[Backport] Security bug 1401571102-based

Manual update of libdav1d to match the version introduced by patch
https://chromium-review.googlesource.com/c/chromium/src/+/4114163:
Roll src/third_party/dav1d/libdav1d/ 87f9a81cd..ed63a7459 (104 commits)

This roll required a few changes to get working:
- "properties" => "built in options" crossfile configuration change due to Meson deprecation.
- generic config creation never worked, so fixed.
- PPC64 configs were never checked in, so switched to generic.
- copyright header changes for generate_sources.
- Updated readme.chromium with potential issues that can arise.

https://chromium.googlesource.com/external/github.com/videolan/dav1d.git/+log/87f9a81cd770..ed63a7459376

$ git log 87f9a81cd..ed63a7459 --date=short --no-merges --format='%ad %ae %s'
2022-12-09 jamrial dav1d: add an option to skip decoding some frame types
2022-12-08 jamrial picture: support creating and freeing refs without tile data
2022-12-07 gramner x86: Add 10bpc 8x32/32x8 itx AVX-512 (Ice Lake) asm
2022-12-07 gramner x86: Add minor DC-only IDCT optimizations
2022-12-13 gramner getbits: Fix assertion failure
2022-12-13 gramner checkasm: Fix integer overflow in refmvs test
2022-01-26 gramner dav1dplay: Update to new libplacebo API
2022-12-09 gramner Add minor getbits improvements
2022-12-09 gramner Add a separate getbits function for getting a single bit
2022-12-09 gramner Remove redundant zeroing in sequence header parsing
2022-12-09 gramner Set the correct default value of initial_display_delay
2022-12-09 jamrial tools: remove the null last entry in inloop_filters_tbl
2022-12-04 lu_zero Do not assume the picture allocation starts as the left edge
2022-11-21 lu_zero ppc: Allocate the correct temp buffer size
2022-11-21 lu_zero ppc: Do not use static const with vec_splats
2022-11-02 charlie.c.hayden Add info to dav1d_send_data docs
2022-10-30 jbeich build: drop -D_DARWIN_C_SOURCE on macOS/iOS after 6b611d36acab
2022-10-30 jbeich build: drop -D_POSIX_C_SOURCE on non-Linux after 6b611d36acab
2022-06-28 victorien threading: Add a pending list for async task insertion
2022-10-26 martin Implement atomic_compare_exchange_strong in the atomic compat headers
2022-10-06 victorien threading: Fix a race around frame completion (frame-mt)
2022-10-07 sebastian Handle host_machine.system() 'ios' and 'tvos' the same way as 'darwin'
2022-09-23 gramner x86: Add 10-bit 8x8/8x16/16x8/16x16 itx AVX-512 (Ice Lake) asm
2022-09-30 gramner Specify hidden visibility for global data symbol declarations
2022-09-28 gramner build: strip() the result of cc.get_define()
2022-09-26 gramner checkasm: Move printf format string to .rodata on x86
2022-09-26 gramner checkasm: Improve 32-bit parameter clobbering on x86-64
2022-09-26 gramner x86: Fix incorrect 32-bit parameter usage in high bit-depth AVX-512 mc
2022-09-09 martin arm: itx: Add clipping to row_clip_min/max in the 10 bpc codepaths
2022-09-15 gramner x86: Fix overflows in 12bpc AVX2 IDCT/IADST
2022-09-15 gramner x86: Fix overflows in 12bpc AVX2 DC-only IDCT
2022-09-15 gramner x86: Fix clipping in high bit-depth AVX2 4x16 IDCT
2022-03-21 martin Don't use gas-preprocessor with clang-cl for arm targets
2022-06-07 david_conrad Fix checking the reference dimesions for the projection process
2022-06-07 david_conrad Fix calculation of OBMC lap dimensions
2022-06-07 david_conrad Support film grain application whose only effect is clipping to video range
2022-06-07 david_conrad Ignore T.35 metadata if the OBU contains no payload
2022-06-07 david_conrad Fix chroma deblock filter size calculation for lossless
2022-06-07 david_conrad Fix rounding in the calculation of initialSubpelX
2022-06-07 david_conrad Fix overflow when saturating dequantized coefficients clipped to 0
2022-06-08 david_conrad Fix overflow in 8-bit NEON ADST
