[PATCH v4 06/16] x86emul: AVX10.1 testing

[Jan Beulich <jbeulich@xxxxxxxx>]

Re-use respective AVX512 tests, by suitably adjusting the predicate
functions. This leaves test names ("Testing ... NN-bit code sequence")
somewhat misleading, but I think we can live with that.

Note that the AVX512{BW,DQ} opmask tests cannot be run as-is for the
AVX10/256 case, as they include 512-bit vector <-> opmask insn tests.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
SDE: -gnr / -gnr256
---
TBD: The testcase.mk change was for AVX10.1/256. It may, however, be
     worthwhile to keep it.

TBD: May want to have separate AVX10 test blobs built, utilizing all of
     the implied AVX512 sub-features.

The AVX512{BW,DQ} opmask tests could of course be cloned (i.e. rebuilt
another time with -mavx512vl passed) accordingly, but the coverage gain
would be pretty marginal.
---
v4: Drop use of vsz<N> fields. Series re-ordering adjustments. Drop stale
    Phi leftovers. Re-base.
v2: Drop SDE 9.27.0 workaround. Re-base over dropping of Xeon Phi
    support.

--- a/tools/tests/x86_emulator/evex-disp8.c
+++ b/tools/tests/x86_emulator/evex-disp8.c
@@ -1026,7 +1026,7 @@ static void test_group(const struct test
         for ( j = 0; j < nr_vl; ++j )
         {
             if ( vl[0] == VL_512 && vl[j] != VL_512 &&
-                 !cpu_policy.feat.avx512vl )
+                 !cpu_policy.feat.avx512vl && !cpu_policy.feat.avx10 )
                 continue;
 
             switch ( tests[i].esz )
@@ -1077,6 +1077,24 @@ static void test_group(const struct test
     }
 }
 
+/* AVX512 (sub)features implied by AVX10. */
+#define avx10_has_avx512f             cpu_has_avx10
+#define avx10_has_avx512bw            cpu_has_avx10
+#define avx10_has_avx512cd            cpu_has_avx10
+#define avx10_has_avx512dq            cpu_has_avx10
+#define avx10_has_avx512_bf16         cpu_has_avx10
+#define avx10_has_avx512_bitalg       cpu_has_avx10
+#define avx10_has_avx512_fp16         cpu_has_avx10
+#define avx10_has_avx512_ifma         cpu_has_avx10
+#define avx10_has_avx512_vbmi         cpu_has_avx10
+#define avx10_has_avx512_vbmi2        cpu_has_avx10
+#define avx10_has_avx512_vnni         cpu_has_avx10
+#define avx10_has_avx512_vpopcntdq    cpu_has_avx10
+
+/* AVX512 sub-feature(s) /not/ implied by AVX10. */
+#define avx10_has_avx512_vp2intersect false
+#define avx10_has_avx512_bmm          false
+
 void evex_disp8_test(void *instr, struct x86_emulate_ctxt *ctxt,
                      const struct x86_emulate_ops *ops)
 {
@@ -1094,7 +1112,7 @@ void evex_disp8_test(void *instr, struct
     } \
 } while ( false )
 
-#define RUN(feat, vl) run(cpu_has_ ## feat, feat, vl)
+#define RUN(feat, vl) run(cpu_has_ ## feat || avx10_has_ ## feat, feat, vl)
 
     RUN(avx512f, all);
     RUN(avx512f, 128);
@@ -1120,11 +1138,15 @@ void evex_disp8_test(void *instr, struct
     RUN(avx512_bmm, all);
     RUN(avx512_bmm, no128);
 
-    if ( cpu_has_avx512f )
+#undef RUN
+
+    if ( cpu_has_avx512f || cpu_has_avx10 )
     {
+#define RUN(feat, vl) run(cpu_has_ ## feat, feat, vl)
         RUN(gfni, all);
         RUN(vaes, all);
         RUN(vpclmulqdq, all);
+#undef RUN
     }
 
     if ( cpu_has_avx10 )
--- a/tools/tests/x86_emulator/testcase.mk
+++ b/tools/tests/x86_emulator/testcase.mk
@@ -7,6 +7,22 @@ $(call cc-options-add,CFLAGS,CC,$(EMBEDD
 CFLAGS += -ffreestanding -nostdinc -I$(XEN_ROOT)/tools/firmware/include
 CFLAGS += -fno-stack-protector -g0 $($(TESTCASE)-cflags)
 
