Merging upstream version 126.0.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

16 files changed, 157 insertions, 322 deletions

diff --git a/third_party/dav1d/src/x86/cdef_avx2.asm b/third_party/dav1d/src/x86/cdef_avx2.asm
index 1f30f8a3b7..95d35fc1c8 100644
--- a/third_party/dav1d/src/x86/cdef_avx2.asm
+++ b/third_party/dav1d/src/x86/cdef_avx2.asm
@@ -398,7 +398,6 @@ SECTION .text
 INIT_YMM avx2
 cglobal cdef_filter_%1x%2_8bpc, 5, 11, 0, dst, stride, left, top, bot, \
                                           pri, sec, dir, damping, edge
-%assign stack_offset_entry stack_offset
     mov          edged, edgem
     cmp          edged, 0xf
     jne .border_block
@@ -1195,9 +1194,9 @@ cglobal cdef_filter_%1x%2_8bpc, 5, 11, 0, dst, stride, left, top, bot, \
 
 .border_block:
  DEFINE_ARGS dst, stride, left, top, bot, pri, sec, stride3, dst4, edge
-%define rstk rsp
-%assign stack_offset stack_offset_entry
-%assign regs_used 11
+    RESET_STACK_STATE
+    %assign stack_offset stack_offset - (regs_used - 11) * gprsize
+    %assign regs_used 11
     ALLOC_STACK 2*16+(%2+4)*32, 16
 %define px rsp+2*16+2*32
 
diff --git a/third_party/dav1d/src/x86/filmgrain16_avx2.asm b/third_party/dav1d/src/x86/filmgrain16_avx2.asm
index a1d4c41f27..eda6035923 100644
--- a/third_party/dav1d/src/x86/filmgrain16_avx2.asm
+++ b/third_party/dav1d/src/x86/filmgrain16_avx2.asm
@@ -646,18 +646,9 @@ INIT_XMM avx2
 INIT_YMM avx2
 .ar2:
 %if WIN64
-    ; xmm6 and xmm7 already saved
-    %assign xmm_regs_used 13 + %2
     %assign stack_size_padded 136
     SUB             rsp, stack_size_padded
-    movaps   [rsp+16*2], xmm8
-    movaps   [rsp+16*3], xmm9
-    movaps   [rsp+16*4], xmm10
-    movaps   [rsp+16*5], xmm11
-    movaps   [rsp+16*6], xmm12
-%if %2
-    movaps   [rsp+16*7], xmm13
-%endif
+    WIN64_PUSH_XMM 13 + %2, 8
 %endif
     DEFINE_ARGS buf, bufy, fg_data, uv, bdmax, shift
     mov          shiftd, [fg_dataq+FGData.ar_coeff_shift]
@@ -747,20 +738,10 @@ INIT_YMM avx2
 
 .ar3:
 %if WIN64
-    ; xmm6 and xmm7 already saved
     %assign stack_offset 32
-    %assign xmm_regs_used 14 + %2
     %assign stack_size_padded 152
     SUB             rsp, stack_size_padded
-    movaps   [rsp+16*2], xmm8
-    movaps   [rsp+16*3], xmm9
-    movaps   [rsp+16*4], xmm10
-    movaps   [rsp+16*5], xmm11
-    movaps   [rsp+16*6], xmm12
-    movaps   [rsp+16*7], xmm13
-%if %2
-    movaps   [rsp+16*8], xmm14
-%endif
+    WIN64_PUSH_XMM 14 + %2, 8
 %endif
     DEFINE_ARGS buf, bufy, fg_data, uv, bdmax, shift
     mov          shiftd, [fg_dataq+FGData.ar_coeff_shift]
diff --git a/third_party/dav1d/src/x86/filmgrain16_sse.asm b/third_party/dav1d/src/x86/filmgrain16_sse.asm
index 6b0daaac0b..25d01caa19 100644
--- a/third_party/dav1d/src/x86/filmgrain16_sse.asm
+++ b/third_party/dav1d/src/x86/filmgrain16_sse.asm
@@ -275,7 +275,6 @@ cglobal generate_grain_y_16bpc, 3, 6, 8, buf, fg_data, bdmax
 
 .ar2:
 %if ARCH_X86_32
-%assign stack_offset_old stack_offset
     ALLOC_STACK -16*8
 %endif
     DEFINE_ARGS buf, fg_data, bdmax, shift
@@ -428,7 +427,6 @@ cglobal generate_grain_y_16bpc, 3, 6, 8, buf, fg_data, bdmax
 %elif ARCH_X86_64
     %define         tmp  rsp+stack_offset-72
 %else
-%assign stack_offset stack_offset_old
     ALLOC_STACK  -16*12
     %define         tmp  rsp
     mov          bdmaxd, bdmaxm
@@ -715,7 +713,6 @@ cglobal generate_grain_uv_%1_16bpc, 1, 7, 8, buf, x, pic_reg, fg_data, h
     DEFINE_ARGS buf, bufy, fg_data, uv, bdmax, shift
 %else
     DEFINE_ARGS buf, bufy, pic_reg, fg_data, uv, shift
-%assign stack_offset_old stack_offset
     ALLOC_STACK  -16*2
     mov           bufyq, r1m
     mov             uvd, r3m
@@ -831,9 +828,7 @@ cglobal generate_grain_uv_%1_16bpc, 1, 7, 8, buf, x, pic_reg, fg_data, h
 %if ARCH_X86_64
     DEFINE_ARGS buf, bufy, fg_data, uv, max, cf3, min, val3, x
 %else
-%assign stack_offset stack_offset_old
-%xdefine rstk rsp
-%assign stack_size_padded 0
+    RESET_STACK_STATE
     DEFINE_ARGS buf, shift, pic_reg, fg_data, uv, bufy, cf3
     mov           bufyq, r1m
     mov             uvd, r3m
@@ -1159,7 +1154,6 @@ cglobal generate_grain_uv_%1_16bpc, 1, 7, 8, buf, x, pic_reg, fg_data, h
 %endif
 %else
     DEFINE_ARGS buf, bufy, pic_reg, fg_data, uv, shift
-%assign stack_offset stack_offset_old
     ALLOC_STACK  -16*14
     mov           bufyq, r1m
     mov             uvd, r3m
diff --git a/third_party/dav1d/src/x86/filmgrain_avx2.asm b/third_party/dav1d/src/x86/filmgrain_avx2.asm
index 55445cf593..91d8ca5c14 100644
--- a/third_party/dav1d/src/x86/filmgrain_avx2.asm
+++ b/third_party/dav1d/src/x86/filmgrain_avx2.asm
@@ -204,18 +204,9 @@ cglobal generate_grain_y_8bpc, 2, 9, 8, buf, fg_data
 
