Skip to content

Commit 975c208

Browse files
e-kude-kud
authored
[X86][AVX10.2] Include changes for COMX and VGETEXP from rev. 2 (#132824)
Address missing changes: - V[,U]COMXSD need to have XD (F3.0F –> F2.0F) - V[,U]COMXS[S,H] need to have XS (F2.[0F,MAP5] -> F3.[0F,MAP5]) - VGETEXPBF16 needs to have T_MAP6 and NP (66.MAP5 -> NP.MAP6) Ref.: https://cdrdv2.intel.com/v1/dl/getContent/828965
1 parent 6e66cfe commit 975c208

15 files changed

+472
-472
lines changed

‎llvm/lib/Target/X86/X86InstrAVX10.td

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1141,7 +1141,7 @@ defm VRSQRT : avx10_fp14_bf16<0x4E, "vrsqrt", X86rsqrt14, SchedWriteFRsqrt>,
11411141
defm VRCP : avx10_fp14_bf16<0x4C, "vrcp", X86rcp14, SchedWriteFRcp>,
11421142
T_MAP6, PS, EVEX_CD8<16, CD8VF>;
11431143
defm VGETEXP : avx10_fp14_bf16<0x42, "vgetexp", X86fgetexp, SchedWriteFRnd>,
1144-
T_MAP5, EVEX_CD8<16, CD8VF>;
1144+
T_MAP6, PS, EVEX_CD8<16, CD8VF>;
11451145

11461146
// VSCALEFBF16
11471147
multiclass avx10_fp_scalef_bf16<bits<8> opc, string OpcodeStr,
@@ -1338,31 +1338,31 @@ multiclass avx10_com_ef_int<bits<8> Opc, X86VectorVTInfo _, SDNode OpNode,
13381338
let Defs = [EFLAGS], Uses = [MXCSR], Predicates = [HasAVX10_2] in {
13391339
defm VUCOMXSDZ : avx10_com_ef<0x2e, FR64X, f64, X86ucomi512,
13401340
"vucomxsd", f64mem, loadf64, SSEPackedDouble>,
1341-
TB, XS, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
1341+
TB, XD, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
13421342
defm VUCOMXSHZ : avx10_com_ef<0x2e, FR16X, f16, X86ucomi512,
13431343
"vucomxsh", f16mem, loadf16, SSEPackedSingle>,
1344-
T_MAP5, XD, EVEX_CD8<16, CD8VT1>;
1344+
T_MAP5, XS, EVEX_CD8<16, CD8VT1>;
13451345
defm VUCOMXSSZ : avx10_com_ef<0x2e, FR32X, f32, X86ucomi512,
13461346
"vucomxss", f32mem, loadf32, SSEPackedSingle>,
1347-
TB, XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
1347+
TB, XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
13481348
defm VCOMXSDZ : avx10_com_ef_int<0x2f, v2f64x_info, X86comi512,
13491349
"vcomxsd", SSEPackedDouble>,
1350-
TB, XS, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
1350+
TB, XD, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
13511351
defm VCOMXSHZ : avx10_com_ef_int<0x2f, v8f16x_info, X86comi512,
13521352
"vcomxsh", SSEPackedSingle>,
1353-
T_MAP5, XD, EVEX_CD8<16, CD8VT1>;
1353+
T_MAP5, XS, EVEX_CD8<16, CD8VT1>;
13541354
defm VCOMXSSZ : avx10_com_ef_int<0x2f, v4f32x_info, X86comi512,
13551355
"vcomxss", SSEPackedSingle>,
1356-
TB, XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
1356+
TB, XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
13571357
defm VUCOMXSDZ : avx10_com_ef_int<0x2e, v2f64x_info, X86ucomi512,
13581358
"vucomxsd", SSEPackedDouble>,
1359-
TB, XS, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
1359+
TB, XD, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
13601360
defm VUCOMXSHZ : avx10_com_ef_int<0x2e, v8f16x_info, X86ucomi512,
13611361
"vucomxsh", SSEPackedSingle>,
1362-
T_MAP5, XD, EVEX_CD8<16, CD8VT1>;
1362+
T_MAP5, XS, EVEX_CD8<16, CD8VT1>;
13631363
defm VUCOMXSSZ : avx10_com_ef_int<0x2e, v4f32x_info, X86ucomi512,
13641364
"vucomxss", SSEPackedSingle>,
1365-
TB, XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
1365+
TB, XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
13661366
}
13671367

