[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH 3/7] tests/tcg: Add tests for prefixed load/store instructions
|
From: |
Gustavo Romero |
|
Subject: |
[PATCH 3/7] tests/tcg: Add tests for prefixed load/store instructions |
|
Date: |
Mon, 5 Oct 2020 01:03:15 -0300 |
From: Michael Roth <mdroth@linux.vnet.ibm.com>
This commit adds various tests to exercise the implementation of prefixed
load/store instructions on POWER10.
It adds a softmmu Makefile so tests can be built for the ppc64-softmmu
target correctly with:
$ make build-tcg-tests-ppc64-softmmu
Moreover it fixes missing BUILD_DIR variable in Makefile.include when
calling Makefile.qemu, otherwise config-$(TARGET).mak is not found (include
fails silently) in Makefile.qemu and build finishes with no errors but no
test is build. It also fixes quiet-command in Makefile.qemu, because it's
not defined when cross-build-guest-tests target calls it, so no command is
executed in fact, hence no test is build after make finishes.
Currently check-tcg target is not working, so '$ make check-tcg' fails
because such a target will try to run the generated test binary as if
it were a raw VM image, which is not the case. It's not a bootable image
either, so the binary can't be loaded by SLOF, so after the build it's
necessary to copy the test binary manually and execute it in a VM or a
real POWER10 machine (e.g. for comparison) with support for prefixed
instructions.
Signed-off-by: Michael Roth <mroth@lamentation.net>
[ gromero - fix to avoid alignment interrupt, don't cross 64-byte boundary
- fix displacement for pl{bz,hz,ha,wz,wa,d} to skip branch insn.
- tweaks in debug output
- add Makefile for ppc64-softmmu target
- build fixes in Makefile.{qemu,include}
- commit log ]
Signed-off-by: Gustavo Romero <gromero@linux.ibm.com>
---
tests/Makefile.include | 3 +
tests/tcg/Makefile.qemu | 2 +
tests/tcg/ppc64 | 1 +
tests/tcg/ppc64le/Makefile.softmmu-target | 13 +
.../test-prefixed-load-store.c | 945 ++++++++++++++++++
5 files changed, 964 insertions(+)
create mode 120000 tests/tcg/ppc64
create mode 100644 tests/tcg/ppc64le/Makefile.softmmu-target
create mode 100644
tests/tcg/ppc64le/instruction-tests/test-prefixed-load-store.c
diff --git a/tests/Makefile.include b/tests/Makefile.include
index 40d909badc..fd9887decf 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -54,6 +54,7 @@ build-tcg-tests-%: $(if $(CONFIG_PLUGIN),test-plugins)
$(call quiet-command,$(MAKE) $(SUBDIR_MAKEFLAGS) \
-f $(SRC_PATH)/tests/tcg/Makefile.qemu \
SRC_PATH=$(SRC_PATH) \
+ BUILD_DIR=$(BUILD_DIR) \
V="$(V)" TARGET="$*" guest-tests, \
"BUILD", "TCG tests for $*")
@@ -61,6 +62,7 @@ run-tcg-tests-%: build-tcg-tests-% all
$(call quiet-command,$(MAKE) $(SUBDIR_MAKEFLAGS) \
-f $(SRC_PATH)/tests/tcg/Makefile.qemu \
SRC_PATH=$(SRC_PATH) SPEED="$(SPEED)" \
+ BUILD_DIR=$(BUILD_DIR) \
V="$(V)" TARGET="$*" run-guest-tests, \
"RUN", "TCG tests for $*")
@@ -68,6 +70,7 @@ clean-tcg-tests-%:
$(call quiet-command,$(MAKE) $(SUBDIR_MAKEFLAGS) \
-f $(SRC_PATH)/tests/tcg/Makefile.qemu \
SRC_PATH=$(SRC_PATH) TARGET="$*" clean-guest-tests, \
+ BUILD_DIR=$(BUILD_DIR) \
"CLEAN", "TCG tests for $*")
.PHONY: build-tcg
diff --git a/tests/tcg/Makefile.qemu b/tests/tcg/Makefile.qemu
index 0332bad10f..b531da19dc 100644
--- a/tests/tcg/Makefile.qemu
+++ b/tests/tcg/Makefile.qemu
@@ -26,6 +26,8 @@ include $(SRC_PATH)/tests/docker/Makefile.include
ifdef CROSS_CC_GUEST
+quiet-command = $(if $(V),$1,$(if $(2),@printf " %-7s %s\n" $2 $3 && $1, @$1))
+
.PHONY: cross-build-guest-tests
cross-build-guest-tests:
$(call quiet-command, \
diff --git a/tests/tcg/ppc64 b/tests/tcg/ppc64
new file mode 120000
index 0000000000..e25d62b735
--- /dev/null
+++ b/tests/tcg/ppc64
@@ -0,0 +1 @@
+ppc64le/
\ No newline at end of file
diff --git a/tests/tcg/ppc64le/Makefile.softmmu-target
b/tests/tcg/ppc64le/Makefile.softmmu-target
new file mode 100644
index 0000000000..2c869a81ab
--- /dev/null
+++ b/tests/tcg/ppc64le/Makefile.softmmu-target
@@ -0,0 +1,13 @@
+# -*- Mode: makefile -*-
+#
+# PPC - included from tests/tcg/Makefile.qemu
+#
+
+VPATH += $(SRC_PATH)/tests/tcg/$(TARGET_NAME)/instruction-tests
+
+PPC_TESTS = test-prefixed-load-store
+
+TESTS += $(PPC_TESTS)
+
+test-prefixed-load-store: test-prefixed-load-store.c