 .ar2:
 %if WIN64
-    ; xmm6 and xmm7 already saved
-    %assign xmm_regs_used 16
     %assign stack_size_padded 168
     SUB             rsp, stack_size_padded
-    movaps   [rsp+16*2], xmm8
-    movaps   [rsp+16*3], xmm9
-    movaps   [rsp+16*4], xmm10
-    movaps   [rsp+16*5], xmm11
-    movaps   [rsp+16*6], xmm12
-    movaps   [rsp+16*7], xmm13
-    movaps   [rsp+16*8], xmm14
-    movaps   [rsp+16*9], xmm15
+    WIN64_PUSH_XMM   16, 8
 %endif
     DEFINE_ARGS buf, fg_data, h, x
     mov             r6d, [fg_dataq+FGData.ar_coeff_shift]
@@ -287,15 +278,9 @@ cglobal generate_grain_y_8bpc, 2, 9, 8, buf, fg_data
 INIT_YMM avx2
 .ar3:
 %if WIN64
-    ; xmm6 and xmm7 already saved
-    %assign stack_offset 16
     ALLOC_STACK   16*14
     %assign stack_size stack_size - 16*4
-    %assign xmm_regs_used 12
-    movaps  [rsp+16*12], xmm8
-    movaps  [rsp+16*13], xmm9
-    movaps  [rsp+16*14], xmm10
-    movaps  [rsp+16*15], xmm11
+    WIN64_PUSH_XMM   12, 8
 %else
     ALLOC_STACK   16*12
 %endif
diff --git a/third_party/dav1d/src/x86/filmgrain_sse.asm b/third_party/dav1d/src/x86/filmgrain_sse.asm
index 0172f98760..d06e349a8c 100644
--- a/third_party/dav1d/src/x86/filmgrain_sse.asm
+++ b/third_party/dav1d/src/x86/filmgrain_sse.asm
@@ -232,7 +232,6 @@ cglobal generate_grain_y_8bpc, 2, 7 + 2 * ARCH_X86_64, 16, buf, fg_data
 
 .ar2:
 %if ARCH_X86_32
-%assign stack_offset_old stack_offset
     ALLOC_STACK -16*8
 %endif
     DEFINE_ARGS buf, fg_data, shift
@@ -333,7 +332,6 @@ cglobal generate_grain_y_8bpc, 2, 7 + 2 * ARCH_X86_64, 16, buf, fg_data
 .ar3:
     DEFINE_ARGS buf, fg_data, shift
 %if ARCH_X86_32
-%assign stack_offset stack_offset_old
     ALLOC_STACK  -16*14
 %elif WIN64
     SUB             rsp, 16*6
@@ -601,7 +599,6 @@ cglobal generate_grain_uv_%1_8bpc, 1, 7 + 3 * ARCH_X86_64, 16, buf, bufy, fg_dat
     DEFINE_ARGS buf, bufy, fg_data, uv, unused, shift
     movifnidn     bufyq, bufymp
 %if ARCH_X86_32
-%assign stack_offset_old stack_offset
     ALLOC_STACK   -2*16
 %endif
     imul            uvd, 28
@@ -738,9 +735,7 @@ cglobal generate_grain_uv_%1_8bpc, 1, 7 + 3 * ARCH_X86_64, 16, buf, bufy, fg_dat
 
 .ar1:
 %if ARCH_X86_32
-%assign stack_offset stack_offset_old
-%assign stack_size_padded 0
-%xdefine rstk rsp
+    RESET_STACK_STATE
 %endif
     DEFINE_ARGS buf, bufy, fg_data, uv, val3, cf3, min, max, x
     imul            uvd, 28
@@ -881,9 +876,6 @@ cglobal generate_grain_uv_%1_8bpc, 1, 7 + 3 * ARCH_X86_64, 16, buf, bufy, fg_dat
 
 .ar2:
 %if ARCH_X86_32
-%assign stack_offset stack_offset_old
-%assign stack_size_padded 0
-%xdefine rstk rsp
     ALLOC_STACK   -8*16
 %endif
     DEFINE_ARGS buf, bufy, fg_data, uv, unused, shift
@@ -1014,9 +1006,7 @@ cglobal generate_grain_uv_%1_8bpc, 1, 7 + 3 * ARCH_X86_64, 16, buf, bufy, fg_dat
 
 .ar3:
 %if ARCH_X86_32
-%assign stack_offset stack_offset_old
-%assign stack_size_padded 0
-%xdefine rstk rsp
+    RESET_STACK_STATE
 %endif
     DEFINE_ARGS buf, bufy, fg_data, uv, unused, shift
     movifnidn     bufyq, bufymp
diff --git a/third_party/dav1d/src/x86/ipred16_avx2.asm b/third_party/dav1d/src/x86/ipred16_avx2.asm
index f4931e977b..7b52abaa10 100644
--- a/third_party/dav1d/src/x86/ipred16_avx2.asm
+++ b/third_party/dav1d/src/x86/ipred16_avx2.asm
@@ -946,7 +946,6 @@ cglobal ipred_smooth_16bpc, 3, 7, 6, dst, stride, tl, w, h, v_weights
     jg .w4_loop
     RET
 .w8:
-%assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM     12
     vpbroadcastw        m0, [tlq] ; bottom
     vbroadcasti128      m7, [tlq+hq*2+2]
@@ -974,7 +973,6 @@ cglobal ipred_smooth_16bpc, 3, 7, 6, dst, stride, tl, w, h, v_weights
     jg .w8_loop
     RET
 .w16:
-%assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM     11
     vpbroadcastw        m0, [tlq] ; bottom
     movu                m7, [tlq+hq*2+2]
@@ -1005,7 +1003,6 @@ cglobal ipred_smooth_16bpc, 3, 7, 6, dst, stride, tl, w, h, v_weights
     jg .w16_loop
     RET
 .w32:
-%assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM     15
     vpbroadcastw        m0, [tlq] ; bottom
     movu                m7, [tlq+hq*2+ 2]
@@ -1047,7 +1044,6 @@ cglobal ipred_smooth_16bpc, 3, 7, 6, dst, stride, tl, w, h, v_weights
     jg .w32_loop
     RET
 .w64:
-%assign stack_offset stack_offset - stack_size_padded
     PROLOGUE 0, 11, 16, dst, stride, tl, tl_base, h, v_weights, dummy, v_weights_base, x, y, dst_base
     mov          dst_baseq, dstq
     mov           tl_baseq, tlq
@@ -1104,7 +1100,6 @@ cglobal ipred_smooth_16bpc, 3, 7, 6, dst, stride, tl, w, h, v_weights
     RET
 
 cglobal ipred_z1_16bpc, 3, 8, 0, dst, stride, tl, w, h, angle, dx, maxbase
-    %assign org_stack_offset stack_offset
     lea                  r6, [ipred_z1_16bpc_avx2_table]
     tzcnt                wd, wm
     movifnidn        angled, anglem
@@ -1312,7 +1307,6 @@ ALIGN function_align
 .w4_end:
     RET
 .w8:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -64, 7
     lea                 r3d, [angleq+216]
     mov                 r3b, hb
@@ -1476,7 +1470,6 @@ ALIGN function_align
     or             maxbased, 16 ; imin(h+15, 31)
     jmp .w16_main
 .w16:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -96, 7
     lea            maxbased, [hq+15]
     test             angled, 0x400
@@ -1622,7 +1615,6 @@ ALIGN function_align
 .w16_end:
     RET
 .w32:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -160, 8
     lea            maxbased, [hq+31]
     mov                 r3d, 63
@@ -1737,7 +1729,6 @@ ALIGN function_align
 .w32_end:
     RET
 .w64:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -256, 10
     lea            maxbased, [hq+63]
     test             angled, 0x400
@@ -2691,7 +2682,6 @@ ALIGN function_align
     jmp .w32_filter_above
 