13681368
//-------------------------------------------------

‎llvm/test/CodeGen/X86/avx10_2_512bf16-intrinsics.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -164,15 +164,15 @@ define <32 x bfloat>@test_int_x86_avx512_mask_getexp_bf16_512(<32 x bfloat> %x0,
164164
; X64-LABEL: test_int_x86_avx512_mask_getexp_bf16_512:
165165
; X64: # %bb.0:
166166
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
167-
; X64-NEXT: vgetexpbf16 %zmm0, %zmm0 # encoding: [0x62,0xf5,0x7d,0x48,0x42,0xc0]
167+
; X64-NEXT: vgetexpbf16 %zmm0, %zmm0 # encoding: [0x62,0xf6,0x7c,0x48,0x42,0xc0]
168168
; X64-NEXT: vmovdqu16 %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xff,0x49,0x6f,0xc8]
169169
; X64-NEXT: vaddbf16 %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf5,0x75,0x48,0x58,0xc0]
170170
; X64-NEXT: retq # encoding: [0xc3]
171171
;
172172
; X86-LABEL: test_int_x86_avx512_mask_getexp_bf16_512:
173173
; X86: # %bb.0:
174174
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
175-
; X86-NEXT: vgetexpbf16 %zmm0, %zmm0 # encoding: [0x62,0xf5,0x7d,0x48,0x42,0xc0]
175+
; X86-NEXT: vgetexpbf16 %zmm0, %zmm0 # encoding: [0x62,0xf6,0x7c,0x48,0x42,0xc0]
176176
; X86-NEXT: vmovdqu16 %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0xff,0x49,0x6f,0xc8]
177177
; X86-NEXT: vaddbf16 %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf5,0x75,0x48,0x58,0xc0]
178178
; X86-NEXT: retl # encoding: [0xc3]

