summaryrefslogtreecommitdiffstats
path: root/tcg/i386
diff options
context:
space:
mode:
authorRichard Henderson2021-12-18 19:48:43 +0100
committerRichard Henderson2022-03-04 19:50:41 +0100
commit965d5d0681a99540bf19f313a66ecb14cee2654a (patch)
treef8191543d3411c2d1c6adc686e065b1d5372143d /tcg/i386
parenttcg/i386: Implement avx512 variable rotate (diff)
downloadqemu-965d5d0681a99540bf19f313a66ecb14cee2654a.tar.gz
qemu-965d5d0681a99540bf19f313a66ecb14cee2654a.tar.xz
qemu-965d5d0681a99540bf19f313a66ecb14cee2654a.zip
tcg/i386: Support avx512vbmi2 vector shift-double instructions
We will use VPSHLD, VPSHLDV and VPSHRDV for 16-bit rotates. 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>
Diffstat (limited to 'tcg/i386')
-rw-r--r--tcg/i386/tcg-target-con-set.h1
-rw-r--r--tcg/i386/tcg-target.c.inc38
-rw-r--r--tcg/i386/tcg-target.opc.h3
3 files changed, 42 insertions, 0 deletions
diff --git a/tcg/i386/tcg-target-con-set.h b/tcg/i386/tcg-target-con-set.h
index 78774d1005..91ceb0e1da 100644
--- a/tcg/i386/tcg-target-con-set.h
+++ b/tcg/i386/tcg-target-con-set.h
@@ -45,6 +45,7 @@ C_O1_I2(r, r, rI)
C_O1_I2(x, x, x)
C_N1_I2(r, r, r)
C_N1_I2(r, r, rW)
+C_O1_I3(x, 0, x, x)
C_O1_I3(x, x, x, x)
C_O1_I4(r, r, re, r, 0)
C_O1_I4(r, r, r, ri, ri)
diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index 712ae3a168..a39f890a7d 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -424,6 +424,15 @@ static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
#define OPC_VPROLVQ (0x15 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
#define OPC_VPRORVD (0x14 | P_EXT38 | P_DATA16 | P_EVEX)
#define OPC_VPRORVQ (0x14 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHLDW (0x70 | P_EXT3A | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHLDD (0x71 | P_EXT3A | P_DATA16 | P_EVEX)
+#define OPC_VPSHLDQ (0x71 | P_EXT3A | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHLDVW (0x70 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHLDVD (0x71 | P_EXT38 | P_DATA16 | P_EVEX)
+#define OPC_VPSHLDVQ (0x71 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHRDVW (0x72 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
+#define OPC_VPSHRDVD (0x73 | P_EXT38 | P_DATA16 | P_EVEX)
+#define OPC_VPSHRDVQ (0x73 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
#define OPC_VPSLLVW (0x12 | P_EXT38 | P_DATA16 | P_VEXW | P_EVEX)
#define OPC_VPSLLVD (0x47 | P_EXT38 | P_DATA16)
#define OPC_VPSLLVQ (0x47 | P_EXT38 | P_DATA16 | P_VEXW)
@@ -2867,6 +2876,15 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
static int const sars_insn[4] = {
OPC_UD2, OPC_PSRAW, OPC_PSRAD, OPC_VPSRAQ
};
+ static int const vpshldi_insn[4] = {
+ OPC_UD2, OPC_VPSHLDW, OPC_VPSHLDD, OPC_VPSHLDQ
+ };
+ static int const vpshldv_insn[4] = {
+ OPC_UD2, OPC_VPSHLDVW, OPC_VPSHLDVD, OPC_VPSHLDVQ
+ };
+ static int const vpshrdv_insn[4] = {
+ OPC_UD2, OPC_VPSHRDVW, OPC_VPSHRDVD, OPC_VPSHRDVQ
+ };
static int const abs_insn[4] = {
/* TODO: AVX512 adds support for MO_64. */
OPC_PABSB, OPC_PABSW, OPC_PABSD, OPC_UD2
@@ -2959,6 +2977,16 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
case INDEX_op_x86_packus_vec:
insn = packus_insn[vece];
goto gen_simd;
+ case INDEX_op_x86_vpshldv_vec:
+ insn = vpshldv_insn[vece];
+ a1 = a2;
+ a2 = args[3];
+ goto gen_simd;
+ case INDEX_op_x86_vpshrdv_vec:
+ insn = vpshrdv_insn[vece];
+ a1 = a2;
+ a2 = args[3];
+ goto gen_simd;
#if TCG_TARGET_REG_BITS == 32
case INDEX_op_dup2_vec:
/* First merge the two 32-bit inputs to a single 64-bit element. */
@@ -3061,7 +3089,12 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
insn = OPC_VPERM2I128;
sub = args[3];
goto gen_simd_imm8;
+ case INDEX_op_x86_vpshldi_vec:
+ insn = vpshldi_insn[vece];
+ sub = args[3];
+ goto gen_simd_imm8;
gen_simd_imm8:
+ tcg_debug_assert(insn != OPC_UD2);
if (type == TCG_TYPE_V256) {
insn |= P_VEXL;
}
@@ -3305,6 +3338,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
case INDEX_op_x86_vperm2i128_vec:
case INDEX_op_x86_punpckl_vec:
case INDEX_op_x86_punpckh_vec:
+ case INDEX_op_x86_vpshldi_vec:
#if TCG_TARGET_REG_BITS == 32
case INDEX_op_dup2_vec:
#endif
@@ -3319,6 +3353,10 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
case INDEX_op_x86_psrldq_vec:
return C_O1_I1(x, x);
+ case INDEX_op_x86_vpshldv_vec:
+ case INDEX_op_x86_vpshrdv_vec:
+ return C_O1_I3(x, 0, x, x);
+
case INDEX_op_x86_vpblendvb_vec:
return C_O1_I3(x, x, x, x);
diff --git a/tcg/i386/tcg-target.opc.h b/tcg/i386/tcg-target.opc.h
index 1312941800..b5f403e35e 100644
--- a/tcg/i386/tcg-target.opc.h
+++ b/tcg/i386/tcg-target.opc.h
@@ -33,3 +33,6 @@ DEF(x86_psrldq_vec, 1, 1, 1, IMPLVEC)
DEF(x86_vperm2i128_vec, 1, 2, 1, IMPLVEC)
DEF(x86_punpckl_vec, 1, 2, 0, IMPLVEC)
DEF(x86_punpckh_vec, 1, 2, 0, IMPLVEC)
+DEF(x86_vpshldi_vec, 1, 2, 1, IMPLVEC)
+DEF(x86_vpshldv_vec, 1, 3, 0, IMPLVEC)
+DEF(x86_vpshrdv_vec, 1, 3, 0, IMPLVEC)