 cglobal ipred_z3_16bpc, 4, 9, 0, dst, stride, tl, w, h, angle, dy, org_w, maxbase
-    %assign org_stack_offset stack_offset
     lea                  r6, [ipred_z3_16bpc_avx2_table]
     tzcnt                hd, hm
     movifnidn        angled, anglem
@@ -2907,7 +2897,6 @@ ALIGN function_align
     RET
 .h8:
     lea                 r4d, [angleq+216]
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -64, 8
     mov                 r4b, wb
     lea                  r7, [strideq*3]
@@ -3155,7 +3144,6 @@ ALIGN function_align
     jmp .h16_main
 ALIGN function_align
 .h16:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -96, 10
     lea            maxbased, [wq+15]
     lea                  r7, [strideq*3]
@@ -3372,7 +3360,6 @@ ALIGN function_align
 .h16_end:
     RET
 .h32:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -160, 9
     lea            maxbased, [wq+31]
     and            maxbased, 31
@@ -3557,7 +3544,6 @@ ALIGN function_align
 .h32_end:
     RET
 .h64:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -256, 10
     lea            maxbased, [wq+63]
     test             angled, 0x400
@@ -3804,7 +3790,6 @@ ALIGN function_align
 ;           5       8           8       i
 
 cglobal ipred_filter_16bpc, 3, 9, 0, dst, stride, tl, w, h, filter
-%assign org_stack_offset stack_offset
 %define base r6-ipred_filter_16bpc_avx2_table
     lea                  r6, [filter_intra_taps]
     tzcnt                wd, wm
@@ -3846,7 +3831,6 @@ cglobal ipred_filter_16bpc, 3, 9, 0, dst, stride, tl, w, h, filter
     RET
 ALIGN function_align
 .w8:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     vbroadcasti128      m14, [base+filter_shuf3]
     vpbroadcastw        m15, r8m ; bitdepth_max
@@ -3883,7 +3867,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w16:
-    %assign stack_offset stack_offset - stack_size_padded
     ALLOC_STACK          32, 16
     vpbroadcastw        m15, r8m ; bitdepth_max
     sub                  hd, 2
@@ -3977,7 +3960,6 @@ ALIGN function_align
     ret
 ALIGN function_align
 .w32:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK          64, 16
     vpbroadcastw        m15, r8m ; bitdepth_max
     sub                  hd, 2
diff --git a/third_party/dav1d/src/x86/ipred_avx2.asm b/third_party/dav1d/src/x86/ipred_avx2.asm
index 58e40935ac..35738e7c0b 100644
--- a/third_party/dav1d/src/x86/ipred_avx2.asm
+++ b/third_party/dav1d/src/x86/ipred_avx2.asm
@@ -772,7 +772,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w32:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM       6
     movu                 m3, [tlq+1]
     punpcklbw            m2, m3, m5
@@ -823,29 +822,17 @@ ALIGN function_align
     jl .w64_loop
     RET
 
-%macro SETUP_STACK_FRAME 3 ; stack_size, regs_used, xmm_regs_used
-    %assign stack_offset 0
-    %assign stack_size_padded 0
-    %assign regs_used %2
-    %xdefine rstk rsp
-    SETUP_STACK_POINTER %1
-    %if regs_used != %2 && WIN64
-        PUSH r%2
-    %endif
-    ALLOC_STACK %1, %3
-%endmacro
-
 cglobal ipred_smooth_h_8bpc, 3, 7, 0, dst, stride, tl, w, h
-%define base r6-ipred_smooth_h_avx2_table
-    lea                  r6, [ipred_smooth_h_avx2_table]
+%define base r5-ipred_smooth_h_avx2_table
+    lea                  r5, [ipred_smooth_h_avx2_table]
     mov                  wd, wm
     vpbroadcastb         m3, [tlq+wq] ; right
     tzcnt                wd, wd
     mov                  hd, hm
-    movsxd               wq, [r6+wq*4]
+    movsxd               wq, [r5+wq*4]
     vpbroadcastd         m4, [base+pb_127_m127]
     vpbroadcastd         m5, [base+pw_128]
-    add                  wq, r6
+    add                  wq, r5
     jmp                  wq
 .w4:
     WIN64_SPILL_XMM       8
@@ -891,7 +878,6 @@ cglobal ipred_smooth_h_8bpc, 3, 7, 0, dst, stride, tl, w, h
     RET
 ALIGN function_align
 .w8:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM       8
     vbroadcasti128       m6, [base+smooth_weights+8*2]
     mova                 m7, [base+ipred_h_shuf]
@@ -927,7 +913,7 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w16:
-    SETUP_STACK_FRAME  32*4, 7, 8
+    ALLOC_STACK        32*4, 8
     lea                  r3, [rsp+64*2-4]
     call .prep ; only worthwhile for for w16 and above
     sub                 tlq, 2
@@ -951,7 +937,7 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w32:
-    SETUP_STACK_FRAME  32*4, 7, 6
+    ALLOC_STACK        32*4
     lea                  r3, [rsp+64*2-2]
     call .prep
     dec                 tlq
@@ -971,19 +957,19 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w64:
-    SETUP_STACK_FRAME  32*4, 7, 9
+    ALLOC_STACK        32*4, 9
     lea                  r3, [rsp+64*2-2]
     call .prep
-    add                  r6, smooth_weights+16*15-ipred_smooth_h_avx2_table
+    add                  r5, smooth_weights+16*15-ipred_smooth_h_avx2_table
     dec                 tlq
-    mova                xm5, [r6-16*7]
-    vinserti128          m5, [r6-16*5], 1
-    mova                xm6, [r6-16*6]
-    vinserti128          m6, [r6-16*4], 1
-    mova                xm7, [r6-16*3]
-    vinserti128          m7, [r6-16*1], 1
-    mova                xm8, [r6-16*2]
-    vinserti128          m8, [r6-16*0], 1
+    mova                xm5, [r5-16*7]
+    vinserti128          m5, [r5-16*5], 1
+    mova                xm6, [r5-16*6]
+    vinserti128          m6, [r5-16*4], 1
+    mova                xm7, [r5-16*3]
+    vinserti128          m7, [r5-16*1], 1
+    mova                xm8, [r5-16*2]
+    vinserti128          m8, [r5-16*0], 1
 .w64_loop:
     vpbroadcastb         m2, [tlq+hq]
     punpcklbw            m2, m3
@@ -1113,7 +1099,6 @@ cglobal ipred_smooth_8bpc, 3, 7, 0, dst, stride, tl, w, h, v_weights
     RET
 ALIGN function_align
 .w8:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      12
     mova                m10, [base+ipred_h_shuf]
     vbroadcasti128      m11, [base+smooth_weights+8*2]
@@ -1157,7 +1142,9 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w16:
-    SETUP_STACK_FRAME  32*4, 7, 14
+    %assign regs_used 4
+    ALLOC_STACK       -32*4, 14
+    %assign regs_used 7
     vbroadcasti128      m11, [tlq+1]
     lea                  r3, [rsp+64*2-4]
     punpcklbw           m10, m11, m0 ; top, bottom
@@ -1197,7 +1184,9 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w32:
-    SETUP_STACK_FRAME  32*4, 7, 11
+    %assign regs_used 4
+    ALLOC_STACK       -32*4, 11
+    %assign regs_used 7
     movu                 m8, [tlq+1]
     lea                  r3, [rsp+64*2-2]
     punpcklbw            m7, m8, m0
@@ -1232,7 +1221,9 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w64:
-    SETUP_STACK_FRAME  32*8, 7, 16
+    %assign regs_used 4
+    ALLOC_STACK       -32*8, 16
+    %assign regs_used 7
     movu                m13, [tlq+1 ]
     movu                m15, [tlq+33]
     add                  r6, smooth_weights+16*15-ipred_smooth_avx2_table
@@ -1316,7 +1307,6 @@ ALIGN function_align
     ret
 