‎llvm/test/CodeGen/X86/avx10_2bf16-intrinsics.ll

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -333,7 +333,7 @@ declare <16 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.256(<16 x bfloat>, <16 x
333333
define <8 x bfloat>@test_int_x86_avx512_getexp_bf16_128(<8 x bfloat> %x0) {
334334
; CHECK-LABEL: test_int_x86_avx512_getexp_bf16_128:
335335
; CHECK: # %bb.0:
336-
; CHECK-NEXT: vgetexpbf16 %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x42,0xc0]
336+
; CHECK-NEXT: vgetexpbf16 %xmm0, %xmm0 # encoding: [0x62,0xf6,0x7c,0x08,0x42,0xc0]
337337
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
338338
%res = call <8 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.128(<8 x bfloat> %x0, <8 x bfloat> zeroinitializer, i8 -1)
339339
ret <8 x bfloat> %res
@@ -343,14 +343,14 @@ define <8 x bfloat>@test_int_x86_avx512_mask_getexp_bf16_128(<8 x bfloat> %x0, <
343343
; X64-LABEL: test_int_x86_avx512_mask_getexp_bf16_128:
344344
; X64: # %bb.0:
345345
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
346-
; X64-NEXT: vgetexpbf16 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x42,0xc8]
346+
; X64-NEXT: vgetexpbf16 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0x42,0xc8]
347347
; X64-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
348348
; X64-NEXT: retq # encoding: [0xc3]
349349
;
350350
; X86-LABEL: test_int_x86_avx512_mask_getexp_bf16_128:
351351
; X86: # %bb.0:
352352
; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
353-
; X86-NEXT: vgetexpbf16 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x42,0xc8]
353+
; X86-NEXT: vgetexpbf16 %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0x42,0xc8]
354354
; X86-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
355355
; X86-NEXT: retl # encoding: [0xc3]
356356
%res = call <8 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.128(<8 x bfloat> %x0, <8 x bfloat> %x1, i8 %x2)
@@ -361,13 +361,13 @@ define <8 x bfloat>@test_int_x86_avx512_maskz_getexp_bf16_128(<8 x bfloat> %x0,
361361
; X64-LABEL: test_int_x86_avx512_maskz_getexp_bf16_128:
362362
; X64: # %bb.0:
363363
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
364-
; X64-NEXT: vgetexpbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0x89,0x42,0xc0]
364+
; X64-NEXT: vgetexpbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x7c,0x89,0x42,0xc0]
365365
; X64-NEXT: retq # encoding: [0xc3]
366366
;
367367
; X86-LABEL: test_int_x86_avx512_maskz_getexp_bf16_128:
368368
; X86: # %bb.0:
369369
; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
370-
; X86-NEXT: vgetexpbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0x89,0x42,0xc0]
370+
; X86-NEXT: vgetexpbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x7c,0x89,0x42,0xc0]
371371
; X86-NEXT: retl # encoding: [0xc3]
372372
%res = call <8 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.128(<8 x bfloat> %x0, <8 x bfloat> zeroinitializer, i8 %x2)
373373
ret <8 x bfloat> %res
@@ -376,7 +376,7 @@ define <8 x bfloat>@test_int_x86_avx512_maskz_getexp_bf16_128(<8 x bfloat> %x0,
376376
define <16 x bfloat>@test_int_x86_avx512_getexp_bf16_256(<16 x bfloat> %x0) {
377377
; CHECK-LABEL: test_int_x86_avx512_getexp_bf16_256:
378378
; CHECK: # %bb.0:
379-
; CHECK-NEXT: vgetexpbf16 %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x42,0xc0]
379+
; CHECK-NEXT: vgetexpbf16 %ymm0, %ymm0 # encoding: [0x62,0xf6,0x7c,0x28,0x42,0xc0]
380380
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
381381
%res = call <16 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.256(<16 x bfloat> %x0, <16 x bfloat> zeroinitializer, i16 -1)
382382
ret <16 x bfloat> %res
@@ -386,14 +386,14 @@ define <16 x bfloat>@test_int_x86_avx512_mask_getexp_bf16_256(<16 x bfloat> %x0,
386386
; X64-LABEL: test_int_x86_avx512_mask_getexp_bf16_256:
387387
; X64: # %bb.0:
388388
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
389-
; X64-NEXT: vgetexpbf16 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x42,0xc8]
389+
; X64-NEXT: vgetexpbf16 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0x42,0xc8]
390390
; X64-NEXT: vmovaps %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc1]
391391
; X64-NEXT: retq # encoding: [0xc3]
392392
;
393393
; X86-LABEL: test_int_x86_avx512_mask_getexp_bf16_256:
394394
; X86: # %bb.0:
395395
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
396-
; X86-NEXT: vgetexpbf16 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x42,0xc8]
396+
; X86-NEXT: vgetexpbf16 %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0x42,0xc8]
397397
; X86-NEXT: vmovaps %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc1]
398398
; X86-NEXT: retl # encoding: [0xc3]
399399
%res = call <16 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.256(<16 x bfloat> %x0, <16 x bfloat> %x1, i16 %x2)
@@ -404,13 +404,13 @@ define <16 x bfloat>@test_int_x86_avx512_maskz_getexp_bf16_256(<16 x bfloat> %x0
404404
; X64-LABEL: test_int_x86_avx512_maskz_getexp_bf16_256:
405405
; X64: # %bb.0:
406406
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
407-
; X64-NEXT: vgetexpbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0xa9,0x42,0xc0]
407+
; X64-NEXT: vgetexpbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x7c,0xa9,0x42,0xc0]
408408
; X64-NEXT: retq # encoding: [0xc3]
409409
;
410410
; X86-LABEL: test_int_x86_avx512_maskz_getexp_bf16_256:
411411
; X86: # %bb.0:
412412
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
413-
; X86-NEXT: vgetexpbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0xa9,0x42,0xc0]
413+
; X86-NEXT: vgetexpbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x7c,0xa9,0x42,0xc0]
414414
; X86-NEXT: retl # encoding: [0xc3]
415415
%res = call <16 x bfloat> @llvm.x86.avx10.mask.getexp.bf16.256(<16 x bfloat> %x0, <16 x bfloat> zeroinitializer, i16 %x2)
416416
ret <16 x bfloat> %res