2022-09-14 martin tools: Allocate the priv structs with proper alignment
2022-09-08 gramner x86: Fix clipping in 10bpc SSE4.1 IDCT asm
2022-09-08 gramner build: Improve Windows linking options
2022-09-08 gramner tools: Improve demuxer probing
2022-08-30 code CI: Disable trimming on some tests
2022-08-30 code CI: Remove git 'safe.directory' config
2022-08-30 code gcovr: Ignore parsing errors
2022-08-30 code crossfiles: Update Android toolchains
2022-08-30 code CI: Update images
(...)
2022-09-01 victorien checkasm: Add short options
2022-09-01 victorien checkasm: Add pattern matching to --test
2022-09-01 victorien checkasm: Remove pattern matching from --bench
2022-08-29 victorien checkasm: Add a --function option
2022-08-30 victorien threading: Fix copy_lpf_progress initialization
2022-08-19 jamrial data: don't overwrite the Dav1dDataProps size value
2022-07-18 gramner Adjust inlining attributes on some functions
2022-07-19 gramner x86: Remove leftover instruction in loopfilter AVX2 asm
2022-06-07 david_conrad Enable pointer authentication in assembly when building arm64e
2022-06-07 david_conrad Don't trash the return stack buffer in the NEON loop filter
2022-07-03 thresh CI: Removed snap package generation
2022-07-06 gramner Eliminate unused C DSP functions at compile time
2022-07-06 gramner cpu: Inline dav1d_get_cpu_flags()
2022-06-22 gramner x86: Add minor loopfilter asm improvements
2022-06-15 gramner checkasm: Speed up signal handling
2022-06-15 gramner checkasm: Improve seed generation on Windows
2022-06-20 gramner ci: Don't specify a specific MacOS version
2022-06-14 gramner x86: Add high bit-depth loopfilter AVX-512 (Ice Lake) asm
2022-06-13 victorien checkasm/lpf: Use operating dimensions
2022-06-03 gramner checkasm: Print the cpu model and cpuid signature on x86
2022-06-03 gramner checkasm: Add a vzeroupper check on x86
2022-06-02 gramner x86: Add a workaround for quirky AVX-512 hardware behavior
2022-05-31 victorien checkasm: Fix uninitialized variable
2022-05-14 code CI: Update coverage collecting
2022-05-05 code CI: Add a build with the minimum requirements
2022-05-05 code CI: Deactivate git 'safe.directory'
2022-03-24 code CI: Update images
2022-05-25 victorien Fix typo
2022-05-19 gramner x86: Add high bit-depth cdef_filter AVX-512 (Ice Lake) asm
2022-05-20 gramner checkasm: Print --help message to stderr instead of stdout
2022-05-20 gramner checkasm: Split cdef test into separate pri/sec/pri+sec parts
2022-05-20 gramner checkasm: Improve benchmarking of functions that modify their input
2022-05-18 b x86/itx_avx2: fix typo
2022-04-22 code CI: Add gcc12 and clang14 builds with mold linker
2022-04-26 code CI: Trigger documentation rebuild if configuration changes
2022-04-24 code meson/doc: Fix doxygen config
2022-04-28 gramner Use a relaxed memory ordering in dav1d_ref_inc()
2022-04-28 gramner Remove redundant code in dav1d_cdf_thread_unref()
2022-04-28 gramner Inline dav1d_ref_inc()
2022-04-24 code x86/itx: Add 32x8 12bpc AVX2 transforms
2022-04-24 code x86/itx: Add 8x32 12bpc AVX2 transforms
2022-04-24 code x86/itx: Deduplicate dconly code
2022-04-23 code lib: Fix typo in documentation
2022-04-07 jamrial obu: don't output invisible but showable key frames more than once
2022-04-07 jamrial obu: check that the frame referenced by existing_frame_idx is showable
2022-04-07 jamrial obu: check refresh_frame_flags is not equal to allFrames on Intra Only frames
2022-03-29 robux4 remove multipass wait from dav1d_decode_frame
2022-04-07 jamrial picture: ensure the new seq header and op param info flags are attached to the next visible picture in display order
2022-03-31 jamrial lib: add a function to query the decoder frame delay
2022-03-31 jamrial lib: split calculating thread count to its own function