 cglobal ipred_z1_8bpc, 3, 8, 0, dst, stride, tl, w, h, angle, dx, maxbase
-    %assign org_stack_offset stack_offset
     lea                  r6, [ipred_z1_avx2_table]
     tzcnt                wd, wm
     movifnidn        angled, anglem
@@ -1415,7 +1405,6 @@ ALIGN function_align
     pmovmskb            r5d, m1
     ret
 .w4_no_upsample:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -16, 11
     mov            maxbased, 7
     test             angled, 0x400 ; !enable_intra_edge_filter
@@ -1522,7 +1511,6 @@ ALIGN function_align
     mov                 r3b, hb
     cmp                 r3d, 8
     ja .w8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || h > 8
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -32, 8
     movu                xm2, [z_filter_s+6]
     mova                xm0, [tlq-1]
@@ -1592,7 +1580,6 @@ ALIGN function_align
     or             maxbased, 8 ; imin(h+7, 15)
     jmp .w8_main
 .w8_no_upsample:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -32, 10
     lea            maxbased, [hq+7]
     test             angled, 0x400
@@ -1696,7 +1683,6 @@ ALIGN function_align
     jmp .w16_main
 ALIGN function_align
 .w16:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -64, 12
     lea            maxbased, [hq+15]
     test             angled, 0x400
@@ -1816,7 +1802,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w32:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -96, 15
     lea                 r3d, [hq+31]
     mov            maxbased, 63
@@ -1960,7 +1945,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w64:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -128, 16
     lea            maxbased, [hq+63]
     test             angled, 0x400 ; !enable_intra_edge_filter
@@ -3001,7 +2985,6 @@ ALIGN function_align
     jmp .w32_filter_above
 