‎llvm/test/MC/Disassembler/X86/avx10.2-bf16-32.txt

Lines changed: 27 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -1719,111 +1719,111 @@
17191719

17201720
# ATT: vgetexpbf16 %xmm3, %xmm2
17211721
# INTEL: vgetexpbf16 xmm2, xmm3
1722-
0x62,0xf5,0x7d,0x08,0x42,0xd3
1722+
0x62,0xf6,0x7c,0x08,0x42,0xd3
17231723

17241724
# ATT: vgetexpbf16 %xmm3, %xmm2 {%k7}
17251725
# INTEL: vgetexpbf16 xmm2 {k7}, xmm3
1726-
0x62,0xf5,0x7d,0x0f,0x42,0xd3
1726+
0x62,0xf6,0x7c,0x0f,0x42,0xd3
17271727

17281728
# ATT: vgetexpbf16 %xmm3, %xmm2 {%k7} {z}
17291729
# INTEL: vgetexpbf16 xmm2 {k7} {z}, xmm3
1730-
0x62,0xf5,0x7d,0x8f,0x42,0xd3
1730+
0x62,0xf6,0x7c,0x8f,0x42,0xd3
17311731

17321732
# ATT: vgetexpbf16 %zmm3, %zmm2
17331733
# INTEL: vgetexpbf16 zmm2, zmm3
1734-
0x62,0xf5,0x7d,0x48,0x42,0xd3
1734+
0x62,0xf6,0x7c,0x48,0x42,0xd3
17351735

17361736
# ATT: vgetexpbf16 %zmm3, %zmm2 {%k7}
17371737
# INTEL: vgetexpbf16 zmm2 {k7}, zmm3
1738-
0x62,0xf5,0x7d,0x4f,0x42,0xd3
1738+
0x62,0xf6,0x7c,0x4f,0x42,0xd3
17391739

17401740
# ATT: vgetexpbf16 %zmm3, %zmm2 {%k7} {z}
17411741
# INTEL: vgetexpbf16 zmm2 {k7} {z}, zmm3
1742-
0x62,0xf5,0x7d,0xcf,0x42,0xd3
1742+
0x62,0xf6,0x7c,0xcf,0x42,0xd3
17431743

17441744
# ATT: vgetexpbf16 %ymm3, %ymm2
17451745
# INTEL: vgetexpbf16 ymm2, ymm3
1746-
0x62,0xf5,0x7d,0x28,0x42,0xd3
1746+
0x62,0xf6,0x7c,0x28,0x42,0xd3
17471747

17481748
# ATT: vgetexpbf16 %ymm3, %ymm2 {%k7}
17491749
# INTEL: vgetexpbf16 ymm2 {k7}, ymm3
1750-
0x62,0xf5,0x7d,0x2f,0x42,0xd3
1750+
0x62,0xf6,0x7c,0x2f,0x42,0xd3
17511751