Created with:
  roll-dep src/third_party/dav1d/libdav1d

Fixed: 1401571

Change-Id: Ic3cef540a87a2cf411abe6071fd4c9963ea61f75
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4114163
Reviewed-by: Wan-Teh Chang <wtc@google.com>
Commit-Queue: Dale Curtis <dalecurtis@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1084574}
Reviewed-on: https://codereview.qt-project.org/c/qt/qtwebengine-chromium/+/468619
Reviewed-by: Michal Klocek <michal.klocek@qt.io>

1 files changed, 123 insertions, 108 deletions

diff --git a/chromium/third_party/dav1d/libdav1d/src/x86/itx16_sse.asm b/chromium/third_party/dav1d/libdav1d/src/x86/itx16_sse.asm
index 4fb30ef4e7a..3833e17c99f 100644
--- a/chromium/third_party/dav1d/libdav1d/src/x86/itx16_sse.asm
+++ b/chromium/third_party/dav1d/libdav1d/src/x86/itx16_sse.asm
@@ -361,18 +361,32 @@ ALIGN function_align
 %macro INV_TXFM_4X4_FN 2 ; type1, type2
     INV_TXFM_FN          %1, %2, 0, 4x4
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
-    movd                 m1, [o(pw_2896x8)]
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
-    add                 r5d, 2048
-    sar                 r5d, 12
+    mov                 r3d, 4
+.dconly:
+    add                 r5d, 128
+    sar                 r5d, 8
+.dconly2:
+    imul                r5d, 2896
+    mova                 m2, [o(pixel_10bpc_max)]
+    add                 r5d, 34816
     movd                 m0, r5d
-    packssdw             m0, m0
-    pmulhrsw             m0, m1
-    pshuflw              m0, m0, q0000
+    pshuflw              m0, m0, q1111
+    pxor                 m3, m3
     punpcklqdq           m0, m0
-    mova                 m1, m0
-    TAIL_CALL m(iadst_4x4_internal_16bpc).end
+.dconly_loop:
+    movq                 m1, [dstq+strideq*0]
+    movhps               m1, [dstq+strideq*1]
+    paddw                m1, m0
+    pminsw               m1, m2
+    pmaxsw               m1, m3
+    movq   [dstq+strideq*0], m1
+    movhps [dstq+strideq*1], m1
+    lea                dstq, [dstq+strideq*2]
+    sub                 r3d, 2
+    jg .dconly_loop
+    RET
 %endif
 %endmacro
 
@@ -662,40 +676,13 @@ cglobal iidentity_4x4_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
 %macro INV_TXFM_4X8_FN 2-3 0 ; type1, type2, eob_offset
     INV_TXFM_FN          %1, %2, %3, 4x8
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
-    mov                 r3d, 2
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
-    add                 r5d, 2048
-    sar                 r5d, 12
-.end:
-    imul                r5d, 2896
-    add                 r5d, 34816
-    movd                 m0, r5d
-    pshuflw              m0, m0, q1111
-    punpcklqdq           m0, m0
-    pxor                 m4, m4
-    mova                 m3, [o(pixel_10bpc_max)]
-    lea                  r2, [strideq*3]
-.loop:
-    movq                 m1, [dstq+strideq*0]
-    movq                 m2, [dstq+strideq*2]
-    movhps               m1, [dstq+strideq*1]
-    movhps               m2, [dstq+r2]
-    paddw                m1, m0
-    paddw                m2, m0
-    REPX     {pminsw x, m3}, m1, m2
-    REPX     {pmaxsw x, m4}, m1, m2
-    movq   [dstq+strideq*0], m1
-    movhps [dstq+strideq*1], m1
-    movq   [dstq+strideq*2], m2
-    movhps [dstq+r2       ], m2
-    lea                dstq, [dstq+strideq*4]
-    dec                 r3d
-    jg .loop
-    RET
+    mov                 r3d, 8
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
+    jmp m(inv_txfm_add_dct_dct_4x4_16bpc).dconly
 %endif
 %endmacro
 
@@ -944,12 +931,12 @@ cglobal iidentity_4x8_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
 %macro INV_TXFM_4X16_FN 2-3 2d ; type1, type2, eob_tbl_suffix
     INV_TXFM_FN          %1, %2, tbl_4x16_%3, 4x16
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
-    mov                 r3d, 4
-    add                 r5d, 6144
-    sar                 r5d, 13
-    jmp m(inv_txfm_add_dct_dct_4x8_16bpc).end
+    mov                 r3d, 16
+    add                 r5d, 384
+    sar                 r5d, 9
+    jmp m(inv_txfm_add_dct_dct_4x4_16bpc).dconly2
 %endif
 %endmacro
 
