qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org, "Alex Bennée" <alex.bennee@linaro.org>
Subject: [PULL 06/30] tcg/i386: Detect AVX512
Date: Thu,  3 Mar 2022 10:59:20 -1000	[thread overview]
Message-ID: <20220303205944.469445-7-richard.henderson@linaro.org> (raw)
In-Reply-To: <20220303205944.469445-1-richard.henderson@linaro.org>

There are some operation sizes in some subsets of AVX512 that
are missing from previous iterations of AVX.  Detect them.

Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/qemu/cpuid.h      | 20 +++++++++++++++++---
 tcg/i386/tcg-target.h     |  4 ++++
 tcg/i386/tcg-target.c.inc | 24 ++++++++++++++++++++++--
 3 files changed, 43 insertions(+), 5 deletions(-)

diff --git a/include/qemu/cpuid.h b/include/qemu/cpuid.h
index 09fc245b91..7adb12d320 100644
--- a/include/qemu/cpuid.h
+++ b/include/qemu/cpuid.h
@@ -45,12 +45,26 @@
 #ifndef bit_AVX2
 #define bit_AVX2        (1 << 5)
 #endif
-#ifndef bit_AVX512F
-#define bit_AVX512F        (1 << 16)
-#endif
 #ifndef bit_BMI2
 #define bit_BMI2        (1 << 8)
 #endif
+#ifndef bit_AVX512F
+#define bit_AVX512F     (1 << 16)
+#endif
+#ifndef bit_AVX512DQ
+#define bit_AVX512DQ    (1 << 17)
+#endif
+#ifndef bit_AVX512BW
+#define bit_AVX512BW    (1 << 30)
+#endif
+#ifndef bit_AVX512VL
+#define bit_AVX512VL    (1u << 31)
+#endif
+
+/* Leaf 7, %ecx */
+#ifndef bit_AVX512VBMI2
+#define bit_AVX512VBMI2 (1 << 6)
+#endif
 
 /* Leaf 0x80000001, %ecx */
 #ifndef bit_LZCNT
diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index ecd0fa6e05..79af353860 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -103,6 +103,10 @@ extern bool have_bmi1;
 extern bool have_popcnt;
 extern bool have_avx1;
 extern bool have_avx2;
+extern bool have_avx512bw;
+extern bool have_avx512dq;
+extern bool have_avx512vbmi2;
+extern bool have_avx512vl;
 extern bool have_movbe;
 
 /* optional instructions */
diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index faa15eecab..7516be5d5f 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -171,6 +171,10 @@ bool have_bmi1;
 bool have_popcnt;
 bool have_avx1;
 bool have_avx2;
+bool have_avx512bw;
+bool have_avx512dq;
+bool have_avx512vbmi2;
+bool have_avx512vl;
 bool have_movbe;
 
 #ifdef CONFIG_CPUID_H
@@ -3839,12 +3843,12 @@ static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
 static void tcg_target_init(TCGContext *s)
 {
 #ifdef CONFIG_CPUID_H
-    unsigned a, b, c, d, b7 = 0;
+    unsigned a, b, c, d, b7 = 0, c7 = 0;
     unsigned max = __get_cpuid_max(0, 0);
 
     if (max >= 7) {
         /* BMI1 is available on AMD Piledriver and Intel Haswell CPUs.  */
-        __cpuid_count(7, 0, a, b7, c, d);
+        __cpuid_count(7, 0, a, b7, c7, d);
         have_bmi1 = (b7 & bit_BMI) != 0;
         have_bmi2 = (b7 & bit_BMI2) != 0;
     }
@@ -3874,6 +3878,22 @@ static void tcg_target_init(TCGContext *s)
             if ((xcrl & 6) == 6) {
                 have_avx1 = (c & bit_AVX) != 0;
                 have_avx2 = (b7 & bit_AVX2) != 0;
+
+                /*
+                 * There are interesting instructions in AVX512, so long
+                 * as we have AVX512VL, which indicates support for EVEX
+                 * on sizes smaller than 512 bits.  We are required to
+                 * check that OPMASK and all extended ZMM state are enabled
+                 * even if we're not using them -- the insns will fault.
+                 */
+                if ((xcrl & 0xe0) == 0xe0
+                    && (b7 & bit_AVX512F)
+                    && (b7 & bit_AVX512VL)) {
+                    have_avx512vl = true;
+                    have_avx512bw = (b7 & bit_AVX512BW) != 0;
+                    have_avx512dq = (b7 & bit_AVX512DQ) != 0;
+                    have_avx512vbmi2 = (c7 & bit_AVX512VBMI2) != 0;
+                }
             }
         }
     }
-- 
2.25.1



  parent reply	other threads:[~2022-03-03 21:13 UTC|newest]