17521752
# ATT: vgetexpbf16 %ymm3, %ymm2 {%k7} {z}
17531753
# INTEL: vgetexpbf16 ymm2 {k7} {z}, ymm3
1754-
0x62,0xf5,0x7d,0xaf,0x42,0xd3
1754+
0x62,0xf6,0x7c,0xaf,0x42,0xd3
17551755

17561756
# ATT: vgetexpbf16 268435456(%esp,%esi,8), %xmm2
17571757
# INTEL: vgetexpbf16 xmm2, xmmword ptr [esp + 8*esi + 268435456]
1758-
0x62,0xf5,0x7d,0x08,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
1758+
0x62,0xf6,0x7c,0x08,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
17591759

17601760
# ATT: vgetexpbf16 291(%edi,%eax,4), %xmm2 {%k7}
17611761
# INTEL: vgetexpbf16 xmm2 {k7}, xmmword ptr [edi + 4*eax + 291]
1762-
0x62,0xf5,0x7d,0x0f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
1762+
0x62,0xf6,0x7c,0x0f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
17631763

17641764
# ATT: vgetexpbf16 (%eax){1to8}, %xmm2
17651765
# INTEL: vgetexpbf16 xmm2, word ptr [eax]{1to8}
1766-
0x62,0xf5,0x7d,0x18,0x42,0x10
1766+
0x62,0xf6,0x7c,0x18,0x42,0x10
17671767

17681768
# ATT: vgetexpbf16 -512(,%ebp,2), %xmm2
17691769
# INTEL: vgetexpbf16 xmm2, xmmword ptr [2*ebp - 512]
1770-
0x62,0xf5,0x7d,0x08,0x42,0x14,0x6d,0x00,0xfe,0xff,0xff
1770+
0x62,0xf6,0x7c,0x08,0x42,0x14,0x6d,0x00,0xfe,0xff,0xff
17711771

17721772
# ATT: vgetexpbf16 2032(%ecx), %xmm2 {%k7} {z}
17731773
# INTEL: vgetexpbf16 xmm2 {k7} {z}, xmmword ptr [ecx + 2032]
1774-
0x62,0xf5,0x7d,0x8f,0x42,0x51,0x7f
1774+
0x62,0xf6,0x7c,0x8f,0x42,0x51,0x7f
17751775

17761776
# ATT: vgetexpbf16 -256(%edx){1to8}, %xmm2 {%k7} {z}
17771777
# INTEL: vgetexpbf16 xmm2 {k7} {z}, word ptr [edx - 256]{1to8}
1778-
0x62,0xf5,0x7d,0x9f,0x42,0x52,0x80
1778+
0x62,0xf6,0x7c,0x9f,0x42,0x52,0x80
17791779

17801780
# ATT: vgetexpbf16 268435456(%esp,%esi,8), %ymm2
17811781
# INTEL: vgetexpbf16 ymm2, ymmword ptr [esp + 8*esi + 268435456]
1782-
0x62,0xf5,0x7d,0x28,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
1782+
0x62,0xf6,0x7c,0x28,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
17831783

17841784
# ATT: vgetexpbf16 291(%edi,%eax,4), %ymm2 {%k7}
17851785
# INTEL: vgetexpbf16 ymm2 {k7}, ymmword ptr [edi + 4*eax + 291]
1786-
0x62,0xf5,0x7d,0x2f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
1786+
0x62,0xf6,0x7c,0x2f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
17871787

17881788
# ATT: vgetexpbf16 (%eax){1to16}, %ymm2
17891789
# INTEL: vgetexpbf16 ymm2, word ptr [eax]{1to16}
1790-
0x62,0xf5,0x7d,0x38,0x42,0x10
1790+
0x62,0xf6,0x7c,0x38,0x42,0x10
17911791