 cglobal ipred_z3_8bpc, 4, 9, 0, dst, stride, tl, w, h, angle, dy, org_w, maxbase
-    %assign org_stack_offset stack_offset
     lea                  r6, [ipred_z3_avx2_table]
     tzcnt                hd, hm
     movifnidn        angled, anglem
@@ -3102,7 +3085,6 @@ ALIGN function_align
     pmovmskb            r5d, m1
     ret
 .h4_no_upsample:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -16, 12
     mov            maxbased, 7
     test             angled, 0x400 ; !enable_intra_edge_filter
@@ -3215,7 +3197,6 @@ ALIGN function_align
     mov                 r4b, wb
     cmp                 r4d, 8
     ja .h8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || w > 8
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -32, 8
     and                 r4d, 4
     mova                xm0, [tlq-15]
@@ -3297,7 +3278,6 @@ ALIGN function_align
     or             maxbased, 8 ; imin(w+7, 15)
     jmp .h8_main
 .h8_no_upsample:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -32, 10
     lea            maxbased, [wq+7]
     test             angled, 0x400
@@ -3455,7 +3435,6 @@ ALIGN function_align
     jmp .h16_main
 ALIGN function_align
 .h16:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -64, 12
     lea            maxbased, [wq+15]
     test             angled, 0x400
@@ -3661,7 +3640,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .h32:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK         -96, 15
     lea            maxbased, [wq+31]
     and            maxbased, 31
@@ -3890,7 +3868,6 @@ ALIGN function_align
     RET
 ALIGN function_align
 .h64:
-    %assign stack_offset org_stack_offset
     ALLOC_STACK        -128, 16
     lea            maxbased, [wq+63]
     test             angled, 0x400 ; !enable_intra_edge_filter
@@ -4221,6 +4198,7 @@ cglobal ipred_filter_8bpc, 3, 7, 0, dst, stride, tl, w, h, filter
     movzx           filterd, byte filterm
 %endif
     shl             filterd, 6
+    WIN64_SPILL_XMM       9, 15
     add             filterq, r6
     lea                  r6, [ipred_filter_avx2_table]
     movq                xm0, [tlq-3] ; _ 6 5 0 1 2 3 4
@@ -4234,7 +4212,6 @@ cglobal ipred_filter_8bpc, 3, 7, 0, dst, stride, tl, w, h, filter
     mov                  hd, hm
     jmp                  wq
 .w4:
-    WIN64_SPILL_XMM       9
     mova                xm8, [base+filter_shuf2]
     sub                 tlq, 3
     sub                 tlq, hq
@@ -4251,8 +4228,7 @@ cglobal ipred_filter_8bpc, 3, 7, 0, dst, stride, tl, w, h, filter
     RET
 ALIGN function_align
 .w8:
-    %assign stack_offset stack_offset - stack_size_padded
-    WIN64_SPILL_XMM      10
+    WIN64_PUSH_XMM       10
     mova                 m8, [base+filter_shuf1]
     FILTER_XMM            7, 0, 6, [base+filter_shuf2]
     vpbroadcastd         m0, [tlq+4]
@@ -4278,26 +4254,18 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w16:
-%if WIN64
-    %assign stack_offset stack_offset - stack_size_padded
-    %assign xmm_regs_used 15
-    %assign stack_size_padded 0x98
-    SUB                 rsp, stack_size_padded
-%endif
     sub                  hd, 2
-    TAIL_CALL .w16_main, 0
-.w16_main:
+    call .w16_main
 %if WIN64
-    movaps       [rsp+0xa8], xmm6
-    movaps       [rsp+0xb8], xmm7
-    movaps       [rsp+0x28], xmm8
-    movaps       [rsp+0x38], xmm9
-    movaps       [rsp+0x48], xmm10
-    movaps       [rsp+0x58], xmm11
-    movaps       [rsp+0x68], xmm12
-    movaps       [rsp+0x78], xmm13
-    movaps       [rsp+0x88], xmm14
+    jmp .end
+%else
+    RET
 %endif
+.w16_main:
+    ; The spills are into the callers stack frame
+    %assign stack_size stack_size + gprsize
+    WIN64_PUSH_XMM       15, 9
+    %assign stack_size stack_size - gprsize
     FILTER_XMM           12, 0, 7, [base+filter_shuf2]
     vpbroadcastd         m0, [tlq+5]
     vpblendd             m0, [tlq-12], 0x14
@@ -4350,7 +4318,6 @@ ALIGN function_align
     ret
 ALIGN function_align
 .w32:
-    sub                 rsp, stack_size_padded
     sub                  hd, 2
     lea                  r3, [dstq+16]
     lea                 r5d, [hq-2]
@@ -4415,6 +4382,7 @@ ALIGN function_align
     shufps              xm6, xm12, xm6, q3131 ; d0 d1 d2 d3
     mova   [dstq+strideq*0], xm0
     mova   [dstq+strideq*1], xm6
+.end:
     RET
 ALIGN function_align
 .main:
diff --git a/third_party/dav1d/src/x86/ipred_sse.asm b/third_party/dav1d/src/x86/ipred_sse.asm
index 976f33a24b..f6b0cad001 100644
--- a/third_party/dav1d/src/x86/ipred_sse.asm
+++ b/third_party/dav1d/src/x86/ipred_sse.asm
@@ -670,10 +670,7 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w32:
-%if WIN64
-    movaps         [rsp+24], xmm7
-    %define xmm_regs_used 8
-%endif
+    WIN64_PUSH_XMM        8, 7
     mova                 m7, m5
 .w32_loop_init:
     mov                 r3d, 2
@@ -705,10 +702,7 @@ ALIGN function_align
     RET
 ALIGN function_align
 .w64:
-%if WIN64
-    movaps         [rsp+24], xmm7
-    %define xmm_regs_used 8
-%endif
+    WIN64_PUSH_XMM        8, 7
     mova                 m7, m5
 .w64_loop_init:
     mov                 r3d, 4
diff --git a/third_party/dav1d/src/x86/looprestoration_sse.asm b/third_party/dav1d/src/x86/looprestoration_sse.asm
index 01eb6fa348..b5c73a51d4 100644
--- a/third_party/dav1d/src/x86/looprestoration_sse.asm
+++ b/third_party/dav1d/src/x86/looprestoration_sse.asm
@@ -42,7 +42,6 @@ pb_0to15:      db  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15
 pb_right_ext_mask: times 24 db 0xff
                    times 8 db 0
 pb_1:          times 16 db 1
-pb_3:          times 16 db 3
 pw_256:        times 8 dw 256
 pw_2056:       times 8 dw 2056
 pw_m16380:     times 8 dw -16380
@@ -290,7 +289,7 @@ cglobal wiener_filter7_8bpc, 0, 7, 8, -384*12-stk_off, _, x, left, lpf, tmpstrid
     call mangle(private_prefix %+ _wiener_filter7_8bpc_ssse3).v
     jmp .v1
 .extend_right:
-    movd            m2, [lpfq-4]
+    movd            m2, [lpfq-1]
 %if ARCH_X86_64
     push            r0
     lea             r0, [pb_right_ext_mask+21]
@@ -302,10 +301,11 @@ cglobal wiener_filter7_8bpc, 0, 7, 8, -384*12-stk_off, _, x, left, lpf, tmpstrid
     movu            m1, [r6+xq+8]
 %endif
 %if cpuflag(ssse3)
-    pshufb          m2, [base+pb_3]
+    pxor            m3, m3
+    pshufb          m2, m3
 %else
     punpcklbw       m2, m2
-    pshuflw         m2, m2, q3333
+    pshuflw         m2, m2, q0000
     punpcklqdq      m2, m2
 %endif
     pand            m4, m0
diff --git a/third_party/dav1d/src/x86/mc16_avx2.asm b/third_party/dav1d/src/x86/mc16_avx2.asm
index 61eeaa1007..42e2a5525e 100644
--- a/third_party/dav1d/src/x86/mc16_avx2.asm
+++ b/third_party/dav1d/src/x86/mc16_avx2.asm
@@ -1337,7 +1337,6 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     cmp                  wd, 4
     je .h_w4
     jl .h_w2
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      13
     shr                 mxd, 16
     sub                srcq, 6
@@ -1415,7 +1414,6 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     cmp                  hd, 4
     cmovle              myd, mxd
     vpbroadcastq         m0, [base+subpel_filters+myq*8]
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      15
     vpbroadcastd         m6, [pd_32]
     vpbroadcastw         m7, r8m
@@ -1590,7 +1588,6 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     jg .v_w8_loop0
     RET
 .hv:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     vpbroadcastw        m15, r8m
     cmp                  wd, 4
@@ -2046,7 +2043,6 @@ cglobal prep_8tap_16bpc, 4, 8, 0, tmp, src, stride, w, h, mx, my
     shr                 mxd, 16
     sub                srcq, 6
     vpbroadcastq         m0, [base+subpel_filters+mxq*8]
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      12
     vbroadcasti128       m6, [subpel_h_shufA]
     vbroadcasti128       m7, [subpel_h_shufB]
@@ -2125,7 +2121,6 @@ cglobal prep_8tap_16bpc, 4, 8, 0, tmp, src, stride, w, h, mx, my
     cmp                  hd, 4
     cmovle              myd, mxd
     vpbroadcastq         m0, [base+subpel_filters+myq*8]
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      15
     vpbroadcastd         m7, [prep_8tap_1d_rnd]
     lea                  r6, [strideq*3]
@@ -2264,7 +2259,6 @@ cglobal prep_8tap_16bpc, 4, 8, 0, tmp, src, stride, w, h, mx, my
 %endif
     RET
 .hv:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     vpbroadcastd        m15, [prep_8tap_2d_rnd]
     cmp                  wd, 4
diff --git a/third_party/dav1d/src/x86/mc16_avx512.asm b/third_party/dav1d/src/x86/mc16_avx512.asm
index 585ba53e08..e5de7ecd96 100644
--- a/third_party/dav1d/src/x86/mc16_avx512.asm
+++ b/third_party/dav1d/src/x86/mc16_avx512.asm
@@ -2377,7 +2377,6 @@ cglobal put_8tap_16bpc, 4, 9, 16, dst, ds, src, ss, w, h, mx, my
     jg .hv_w16_loop
     RET
 .hv_w32:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM 32
     vbroadcasti32x4     m20, [spel_h_shufA]
     vbroadcasti32x4     m21, [spel_h_shufB]
@@ -3175,7 +3174,6 @@ cglobal prep_8tap_16bpc, 3, 8, 16, tmp, src, stride, w, h, mx, my, stride3
     jg .hv_w8_loop
     RET
 .hv_w16:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM 27
     vbroadcasti32x8      m5, [srcq+strideq*0+ 8]
     vinserti32x8         m4, m5, [srcq+strideq*0+ 0], 0
@@ -3313,7 +3311,6 @@ cglobal prep_8tap_16bpc, 3, 8, 16, tmp, src, stride, w, h, mx, my, stride3
     RET
 .hv_w32:
 %if WIN64
-    %assign stack_offset stack_offset - stack_size_padded
     PUSH                 r8
     %assign regs_used regs_used + 1
     WIN64_SPILL_XMM 32
diff --git a/third_party/dav1d/src/x86/mc16_sse.asm b/third_party/dav1d/src/x86/mc16_sse.asm
index fde8e372a3..b0c42597f7 100644
--- a/third_party/dav1d/src/x86/mc16_sse.asm
+++ b/third_party/dav1d/src/x86/mc16_sse.asm
@@ -1302,10 +1302,7 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     jg .h_w4_loop
     RET
 .h_w8:
-%if WIN64
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      12
-%endif
     shr                 mxd, 16
     movq                 m3, [base+subpel_filters+mxq*8]
     movifnidn          dstq, dstmp
@@ -1383,14 +1380,7 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     cmp                  hd, 6
     cmovb               myd, mxd
     movq                 m3, [base+subpel_filters+myq*8]
-%if STACK_ALIGNMENT < 16
-    %xdefine           rstk  rsp
-%else
-    %assign stack_offset stack_offset - stack_size_padded
-%endif
-%if WIN64
     WIN64_SPILL_XMM      15
-%endif
     movd                 m7, r8m
     movifnidn          dstq, dstmp
     movifnidn           dsq, dsmp
@@ -1604,11 +1594,7 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     jg .v_w4_loop0
     RET
 .hv:
-%if STACK_ALIGNMENT < 16
-    %xdefine           rstk  rsp
-%else
-    %assign stack_offset stack_offset - stack_size_padded
-%endif
+    RESET_STACK_STATE
 %if ARCH_X86_32
     movd                 m4, r8m
     mova                 m6, [base+pd_512]
@@ -1750,11 +1736,7 @@ cglobal put_8tap_16bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my
     cmovb               myd, mxd
     movq                 m3, [base+subpel_filters+myq*8]
 %if ARCH_X86_32
-%if STACK_ALIGNMENT < 16
-    %xdefine           rstk  rsp
-%else
-    %assign stack_offset stack_offset - stack_size_padded
-%endif
+    RESET_STACK_STATE
     mov                dstq, dstmp
     mov                 dsq, dsmp
     mova                 m0, [base+spel_h_shufA]
@@ -2182,11 +2164,6 @@ cglobal prep_8tap_16bpc, 4, 8, 0, tmp, src, ss, w, h, mx, my
     cmp                  hd, 4
     cmove               myd, mxd
     movq                 m3, [base+subpel_filters+myq*8]
-%if STACK_ALIGNMENT < 16
-    %xdefine           rstk  rsp
-%else
-    %assign stack_offset stack_offset - stack_size_padded
-%endif
     WIN64_SPILL_XMM      15
     movddup              m7, [base+prep_8tap_1d_rnd]
     movifnidn           ssq, r2mp
@@ -2339,11 +2316,7 @@ cglobal prep_8tap_16bpc, 4, 8, 0, tmp, src, ss, w, h, mx, my
     jg .v_loop0
     RET
 .hv:
-%if STACK_ALIGNMENT < 16
-    %xdefine           rstk  rsp
-%else
-    %assign stack_offset stack_offset - stack_size_padded
-%endif
+    RESET_STACK_STATE
     movzx               t3d, mxb
     shr                 mxd, 16
     cmp                  wd, 4
diff --git a/third_party/dav1d/src/x86/mc_avx2.asm b/third_party/dav1d/src/x86/mc_avx2.asm
index 3b208033bd..58e3cb5af1 100644
--- a/third_party/dav1d/src/x86/mc_avx2.asm
+++ b/third_party/dav1d/src/x86/mc_avx2.asm
@@ -1259,7 +1259,6 @@ cglobal prep_bilin_8bpc, 3, 7, 0, tmp, src, stride, w, h, mxy, stride3
 .hv:
     ; (16 * src[x] + (my * (src[x + src_stride] - src[x])) + 8) >> 4
     ; = src[x] + (((my * (src[x + src_stride] - src[x])) + 8) >> 4)
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM       7
     movzx                wd, word [r6+wq*2+table_offset(prep, _bilin_hv)]
     shl                mxyd, 11
@@ -1620,7 +1619,6 @@ cglobal put_8tap_8bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
     jg .h_loop
     RET
 .v:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     movzx               mxd, myb
     shr                 myd, 16
@@ -1834,7 +1832,6 @@ cglobal put_8tap_8bpc, 4, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
     jg .v_w16_loop0
     RET
 .hv:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     cmp                  wd, 4
     jg .hv_w8
@@ -2247,7 +2244,6 @@ cglobal prep_8tap_8bpc, 3, 8, 0, tmp, src, stride, w, h, mx, my, stride3
     jg .h_loop
     RET
 .v:
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM      16
     movzx               mxd, myb ; Select 4-tap/8-tap filter multipliers.
     shr                 myd, 16  ; Note that the code is 8-tap only, having
@@ -2430,8 +2426,6 @@ cglobal prep_8tap_8bpc, 3, 8, 0, tmp, src, stride, w, h, mx, my, stride3
     jg .v_w16_loop0
     RET
 .hv:
-    %assign stack_offset stack_offset - stack_size_padded
-    %assign stack_size_padded 0
     WIN64_SPILL_XMM      16
     cmp                  wd, 4
     je .hv_w4
@@ -4108,10 +4102,9 @@ cglobal warp_affine_8x8t_8bpc, 0, 14, 0, tmp, ts
 cglobal warp_affine_8x8_8bpc, 0, 14, 0, dst, ds, src, ss, abcd, mx, tmp2, alpha, \
                                         beta, filter, tmp1, delta, my, gamma
 %if WIN64
-    sub                 rsp, 0xa0
     %assign xmm_regs_used 16
     %assign stack_size_padded 0xa0
-    %assign stack_offset stack_offset+stack_size_padded
+    SUB                 rsp, stack_size_padded
 %endif
     call .main
     jmp .start
@@ -4134,21 +4127,13 @@ cglobal warp_affine_8x8_8bpc, 0, 14, 0, dst, ds, src, ss, abcd, mx, tmp2, alpha,
     RET
 ALIGN function_align
 .main:
-    ; Stack args offset by one (r4m -> r5m etc.) due to call
-%if WIN64
-    mov               abcdq, r5m
-    mov                 mxd, r6m
-    movaps [rsp+stack_offset+0x10], xmm6
-    movaps [rsp+stack_offset+0x20], xmm7
-    movaps       [rsp+0x28], xmm8
-    movaps       [rsp+0x38], xmm9
-    movaps       [rsp+0x48], xmm10
-    movaps       [rsp+0x58], xmm11
-    movaps       [rsp+0x68], xmm12
-    movaps       [rsp+0x78], xmm13
-    movaps       [rsp+0x88], xmm14
-    movaps       [rsp+0x98], xmm15
-%endif
+    ; Stack is offset due to call
+    %assign stack_offset stack_offset + gprsize
+    %assign stack_size stack_size + gprsize
+    %assign stack_size_padded stack_size_padded + gprsize
+    movifnidn         abcdq, abcdmp
+    movifnidn           mxd, mxm
+    WIN64_PUSH_XMM
     movsx            alphad, word [abcdq+2*0]
     movsx             betad, word [abcdq+2*1]
     mova                m12, [warp_8x8_shufA]
@@ -4162,7 +4147,7 @@ ALIGN function_align
     lea               tmp2d, [alphaq*3]
     sub                srcq, tmp1q    ; src -= src_stride*3 + 3
     sub               betad, tmp2d    ; beta -= alpha*3
-    mov                 myd, r7m
+    mov                 myd, r6m
     call .h
     psrld                m1, m0, 16
     call .h
diff --git a/third_party/dav1d/src/x86/mc_avx512.asm b/third_party/dav1d/src/x86/mc_avx512.asm
index 7897f1decc..f9043f1ad3 100644
--- a/third_party/dav1d/src/x86/mc_avx512.asm
+++ b/third_party/dav1d/src/x86/mc_avx512.asm
@@ -1276,7 +1276,6 @@ cglobal prep_bilin_8bpc, 3, 7, 0, tmp, src, stride, w, h, mxy, stride3
 .hv:
     ; (16 * src[x] + (my * (src[x + src_stride] - src[x])) + 8) >> 4
     ; = src[x] + (((my * (src[x + src_stride] - src[x])) + 8) >> 4)
-    %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM       7
     movzx                wd, word [t2+wq*2+table_offset(prep, _bilin_hv)]
     shl                mxyd, 11
@@ -2853,8 +2852,6 @@ cglobal prep_8tap_8bpc, 3, 8, 0, tmp, src, stride, w, h, mx, my, stride3
     jg .v_loop0
     RET
 .hv:
-    %assign stack_offset stack_offset - stack_size_padded
-    %assign stack_size_padded 0
     WIN64_SPILL_XMM      16
     cmp                  wd, 4
     je .hv_w4
diff --git a/third_party/dav1d/src/x86/mc_sse.asm b/third_party/dav1d/src/x86/mc_sse.asm
index 54939c647a..a447a80161 100644
--- a/third_party/dav1d/src/x86/mc_sse.asm
+++ b/third_party/dav1d/src/x86/mc_sse.asm
@@ -1199,7 +1199,6 @@ cglobal prep_bilin_8bpc, 3, 7, 0, tmp, src, stride, w, h, mxy, stride3
     RET
 .v:
 %if notcpuflag(ssse3)
- %assign stack_offset stack_offset - stack_size_padded
     WIN64_SPILL_XMM 8
 %endif
     movzx                wd, word [r6+wq*2+table_offset(prep, _bilin_v)]
@@ -1375,7 +1374,6 @@ cglobal prep_bilin_8bpc, 3, 7, 0, tmp, src, stride, w, h, mxy, stride3
     ; (16 * src[x] + (my * (src[x + src_stride] - src[x])) + 8) >> 4
     ; = src[x] + (((my * (src[x + src_stride] - src[x])) + 8) >> 4)
     movzx                wd, word [r6+wq*2+table_offset(prep, _bilin_hv)]
-%assign stack_offset stack_offset - stack_size_padded
 %if cpuflag(ssse3)
     imul               mxyd, 0x08000800
     WIN64_SPILL_XMM 8
@@ -1592,7 +1590,6 @@ FN put_8tap, regular,        REGULAR, REGULAR
 %endif
 