Thread overview: 33+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-03-03 20:59 [PULL 00/30] tcg patch queue Richard Henderson
2022-03-03 20:59 ` [PULL 01/30] tcg/optimize: only read val after const check Richard Henderson
2022-03-03 20:59 ` [PULL 02/30] tcg: Set MAX_OPC_PARAM_IARGS to 7 Richard Henderson
2022-03-03 20:59 ` [PULL 03/30] tcg: Add opcodes for vector nand, nor, eqv Richard Henderson
2022-03-03 20:59 ` [PULL 04/30] tcg/ppc: Implement vector NAND, NOR, EQV Richard Henderson
2022-03-03 20:59 ` [PULL 05/30] tcg/s390x: " Richard Henderson
2022-03-03 20:59 ` Richard Henderson [this message]
2022-03-03 20:59 ` [PULL 07/30] tcg/i386: Add tcg_out_evex_opc Richard Henderson
2022-03-03 20:59 ` [PULL 08/30] tcg/i386: Use tcg_can_emit_vec_op in expand_vec_cmp_noinv Richard Henderson
2022-03-03 20:59 ` [PULL 09/30] tcg/i386: Implement avx512 variable shifts Richard Henderson
2022-03-03 20:59 ` [PULL 10/30] tcg/i386: Implement avx512 scalar shift Richard Henderson
2022-03-03 20:59 ` [PULL 11/30] tcg/i386: Implement avx512 immediate sari shift Richard Henderson
2022-03-03 20:59 ` [PULL 12/30] tcg/i386: Implement avx512 immediate rotate Richard Henderson
2022-03-03 20:59 ` [PULL 13/30] tcg/i386: Implement avx512 variable rotate Richard Henderson
2022-03-03 20:59 ` [PULL 14/30] tcg/i386: Support avx512vbmi2 vector shift-double instructions Richard Henderson
2022-03-03 20:59 ` [PULL 15/30] tcg/i386: Expand vector word rotate as avx512vbmi2 shift-double Richard Henderson
2022-03-03 20:59 ` [PULL 16/30] tcg/i386: Remove rotls_vec from tcg_target_op_def Richard Henderson
2022-03-03 20:59 ` [PULL 17/30] tcg/i386: Expand scalar rotate with avx512 insns Richard Henderson
2022-03-03 20:59 ` [PULL 18/30] tcg/i386: Implement avx512 min/max/abs Richard Henderson
2022-03-03 20:59 ` [PULL 19/30] tcg/i386: Implement avx512 multiply Richard Henderson
2022-03-03 20:59 ` [PULL 20/30] tcg/i386: Implement more logical operations for avx512 Richard Henderson
2022-03-03 20:59 ` [PULL 21/30] tcg/i386: Implement bitsel " Richard Henderson
2022-03-03 20:59 ` [PULL 22/30] tcg: Add TCG_TARGET_SIGNED_ADDR32 Richard Henderson
2022-03-03 20:59 ` [PULL 23/30] accel/tcg: Split out g2h_tlbe Richard Henderson
2022-03-03 20:59 ` [PULL 24/30] accel/tcg: Support TCG_TARGET_SIGNED_ADDR32 for softmmu Richard Henderson
2022-03-03 20:59 ` [PULL 25/30] accel/tcg: Add guest_base_signed_addr32 for user-only Richard Henderson
2022-03-03 20:59 ` [PULL 26/30] linux-user: Support TCG_TARGET_SIGNED_ADDR32 Richard Henderson
2022-03-03 20:59 ` [PULL 27/30] tcg/aarch64: " Richard Henderson
2022-03-03 20:59 ` [PULL 28/30] tcg/mips: " Richard Henderson
2022-03-03 20:59 ` [PULL 29/30] tcg/riscv: " Richard Henderson
2022-03-03 20:59 ` [PULL 30/30] tcg/loongarch64: " Richard Henderson
2022-03-04 15:22 ` [PULL 00/30] tcg patch queue Peter Maydell
2022-03-04 18:47   ` Richard Henderson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20220303205944.469445-7-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=alex.bennee@linaro.org \
    --cc=peter.maydell@linaro.org \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).