17921792
# ATT: vgetexpbf16 -1024(,%ebp,2), %ymm2
17931793
# INTEL: vgetexpbf16 ymm2, ymmword ptr [2*ebp - 1024]
1794-
0x62,0xf5,0x7d,0x28,0x42,0x14,0x6d,0x00,0xfc,0xff,0xff
1794+
0x62,0xf6,0x7c,0x28,0x42,0x14,0x6d,0x00,0xfc,0xff,0xff
17951795

17961796
# ATT: vgetexpbf16 4064(%ecx), %ymm2 {%k7} {z}
17971797
# INTEL: vgetexpbf16 ymm2 {k7} {z}, ymmword ptr [ecx + 4064]
1798-
0x62,0xf5,0x7d,0xaf,0x42,0x51,0x7f
1798+
0x62,0xf6,0x7c,0xaf,0x42,0x51,0x7f
17991799

18001800
# ATT: vgetexpbf16 -256(%edx){1to16}, %ymm2 {%k7} {z}
18011801
# INTEL: vgetexpbf16 ymm2 {k7} {z}, word ptr [edx - 256]{1to16}
1802-
0x62,0xf5,0x7d,0xbf,0x42,0x52,0x80
1802+
0x62,0xf6,0x7c,0xbf,0x42,0x52,0x80
18031803

18041804
# ATT: vgetexpbf16 268435456(%esp,%esi,8), %zmm2
18051805
# INTEL: vgetexpbf16 zmm2, zmmword ptr [esp + 8*esi + 268435456]
1806-
0x62,0xf5,0x7d,0x48,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
1806+
0x62,0xf6,0x7c,0x48,0x42,0x94,0xf4,0x00,0x00,0x00,0x10
18071807

18081808
# ATT: vgetexpbf16 291(%edi,%eax,4), %zmm2 {%k7}
18091809
# INTEL: vgetexpbf16 zmm2 {k7}, zmmword ptr [edi + 4*eax + 291]
1810-
0x62,0xf5,0x7d,0x4f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
1810+
0x62,0xf6,0x7c,0x4f,0x42,0x94,0x87,0x23,0x01,0x00,0x00
18111811

18121812
# ATT: vgetexpbf16 (%eax){1to32}, %zmm2
18131813
# INTEL: vgetexpbf16 zmm2, word ptr [eax]{1to32}
1814-
0x62,0xf5,0x7d,0x58,0x42,0x10
1814+
0x62,0xf6,0x7c,0x58,0x42,0x10
18151815

18161816
# ATT: vgetexpbf16 -2048(,%ebp,2), %zmm2
18171817
# INTEL: vgetexpbf16 zmm2, zmmword ptr [2*ebp - 2048]
1818-
0x62,0xf5,0x7d,0x48,0x42,0x14,0x6d,0x00,0xf8,0xff,0xff
1818+
0x62,0xf6,0x7c,0x48,0x42,0x14,0x6d,0x00,0xf8,0xff,0xff
18191819

18201820
# ATT: vgetexpbf16 8128(%ecx), %zmm2 {%k7} {z}
18211821
# INTEL: vgetexpbf16 zmm2 {k7} {z}, zmmword ptr [ecx + 8128]
1822-
0x62,0xf5,0x7d,0xcf,0x42,0x51,0x7f
1822+
0x62,0xf6,0x7c,0xcf,0x42,0x51,0x7f
18231823

18241824
# ATT: vgetexpbf16 -256(%edx){1to32}, %zmm2 {%k7} {z}
18251825
# INTEL: vgetexpbf16 zmm2 {k7} {z}, word ptr [edx - 256]{1to32}
1826-
0x62,0xf5,0x7d,0xdf,0x42,0x52,0x80
1826+
0x62,0xf6,0x7c,0xdf,0x42,0x52,0x80
18271827

18281828
# ATT: vgetmantbf16 $123, %zmm3, %zmm2
18291829
# INTEL: vgetmantbf16 zmm2, zmm3, 123

0 commit comments

Comments
 (0)