@@ -1297,13 +1284,13 @@ cglobal iidentity_4x16_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, 0, 8x4, 8, 0-4*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
-    add                 r5d, 2048
-    sar                 r5d, 12
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
+    add                 r5d, 128
+    sar                 r5d, 8
     imul                r5d, 2896
     add                 r5d, 34816
     movd                 m0, r5d
@@ -1783,12 +1770,12 @@ cglobal iidentity_8x4_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, %3, 8x8, 8, 0-5*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 2
 .end:
-    add                 r5d, 6144
-    sar                 r5d, 13
+    add                 r5d, 384
+    sar                 r5d, 9
 .end2:
     imul                r5d, 2896
     add                 r5d, 34816
@@ -2146,11 +2133,11 @@ cglobal iidentity_8x8_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, tbl_8x16_%3, 8x16, 8, 0-17*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
     mov                 r3d, 4
 %if stack_size_padded > 0
     ; adjust to caller's stack allocation
@@ -2477,12 +2464,12 @@ cglobal iidentity_8x16_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, 0, 16x4, 8, 0-12*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 4
 .dconly:
-    add                 r5d, 6144
-    sar                 r5d, 13
+    add                 r5d, 384
+    sar                 r5d, 9
 .dconly2:
     imul                r5d, 2896
     add                 r5d, 34816
@@ -2755,6 +2742,8 @@ cglobal idct_16x4_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     ret
 .round:
 %if ARCH_X86_64
+    REPX    {pmaxsd x, m12}, m0, m1, m2, m3, m4, m5, m6, m7
+    REPX    {pminsd x, m13}, m0, m1, m2, m3, m4, m5, m6, m7
     pcmpeqd              m8, m8
     REPX      {psubd x, m8}, m0, m1, m2, m3, m4, m5, m6, m7
     mova                 m8, [r3+1*16]
@@ -2785,6 +2774,14 @@ cglobal idct_16x4_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     ; and out0-15 is now in m0-15
 %else
     mova         [r3+ 0*16], m0
+    mova                 m0, [o(clip_18b_min)]
+    REPX     {pmaxsd x, m0}, m1, m2, m3, m4, m5, m6, m7
+    pmaxsd               m0, [r3+ 0*16]
+    mova         [r3+ 0*16], m7
+    mova                 m7, [o(clip_18b_max)]
+    REPX     {pminsd x, m7}, m0, m1, m2, m3, m4, m5, m6
+    pminsd               m7, [r3+ 0*16]
+    mova         [r3+ 0*16], m0
     pcmpeqd              m0, m0
     REPX      {psubd x, m0}, m1, m2, m3, m4, m5, m6, m7
     mova         [r3+ 1*16], m1
@@ -3472,12 +3469,12 @@ cglobal iidentity_16x4_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, %3, 16x8, 8, 0-13*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 8
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
 %if ARCH_X86_32
     add                 rsp, 1*16
 %endif
@@ -3939,11 +3936,11 @@ cglobal iidentity_16x8_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     INV_TXFM_FN          %1, %2, tbl_16x16_%3, 16x16, 8, 0-17*16
 %endif
 %ifidn %1_%2, dct_dct
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 16
-    add                 r5d, 10240
-    sar                 r5d, 14
+    add                 r5d, 640
+    sar                 r5d, 10
     add                 rsp, (5+ARCH_X86_64*3+WIN64)*16
     jmp m(inv_txfm_add_dct_dct_16x4_16bpc).dconly2
 %endif
@@ -4057,6 +4054,8 @@ cglobal idct_16x16_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     ret
 .round:
 %if ARCH_X86_64
+    REPX    {pmaxsd x, m12}, m0, m1, m2, m3, m4, m5, m6, m7
+    REPX    {pminsd x, m13}, m0, m1, m2, m3, m4, m5, m6, m7
     psrld                m8, m11, 10        ; 2
     REPX      {paddd x, m8}, m0, m1, m2, m3, m4, m5, m6, m7
     mova                 m8, [r3+1*16]
@@ -4087,6 +4086,14 @@ cglobal idct_16x16_internal_16bpc, 0, 0, 0, dst, stride, c, eob, tx2
     ; and out0-15 is now in m0-15
 %else
     mova         [r3+ 0*16], m0
+    mova                 m0, [o(clip_18b_min)]
+    REPX     {pmaxsd x, m0}, m1, m2, m3, m4, m5, m6, m7
+    pmaxsd               m0, [r3+ 0*16]
+    mova         [r3+ 0*16], m7
+    mova                 m7, [o(clip_18b_max)]
+    REPX     {pminsd x, m7}, m0, m1, m2, m3, m4, m5, m6
+    pminsd               m7, [r3+ 0*16]
+    mova         [r3+ 0*16], m0
     mova                 m0, [o(pd_2)]
     REPX      {paddd x, m0}, m1, m2, m3, m4, m5, m6, m7
     paddd                m0, [r3+ 0*16]
@@ -5162,11 +5169,11 @@ cglobal inv_txfm_add_dct_dct_8x32_16bpc, 4, 7, 15, 0-36*16, \
     call m(idct_8x8_internal_16bpc).round1_and_write_8x8
     ret
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 8
-    add                 r5d, 10240
-    sar                 r5d, 14
+    add                 r5d, 640
+    sar                 r5d, 10
     add                 rsp, (31+2*ARCH_X86_64)*16
     jmp m(inv_txfm_add_dct_dct_8x8_16bpc).end2
 
