw32.h
+++ b/gcc/config/i386/mingw32.h
@@ -30,7 +30,7 @@ along with GCC; see the file COPYING3. If not see
#define TARGET_SUBTARGET_DEFAULT \
(MASK_80387 | MASK_IEEE_FP | MASK_FLOAT_RETURNS \
| MASK_STACK_PROBE | MASK_ALIGN_DOUBLE \
- | MASK_MS_BITFIELD_LAYOUT)
+ | MASK_MS_BITFIELD_LAYOUT | MASK_NO_ALIGN_VECTOR_INSN)
#ifndef TARGET_USING_MCFGTHREAD
#define TARGET_USING_MCFGTHREAD 0
diff --git a/gcc/config/i386/predicates.md b/gcc/config/i386/predicates.md
index c4141a96735..0ee5a9054ee 100644
--- a/gcc/config/i386/predicates.md
+++ b/gcc/config/i386/predicates.md
@@ -1644,7 +1644,7 @@
;; less than its natural alignment.
(define_predicate "misaligned_operand"
(and (match_code "mem")
- (match_test "MEM_ALIGN (op) < GET_MODE_BITSIZE (mode)")))
+ (match_test "TARGET_NO_ALIGN_VECTOR_INSN || MEM_ALIGN (op) < GET_MODE_BITSIZE (mode)")))
;; Return true if OP is a parallel for an mov{d,q,dqa,ps,pd} vec_select,
;; where one of the two operands of the vec_concat is const0_operand.
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 076064f97e6..b224b578676 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -1749,7 +1749,8 @@
(vec_concat:V2DF (vec_select:DF (match_dup 2)
(parallel [(const_int 0)]))
(match_operand:DF 3 "memory_operand")))]
- "TARGET_SSE2 && TARGET_SSE_UNALIGNED_LOAD_OPTIMAL
+ "TARGET_SSE2
+ && (TARGET_SSE_UNALIGNED_LOAD_OPTIMAL || TARGET_NO_ALIGN_VECTOR_INSN)
&& ix86_operands_ok_for_move_multiple (operands, true, DFmode)"
[(set (match_dup 2) (match_dup 5))]
"operands[5] = adjust_address (operands[1], V2DFmode, 0);")
@@ -1760,7 +1761,8 @@
(set (match_operand:V2DF 2 "sse_reg_operand")
(vec_concat:V2DF (match_operand:DF 4 "sse_reg_operand")
(match_operand:DF 3 "memory_operand")))]
- "TARGET_SSE2 && TARGET_SSE_UNALIGNED_LOAD_OPTIMAL
+ "TARGET_SSE2
+ && (TARGET_SSE_UNALIGNED_LOAD_OPTIMAL || TARGET_NO_ALIGN_VECTOR_INSN)
&& REGNO (operands[4]) == REGNO (operands[2])
&& ix86_operands_ok_for_move_multiple (operands, true, DFmode)"
[(set (match_dup 2) (match_dup 5))]
@@ -1774,7 +1776,8 @@
(set (match_operand:DF 2 "memory_operand")
(vec_select:DF (match_operand:V2DF 3 "sse_reg_operand")
(parallel [(const_int 1)])))]
- "TARGET_SSE2 && TARGET_SSE_UNALIGNED_STORE_OPTIMAL
+ "TARGET_SSE2
+ && (TARGET_SSE_UNALIGNED_STORE_OPTIMAL || TARGET_NO_ALIGN_VECTOR_INSN)
&& ix86_operands_ok_for_move_multiple (operands, false, DFmode)"
[(set (match_dup 4) (match_dup 1))]
"operands[4] = adjust_address (operands[0], V2DFmode, 0);")
@@ -10434,7 +10437,8 @@
(vec_select:V2SF
(match_operand:V4SF 1 "nonimmediate_operand" " v,v,m")
(parallel [(const_int 0) (const_int 1)])))]
- "TARGET_SSE && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
+ "TARGET_SSE && TARGET_ALIGN_VECTOR_INSN
+ && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
"@
%vmovlps\t{%1, %0|%q0, %1}
%vmovaps\t{%1, %0|%0, %1}
@@ -10443,6 +10447,21 @@
(set_attr "prefix" "maybe_vex")
(set_attr "mode" "V2SF,V4SF,V2SF")])
+(define_insn "sse_storelps_unalign"
+ [(set (match_operand:V2SF 0 "nonimmediate_operand" "=m,v,v")
+ (vec_select:V2SF
+ (match_operand:V4SF 1 "nonimmediate_operand" " v,v,m")
+ (parallel [(const_int 0) (const_int 1)])))]
+ "TARGET_SSE && TARGET_NO_ALIGN_VECTOR_INSN
+ && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
+ "@
+ %vmovlps\t{%1, %0|%q0, %1}
+ %vmovups\t{%1, %0|%0, %1}
+ %vmovlps\t{%1, %d0|%d0, %q1}"
+ [(set_attr "type" "ssemov")
+ (set_attr "prefix" "maybe_vex")
+ (set_attr "mode" "V2SF,V4SF,V2SF")])
+
(define_expand "sse_loadlps_exp"
[(set (match_operand:V4SF 0 "nonimmediate_operand")
(vec_concat:V4SF
@@ -13338,7 +13357,7 @@
(vec_select:DF
(match_operand:V2DF 1 "nonimmediate_operand" "x,x,m")
(parallel [(const_int 0)])))]
- "!TARGET_SSE2 && TARGET_SSE
+ "!TARGET_SSE2 && TARGET_SSE && TARGET_ALIGN_VECTOR_INSN
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
"@
movlps\t{%1, %0|%0, %1}
@@ -13347,6 +13366,20 @@
[(set_attr "type" "ssemov")
(set_attr "mode" "V2SF,V4SF,V2SF")])
+(define_insn "*vec_extractv2df_0_sse_unalign"
+ [(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x")
+ (vec_select:DF
+ (match_operand:V2DF 1 "nonimmediate_operand" "x,x,m")
+ (parallel [(const_int 0)])))]
+ "!TARGET_SSE2 && TARGET_SSE && TARGET_NO_ALIGN_VECTOR_INSN
+ && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
+ "@
+ movlps\t{%1, %0|%0, %1}
+ movups\t{%1, %0|%0, %1}
+ movlps\t{%1, %0|%0, %q1}"
+ [(set_attr "type" "ssemov")
+ (set_attr "mode" "V2SF,V4SF,V2SF")])
+
(define_expand "sse2_loadhpd_exp"
[(set (match_operand:V2DF 0 "nonimmediate_operand")
(vec_concat:V2DF
--
2.38.1