+ifneq ($(filter -mavx512%,$($(TESTCASE)-cflags)),)
+
+cflags-vsz64 :=
+cflags-vsz32 := -mprefer-vector-width=256
+cflags-vsz16 := -mprefer-vector-width=128
+# Scalar tests don't set VEC_SIZE (and VEC_MAX is used by S/G ones only)
+cflags-vsz   := -mprefer-vector-width=128
+
+ifneq ($(filter -DVEC_SIZE=%,$($(TESTCASE)-cflags)),)
+CFLAGS += $(cflags-vsz$(patsubst -DVEC_SIZE=%,%,$(filter 
-DVEC_SIZE=%,$($(TESTCASE)-cflags))))
+else
+CFLAGS += $(cflags-vsz$(patsubst -DVEC_MAX=%,%,$(filter 
-DVEC_MAX=%,$($(TESTCASE)-cflags))))
+endif
+
+endif
+
 LDFLAGS_DIRECT += $(shell { $(LD) -v --warn-rwx-segments; } >/dev/null 2>&1 && 
echo --no-warn-rwx-segments)
 
 .PHONY: all
--- a/tools/tests/x86_emulator/test_x86_emulator.c
+++ b/tools/tests/x86_emulator/test_x86_emulator.c
@@ -124,52 +124,56 @@ static bool simd_check_avx_pclmul(void)
 
 static bool simd_check_avx512f(void)
 {
-    return cpu_has_avx512f;
+    return cpu_has_avx512f || cpu_has_avx10;
 }
 #define simd_check_avx512f_opmask simd_check_avx512f
 #define simd_check_avx512f_sg simd_check_avx512f
 
 static bool simd_check_avx512f_vl(void)
 {
-    return cpu_has_avx512f && cpu_policy.feat.avx512vl;
+    return (cpu_has_avx512f && cpu_policy.feat.avx512vl) ||
+           cpu_has_avx10;
 }
 #define simd_check_avx512vl_sg simd_check_avx512f_vl
 
 static bool simd_check_avx512dq(void)
 {
-    return cpu_has_avx512dq;
+    return cpu_has_avx512dq || cpu_has_avx10;
 }
 #define simd_check_avx512dq_opmask simd_check_avx512dq
 
 static bool simd_check_avx512dq_vl(void)
 {
-    return cpu_has_avx512dq && cpu_policy.feat.avx512vl;
+    return (cpu_has_avx512dq && cpu_policy.feat.avx512vl) ||
+           cpu_has_avx10;
 }
 
 static bool simd_check_avx512bw(void)
 {
-    return cpu_has_avx512bw;
+    return cpu_has_avx512bw || cpu_has_avx10;
 }
 #define simd_check_avx512bw_opmask simd_check_avx512bw
 
 static bool simd_check_avx512bw_vl(void)
 {
-    return cpu_has_avx512bw && cpu_policy.feat.avx512vl;
+    return (cpu_has_avx512bw && cpu_policy.feat.avx512vl) ||
+           cpu_has_avx10;
 }
 
 static bool simd_check_avx512vbmi(void)
 {
-    return cpu_has_avx512_vbmi;
+    return cpu_has_avx512_vbmi || cpu_has_avx10;
 }
 
 static bool simd_check_avx512vbmi_vl(void)
 {
-    return cpu_has_avx512_vbmi && cpu_policy.feat.avx512vl;
+    return (cpu_has_avx512_vbmi && cpu_policy.feat.avx512vl) ||
+           cpu_has_avx10;
 }
 
 static bool simd_check_avx512vbmi2(void)
 {
-    return cpu_has_avx512_vbmi2;
+    return cpu_has_avx512_vbmi2 || cpu_has_avx10;
 }
 
 static bool simd_check_sse4_sha(void)
@@ -250,17 +254,18 @@ static bool simd_check_avx512bw_gf_vl(vo
 
 static bool simd_check_avx512vnni(void)
 {
-    return cpu_has_avx512_vnni;
+    return cpu_has_avx512_vnni || cpu_has_avx10;
 }
 
 static bool simd_check_avx512fp16(void)
 {
-    return cpu_has_avx512_fp16;
+    return cpu_has_avx512_fp16 || cpu_has_avx10;
 }
 
 static bool simd_check_avx512fp16_vl(void)
 {
-    return cpu_has_avx512_fp16 && cpu_policy.feat.avx512vl;
+    return (cpu_has_avx512_fp16 && cpu_policy.feat.avx512vl) ||
+           cpu_has_avx10;
 }
 