@@ -5339,12 +5346,12 @@ cglobal inv_txfm_add_dct_dct_16x32_16bpc, 4, 7, 16, 0-77*16, \
 %endif
     RET
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 32
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
     add                 rsp, (65+4*ARCH_X86_64)*16
     jmp m(inv_txfm_add_dct_dct_16x4_16bpc).dconly
 
@@ -5944,6 +5951,8 @@ cglobal inv_txfm_add_dct_dct_32x8_16bpc, 4, 7, 16, 0-(24+8*ARCH_X86_32)*16, \
     ; final sumsub for idct16 as well as idct32, plus final downshift
 %macro IDCT32_END 6 ; in/out1, out2-4, tmp, shift, idx
     mova                m%4, [r3+16*(23-%1)]
+    pmaxsd              m%1, m12
+    pminsd              m%1, m13
     psubd               m%3, m%1, m%4 ; idct16 out15 - n
     paddd               m%1, m%4      ; idct16 out0  + n
     pmaxsd              m%1, m12
@@ -6019,6 +6028,8 @@ cglobal inv_txfm_add_dct_dct_32x8_16bpc, 4, 7, 16, 0-(24+8*ARCH_X86_32)*16, \
 .loop_dct32_end:
     mova                 m0, [r3+16*16]
     mova                 m6, [r3+16*24]
+    pmaxsd               m0, m2
+    pminsd               m0, m3
     psubd                m5, m0, m6 ; idct16 out15 - n
     paddd                m0, m6     ; idct16 out0  + n
     pmaxsd               m0, m2
@@ -6045,12 +6056,12 @@ cglobal inv_txfm_add_dct_dct_32x8_16bpc, 4, 7, 16, 0-(24+8*ARCH_X86_32)*16, \
 %endif
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 8
 .dconly1:
-    add                 r5d, 10240
-    sar                 r5d, 14
+    add                 r5d, 640
+    sar                 r5d, 10
 .dconly2:
     imul                r5d, 2896
     add                 r5d, 34816
@@ -6344,14 +6355,14 @@ cglobal inv_txfm_add_dct_dct_32x16_16bpc, 4, 7, 16, 0-(24+8*ARCH_X86_32)*16, \
 %endif
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 16
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
-    add                 r5d, 6144
-    sar                 r5d, 13
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
+    add                 r5d, 384
+    sar                 r5d, 9
     jmp m(inv_txfm_add_dct_dct_32x8_16bpc).dconly2
 
 cglobal inv_txfm_add_dct_dct_32x32_16bpc, 4, 7, 16, 0-(5*32+1)*16, \
@@ -6565,7 +6576,7 @@ cglobal inv_txfm_add_dct_dct_32x32_16bpc, 4, 7, 16, 0-(5*32+1)*16, \
     jmp m(inv_txfm_add_dct_dct_16x32_16bpc).loop_pass2_entry
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 32
     add                 rsp, (5*32+1-(24+8*ARCH_X86_32))*16
@@ -6838,11 +6849,11 @@ cglobal inv_txfm_add_dct_dct_16x64_16bpc, 4, 7, 16, \
     ret
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 64
-    add                 r5d, 10240
-    sar                 r5d, 14
+    add                 r5d, 640
+    sar                 r5d, 10
     add                 rsp, (12+2*64)*16+(4+4*ARCH_X86_32)*gprsize-(8+4*ARCH_X86_32)*16
     jmp m(inv_txfm_add_dct_dct_16x4_16bpc).dconly2
 