 cglobal put_8tap_8bpc, 1, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
-%assign org_stack_offset stack_offset
     imul                mxd, mxm, 0x010101
     add                 mxd, t0d ; 8tap_h, mx, 4tap_h
 %if ARCH_X86_64
@@ -1618,7 +1615,6 @@ cglobal put_8tap_8bpc, 1, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
     movzx                wd, word [base_reg+wq*2+table_offset(put,)]
     add                  wq, base_reg
 ; put_bilin mangling jump
-%assign stack_offset org_stack_offset
     movifnidn           dsq, dsmp
     movifnidn           ssq, ssmp
 %if WIN64
@@ -1792,7 +1788,6 @@ cglobal put_8tap_8bpc, 1, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
     cmovs               ssd, mxd
     movq                 m0, [base_reg+ssq*8+subpel_filters-put_ssse3]
 %else
- %assign stack_offset org_stack_offset
     WIN64_SPILL_XMM      16
     movzx               mxd, myb
     shr                 myd, 16
@@ -2048,7 +2043,7 @@ cglobal put_8tap_8bpc, 1, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
 %undef subpel2
 %undef subpel3
 .hv:
-    %assign stack_offset org_stack_offset
+    RESET_STACK_STATE
     cmp                  wd, 4
     jg .hv_w8
 %if ARCH_X86_32
@@ -2369,7 +2364,7 @@ cglobal put_8tap_8bpc, 1, 9, 0, dst, ds, src, ss, w, h, mx, my, ss3
 %undef subpelv2
 %undef subpelv3
 .hv_w8:
-    %assign stack_offset org_stack_offset
+    RESET_STACK_STATE
 %define hv8_line_1 0
 %define hv8_line_2 1
 %define hv8_line_3 2
@@ -2843,7 +2838,6 @@ FN prep_8tap, regular,        REGULAR, REGULAR
  %define base 0
 %endif
 cglobal prep_8tap_8bpc, 1, 9, 0, tmp, src, stride, w, h, mx, my, stride3
-%assign org_stack_offset stack_offset
     imul                mxd, mxm, 0x010101
     add                 mxd, t0d ; 8tap_h, mx, 4tap_h
     imul                myd, mym, 0x010101
@@ -2862,7 +2856,6 @@ cglobal prep_8tap_8bpc, 1, 9, 0, tmp, src, stride, w, h, mx, my, stride3
     add                  wq, base_reg
     movifnidn       strided, stridem
     lea                  r6, [strideq*3]
-    %assign stack_offset org_stack_offset
 %if WIN64
     pop                  r8
     pop                  r7
@@ -3095,7 +3088,6 @@ cglobal prep_8tap_8bpc, 1, 9, 0, tmp, src, stride, w, h, mx, my, stride3
     mov                 mxd, myd
     and                 mxd, 0x7f
 %else
- %assign stack_offset org_stack_offset
     WIN64_SPILL_XMM      16
     movzx               mxd, myb
 %endif
@@ -3359,7 +3351,7 @@ cglobal prep_8tap_8bpc, 1, 9, 0, tmp, src, stride, w, h, mx, my, stride3
 %undef subpel2
 %undef subpel3
 .hv:
-    %assign stack_offset org_stack_offset
+    RESET_STACK_STATE
     cmp                  wd, 4
     jg .hv_w8
     and                 mxd, 0x7f
@@ -3659,7 +3651,7 @@ cglobal prep_8tap_8bpc, 1, 9, 0, tmp, src, stride, w, h, mx, my, stride3
 %undef subpelv2
 %undef subpelv3
 .hv_w8:
-    %assign stack_offset org_stack_offset
+    RESET_STACK_STATE
 %define hv8_line_1 0
 %define hv8_line_2 1
 %define hv8_line_3 2
diff --git a/third_party/dav1d/src/x86/msac.asm b/third_party/dav1d/src/x86/msac.asm
index 9f05c921a6..4156efe914 100644
--- a/third_party/dav1d/src/x86/msac.asm
+++ b/third_party/dav1d/src/x86/msac.asm
@@ -143,10 +143,9 @@ cglobal msac_decode_symbol_adapt4, 0, 6, 6
     mov           esp, [esp]
 %endif
 %endif
-    not            t4
     sub           t2d, t1d ; rng
     shl            t1, gprsize*8-16
-    add            t4, t1  ; ~dif
+    sub            t4, t1  ; dif - v
 .renorm3:
     mov           t1d, [t0+msac.cnt]
     movifnidn      t7, t0
@@ -157,33 +156,31 @@ cglobal msac_decode_symbol_adapt4, 0, 6, 6
     shl           t2d, cl
     shl            t4, cl
     mov [t7+msac.rng], t2d
-    not            t4
     sub           t1d, ecx
     jae .end ; no refill required
 