 static void simd_set_regs(struct cpu_user_regs *regs)
@@ -3238,11 +3243,22 @@ int main(int argc, char **argv)
         rc = x86_emulate(&ctxt, &emulops);
         if ( rc != X86EMUL_OKAY || !check_eip(evex_vmovq_from_mem) )
             goto fail;
-        asm ( "vmovq %1, %%xmm1\n\t"
-              "vpcmpeqq %%zmm0, %%zmm1, %%k0\n"
-              "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
-        if ( rc != 0xff )
-            goto fail;
+        if ( simd_check_avx512f() )
+        {
+            asm ( "vmovq %1, %%xmm1\n\t"
+                  "vpcmpeqq %%zmm0, %%zmm1, %%k0\n"
+                  "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0x00ff )
+                goto fail;
+        }
+        else
+        {
+            asm ( "vmovq %1, %%xmm1\n\t"
+                  "vpcmpeqq %%xmm0, %%xmm1, %%k0\n"
+                  "kmovb %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0x03 )
+                goto fail;
+        }
         printf("okay\n");
     }
     else
@@ -3820,11 +3836,22 @@ int main(int argc, char **argv)
         rc = x86_emulate(&ctxt, &emulops);
         if ( rc != X86EMUL_OKAY || !check_eip(evex_vmovd_from_mem) )
             goto fail;
-        asm ( "vmovd %1, %%xmm0\n\t"
-              "vpcmpeqd %%zmm4, %%zmm0, %%k0\n\t"
-              "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
-        if ( rc != 0xffff )
-            goto fail;
+        if ( simd_check_avx512f() )
+        {
+            asm ( "vmovd %1, %%xmm0\n\t"
+                  "vpcmpeqd %%zmm4, %%zmm0, %%k0\n\t"
+                  "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0xffff )
+                goto fail;
+        }
+        else
+        {
+            asm ( "vmovd %1, %%xmm0\n\t"
+                  "vpcmpeqd %%xmm4, %%xmm0, %%k0\n\t"
+                  "kmovb %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0x0f )
+                goto fail;
+        }
         printf("okay\n");
     }
     else
@@ -4037,11 +4064,22 @@ int main(int argc, char **argv)
         rc = x86_emulate(&ctxt, &emulops);
         if ( (rc != X86EMUL_OKAY) || !check_eip(evex_vmovd_from_reg) )
             goto fail;
-        asm ( "vmovd %1, %%xmm0\n\t"
-              "vpcmpeqd %%zmm1, %%zmm0, %%k0\n\t"
-              "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
-        if ( rc != 0xffff )
-            goto fail;
+        if ( simd_check_avx512f() )
+        {
+            asm ( "vmovd %1, %%xmm0\n\t"
+                  "vpcmpeqd %%zmm1, %%zmm0, %%k0\n\t"
+                  "kmovw %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0xffff )
+                goto fail;
+        }
+        else
+        {
+            asm ( "vmovd %1, %%xmm0\n\t"
+                  "vpcmpeqd %%xmm1, %%xmm0, %%k0\n\t"
+                  "kmovb %%k0, %0" : "=r" (rc) : "m" (res[8]) );
+            if ( rc != 0x0f )
+                goto fail;
+        }
         printf("okay\n");
     }
     else
@@ -5524,14 +5562,28 @@ int main(int argc, char **argv)
         rc = x86_emulate(&ctxt, &emulops);
         if ( (rc != X86EMUL_OKAY) || !check_eip(vmovsh_from_mem) )
             goto fail;
-        asm volatile ( "kmovw     %2, %%k1\n\t"
-                       "vmovdqu16 %1, %%zmm4%{%%k1%}%{z%}\n\t"
-                       "vpcmpeqw  %%zmm4, %%zmm5, %%k0\n\t"
-                       "kmovw     %%k0, %0"
-                       : "=g" (rc)
-                       : "m" (res[2]), "r" (1) );
-        if ( rc != 0xffff )
-            goto fail;
+        if ( simd_check_avx512fp16() )
+        {
+            asm volatile ( "kmovw     %2, %%k1\n\t"
+                           "vmovdqu16 %1, %%zmm4%{%%k1%}%{z%}\n\t"
+                           "vpcmpeqw  %%zmm4, %%zmm5, %%k0\n\t"
+                           "kmovw     %%k0, %0"
+                           : "=g" (rc)
+                           : "m" (res[2]), "r" (1) );
+            if ( rc != 0xffff )
+                goto fail;
+        }
+        else
+        {
+            asm volatile ( "kmovb     %2, %%k1\n\t"
+                           "vmovdqu16 %1, %%xmm4%{%%k1%}%{z%}\n\t"
+                           "vpcmpeqw  %%xmm4, %%xmm5, %%k0\n\t"
+                           "kmovb     %%k0, %0"
+                           : "=g" (rc)
+                           : "m" (res[2]), "r" (1) );
+            if ( rc != 0xff )
+                goto fail;
+        }
         printf("okay\n");
 
         printf("%-40s", "Testing vmovsh %xmm4,2(%eax){%k3}...");
--- a/tools/tests/x86_emulator/x86-emulate.c
+++ b/tools/tests/x86_emulator/x86-emulate.c
@@ -244,7 +244,7 @@ int emul_test_get_fpu(
             break;
     case X86EMUL_FPU_opmask:
     case X86EMUL_FPU_zmm:
-        if ( cpu_has_avx512f )
+        if ( cpu_has_avx512f || cpu_has_avx10 )
             break;
     default:
         return X86EMUL_UNHANDLEABLE;