@@ -7098,14 +7109,14 @@ cglobal inv_txfm_add_dct_dct_32x64_16bpc, 4, 7, 16, \
     jmp m(inv_txfm_add_dct_dct_16x64_16bpc).loop_pass2
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 64
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
-    add                 r5d, 6144
-    sar                 r5d, 13
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
+    add                 r5d, 384
+    sar                 r5d, 9
     add                 rsp, (32+4*64)*16+(4+4*ARCH_X86_32)*gprsize-(24+8*ARCH_X86_32)*16
     jmp m(inv_txfm_add_dct_dct_32x8_16bpc).dconly2
 
@@ -7537,6 +7548,8 @@ cglobal inv_txfm_add_dct_dct_64x16_16bpc, 4, 7, 16, 0-(64+8*ARCH_X86_32)*16, \
     mova                 m5, [r3-16* 4] ; idct64 48 + n
     mova                 m6, [r4-16*20] ; idct64 47 - n
     mova                 m7, [r3-16*20] ; idct64 32 + n
+    pmaxsd               m0, m12
+    pminsd               m0, m13
     paddd                m8, m0, m1     ; idct16 out0  + n
     psubd                m0, m1         ; idct16 out15 - n
     REPX    {pmaxsd x, m12}, m8, m0
@@ -7565,11 +7578,13 @@ cglobal inv_txfm_add_dct_dct_64x16_16bpc, 4, 7, 16, 0-(64+8*ARCH_X86_32)*16, \
     mova         [r4-16* 4], m6
     mova         [r3+16*12], m8
 %else
+    mova                 m5, [o(clip_18b_min)]
+    mova                 m6, [o(clip_18b_max)]
     mova                 m1, [r3+16*44] ; idct16 15 - n
+    pmaxsd               m0, m5
+    pminsd               m0, m6
     paddd                m4, m0, m1     ; idct16 out0  + n
     psubd                m0, m1         ; idct16 out15 - n
-    mova                 m5, [o(clip_18b_min)]
-    mova                 m6, [o(clip_18b_max)]
     REPX     {pmaxsd x, m5}, m4, m0
     REPX     {pminsd x, m6}, m4, m0
     paddd                m1, m4, m3     ; idct32 out0  + n
@@ -7632,12 +7647,12 @@ cglobal inv_txfm_add_dct_dct_64x16_16bpc, 4, 7, 16, 0-(64+8*ARCH_X86_32)*16, \
     ret
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 16
 .dconly1:
-    add                 r5d, 10240
-    sar                 r5d, 14
+    add                 r5d, 640
+    sar                 r5d, 10
 .dconly2:
     imul                r5d, 2896
     add                 r5d, 34816
@@ -7876,14 +7891,14 @@ cglobal inv_txfm_add_dct_dct_64x32_16bpc, 4, 7, 16, \
     ret
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 32
-    add                 r5d, 2048
-    sar                 r5d, 12
-    imul                r5d, 2896
-    add                 r5d, 6144
-    sar                 r5d, 13
+    add                 r5d, 128
+    sar                 r5d, 8
+    imul                r5d, 181
+    add                 r5d, 384
+    sar                 r5d, 9
     add                 rsp, (1+8*32+1*WIN64)*16
     jmp m(inv_txfm_add_dct_dct_64x16_16bpc).dconly2
 
@@ -8112,7 +8127,7 @@ cglobal inv_txfm_add_dct_dct_64x64_16bpc, 4, 7, 16, \
     ret
 
 .dconly:
-    imul                r5d, [cq], 2896
+    imul                r5d, [cq], 181
     mov                [cq], eobd ; 0
     mov                 r3d, 64
     add                 rsp, (64+8*ARCH_X86_32+8*64+1*ARCH_X86_64)*16 + \