 ; refill:
-    mov            t2, [t7+msac.buf]
-    mov           rcx, [t7+msac.end]
 %if ARCH_X86_64 == 0
     push           t5
 %endif
-    lea            t5, [t2+gprsize]
-    cmp            t5, rcx
+    mov            t2, [t7+msac.buf]
+    mov            t5, [t7+msac.end]
+    lea           rcx, [t2+gprsize]
+    sub           rcx, t5
     ja .refill_eob
-    mov            t2, [t2]
-    lea           ecx, [t1+23]
-    add           t1d, 16
-    shr           ecx, 3   ; shift_bytes
-    bswap          t2
-    sub            t5, rcx
-    shl           ecx, 3   ; shift_bits
-    shr            t2, cl
-    sub           ecx, t1d ; shift_bits - 16 - cnt
-    mov           t1d, gprsize*8-16
-    shl            t2, cl
-    mov [t7+msac.buf], t5
-    sub           t1d, ecx ; cnt + gprsize*8 - shift_bits
-    xor            t4, t2
+    mov            t5, [t2]
+    lea           ecx, [t1+16-gprsize*8]
+    not            t5
+    bswap          t5
+    shr            t5, cl
+    neg           ecx
+    shr           ecx, 3 ; num_bytes_read
+    or             t4, t5
+.refill_end:
+    add            t2, rcx
+    lea           t1d, [t1+rcx*8] ; cnt += num_bits_read
+    mov [t7+msac.buf], t2
+.refill_end2:
 %if ARCH_X86_64 == 0
     pop            t5
 %endif
@@ -191,29 +188,35 @@ cglobal msac_decode_symbol_adapt4, 0, 6, 6
     mov [t7+msac.cnt], t1d
     mov [t7+msac.dif], t4
     RET
+.pad_with_ones:
+    lea           ecx, [t1-16]
+%if ARCH_X86_64
+    ror           rcx, cl
+%else
+    shr           ecx, cl
+%endif
+    or             t4, rcx
+    jmp .refill_end2
 .refill_eob: ; avoid overreading the input buffer
-    mov            t5, rcx
-    mov           ecx, gprsize*8-24
-    sub           ecx, t1d ; c
-.refill_eob_loop:
     cmp            t2, t5
-    jae .refill_eob_end    ; eob reached
-    movzx         t1d, byte [t2]
-    inc            t2
-    shl            t1, cl
-    xor            t4, t1
-    sub           ecx, 8
-    jge .refill_eob_loop
-.refill_eob_end:
-    mov           t1d, gprsize*8-24
-%if ARCH_X86_64 == 0
-    pop            t5
-%endif
-    sub           t1d, ecx
-    mov [t7+msac.buf], t2
-    mov [t7+msac.dif], t4
-    mov [t7+msac.cnt], t1d
-    RET
+    jae .pad_with_ones ; eob reached
+    ; We can safely do a register-sized load of the last bytes of the buffer
+    ; as this code is only reached if the msac buffer size is >= gprsize.
+    mov            t5, [t5-gprsize]
+    shl           ecx, 3
+    shr            t5, cl
+    lea           ecx, [t1+16-gprsize*8]
+    not            t5
+    bswap          t5
+    shr            t5, cl
+    neg           ecx
+    or             t4, t5
+    mov           t5d, [t7+msac.end]
+    shr           ecx, 3
+    sub           t5d, t2d ; num_bytes_left
+    cmp           ecx, t5d
+    cmovae        ecx, t5d ; num_bytes_read
+    jmp .refill_end
 
 cglobal msac_decode_symbol_adapt8, 0, 6, 6
     DECODE_SYMBOL_ADAPT_INIT
@@ -366,7 +369,6 @@ cglobal msac_decode_bool_adapt, 0, 6, 0
 %if ARCH_X86_64 == 0
     movzx         eax, al
 %endif
-    not            t4
     test          t3d, t3d
     jz m(msac_decode_symbol_adapt4, SUFFIX).renorm3
 %if UNIX64 == 0
@@ -420,7 +422,6 @@ cglobal msac_decode_bool_equi, 0, 6, 0
     mov           ecx, 0xbfff
     setb           al ; the upper 32 bits contains garbage but that's OK
     sub           ecx, t2d
-    not            t4
     ; In this case of this function, (d =) 16 - clz(v) = 2 - (v >> 14)
     ;   i.e. (0 <= d <= 2) and v < (3 << 14)
     shr           ecx, 14           ; d
@@ -447,7 +448,6 @@ cglobal msac_decode_bool, 0, 6, 0
     cmovb         t2d, t1d
     cmovb          t4, t3
     setb           al
-    not            t4
 %if ARCH_X86_64 == 0
     movzx         eax, al
 %endif
@@ -497,48 +497,45 @@ cglobal msac_decode_bool, 0, 6, 0
     tzcnt         eax, eax
     movzx         ecx, word [buf+rax+16]
     movzx         t2d, word [buf+rax+14]
-    not            t4
 %if ARCH_X86_64
     add           t6d, 5
 %endif
     sub           eax, 5   ; setup for merging the tok_br and tok branches
     sub           t2d, ecx
     shl           rcx, gprsize*8-16
-    add            t4, rcx
+    sub            t4, rcx
     bsr           ecx, t2d
     xor           ecx, 15
     shl           t2d, cl
     shl            t4, cl
     movd           m2, t2d
     mov [t7+msac.rng], t2d
-    not            t4
     sub           t5d, ecx
     jae %%end
-    mov            t2, [t7+msac.buf]
-    mov           rcx, [t7+msac.end]
 %if UNIX64 == 0
     push           t8
 %endif
-    lea            t8, [t2+gprsize]
-    cmp            t8, rcx
+    mov            t2, [t7+msac.buf]
+    mov            t8, [t7+msac.end]
+    lea           rcx, [t2+gprsize]
+    sub           rcx, t8
     ja %%refill_eob
-    mov            t2, [t2]
-    lea           ecx, [t5+23]
-    add           t5d, 16
+    mov            t8, [t2]
+    lea           ecx, [t5+16-gprsize*8]
+    not            t8
+    bswap          t8
+    shr            t8, cl
+    neg           ecx
     shr           ecx, 3
-    bswap          t2
-    sub            t8, rcx
-    shl           ecx, 3
-    shr            t2, cl
-    sub           ecx, t5d
-    mov           t5d, gprsize*8-16
-    shl            t2, cl
-    mov [t7+msac.buf], t8
+    or             t4, t8
+%%refill_end:
+    add            t2, rcx
+    lea           t5d, [t5+rcx*8]
+    mov [t7+msac.buf], t2
+%%refill_end2:
 %if UNIX64 == 0
     pop            t8
 %endif
-    sub           t5d, ecx
-    xor            t4, t2
 %%end:
     movp           m3, t4
 %if ARCH_X86_64
@@ -559,27 +556,34 @@ cglobal msac_decode_bool, 0, 6, 0
     shr           eax, 1
     mov [t7+msac.cnt], t5d
     RET
+%%pad_with_ones:
+    ; ensure that dif is padded with at least 15 bits of ones at the end
+    lea           ecx, [t5-16]
+%if ARCH_X86_64
+    ror           rcx, cl
+%else
+    shr           ecx, cl
+%endif
+    or             t4, rcx
+    jmp %%refill_end2
 %%refill_eob:
-    mov            t8, rcx
-    mov           ecx, gprsize*8-24
-    sub           ecx, t5d
-%%refill_eob_loop:
     cmp            t2, t8
-    jae %%refill_eob_end
-    movzx         t5d, byte [t2]
-    inc            t2
-    shl            t5, cl
-    xor            t4, t5
-    sub           ecx, 8
-    jge %%refill_eob_loop
-%%refill_eob_end:
-%if UNIX64 == 0
-    pop            t8
-%endif
-    mov           t5d, gprsize*8-24
-    mov [t7+msac.buf], t2
-    sub           t5d, ecx
-    jmp %%end
+    jae %%pad_with_ones
+    mov            t8, [t8-gprsize]
+    shl           ecx, 3
+    shr            t8, cl
+    lea           ecx, [t5+16-gprsize*8]
+    not            t8
+    bswap          t8
+    shr            t8, cl
+    neg           ecx
+    or             t4, t8
+    mov           t8d, [t7+msac.end]
+    shr           ecx, 3
+    sub           t8d, t2d
+    cmp           ecx, t8d
+    cmovae        ecx, t8d
+    jmp %%refill_end
 %endmacro
 
 cglobal msac_decode_hi_tok, 0, 7 + ARCH_X86_64, 6