+ gcc $< -o test-prefixed-load-store
diff --git a/tests/tcg/ppc64le/instruction-tests/test-prefixed-load-store.c
b/tests/tcg/ppc64le/instruction-tests/test-prefixed-load-store.c
new file mode 100644
index 0000000000..f5948ada85
--- /dev/null
+++ b/tests/tcg/ppc64le/instruction-tests/test-prefixed-load-store.c
@@ -0,0 +1,945 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <assert.h>
+#include <endian.h>
+#include <string.h>
+
+bool debug = false;
+
+#define dprintf(...) \
+ do { \
+ if (debug == true) { \
+ fprintf(stderr, "%s: ", __func__); \
+ fprintf(stderr, __VA_ARGS__); \
+ } \
+ } while (0);
+
+bool le;
+
+#define PSTB(_RS, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1 << 26 | 2 << 24 | (" #_R ") << 20 | (" #_d0 ");" \
+ ".long 38 << 26 | (" #_RS ") << 21 | (" #_RA ") << 16 | (" #_d1 ");"
+#define PSTH(_RS, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1 << 26 | 2 << 24 | (" #_R ") << 20 | (" #_d0 ");" \
+ ".long 44 << 26 | (" #_RS ") << 21 | (" #_RA ") << 16 | (" #_d1 ");"
+#define PSTW(_RS, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1 << 26 | 2 << 24 | (" #_R ") << 20 | (" #_d0 ");" \
+ ".long 36 << 26 | (" #_RS ") << 21 | (" #_RA ") << 16 | (" #_d1 ");"
+#define PSTD(_RS, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1 << 26 | (" #_R ") << 20 | (" #_d0 ");" \
+ ".long 61 << 26 | (" #_RS ") << 21 | (" #_RA ") << 16 | (" #_d1 ");"
+
+#define PST_CALL(op, src, dest_ptr, offset_upper18, offset_lower16, r) \
+ do { \
+ asm( \
+ op(%1, %0, offset_upper18, offset_lower16, r) \
+ : "+r" (dest_ptr) \
+ : "r" (src)); \
+ } while (0);
+
+void check_pst(uint64_t src, uint64_t dest, uint64_t dest_orig, int width) {
+ uint64_t dest_orig_mask;
+ uint64_t src_mask = (width == 8) ? -1UL : (1UL << (8*width)) - 1;
+
+ if (le) {
+ dest_orig_mask = -1UL << (8*width);
+ assert(dest == ((dest_orig & dest_orig_mask) | ((src & src_mask))));
+ } else {
+ dest_orig_mask = (-1UL << (8*width)) >> (8*width);
+ assert(dest == ((dest_orig & dest_orig_mask) | ((src & src_mask) <<
(8*(8-width)))));
+ }
+}
+
+void test_pst_offset(int width) {
+ uint64_t dest_orig = 0x2726252423222120;
+ uint64_t src = 0x1716151413111110;
+ uint64_t dest = dest_orig;
+ void *dest_ptr, *dest_ptr_offset;
+
+ dest_ptr = &dest;
+
+ switch (width) {
+ case 1:
+ dest_ptr_offset = dest_ptr - 1;
+ PST_CALL(PSTB, src, dest_ptr_offset, 0, 1, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0xFFFF;
+ PST_CALL(PSTB, src, dest_ptr_offset, 0, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr + 1;
+ PST_CALL(PSTB, src, dest_ptr_offset, 0x3FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0x1FFFFFFFF;
+ PST_CALL(PSTB, src, dest_ptr_offset, 0x1FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ break;
+ case 2:
+ dest_ptr_offset = dest_ptr - 1;
+ PST_CALL(PSTH, src, dest_ptr_offset, 0, 1, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0xFFFF;
+ PST_CALL(PSTH, src, dest_ptr_offset, 0, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr + 1;
+ PST_CALL(PSTH, src, dest_ptr_offset, 0x3FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0x1FFFFFFFF;
+ PST_CALL(PSTH, src, dest_ptr_offset, 0x1FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ break;
+ case 4:
+ dest_ptr_offset = dest_ptr - 1;
+ PST_CALL(PSTW, src, dest_ptr_offset, 0, 1, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0xFFFF;
+ PST_CALL(PSTW, src, dest_ptr_offset, 0, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr + 1;
+ PST_CALL(PSTW, src, dest_ptr_offset, 0x3FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0x1FFFFFFFF;
+ PST_CALL(PSTW, src, dest_ptr_offset, 0x1FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ break;
+ case 8:
+ dest_ptr_offset = dest_ptr - 1;
+ PST_CALL(PSTD, src, dest_ptr_offset, 0, 1, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0xFFFF;
+ PST_CALL(PSTD, src, dest_ptr_offset, 0, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr + 1;
+ PST_CALL(PSTD, src, dest_ptr_offset, 0x3FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ dest_ptr_offset = dest_ptr - 0x1FFFFFFFF;
+ PST_CALL(PSTD, src, dest_ptr_offset, 0x1FFFF, 0xFFFF, 0);
+ check_pst(src, dest, dest_orig, width);
+ break;
+ default:
+ assert(false);
+ }
+}
+
+void test_pst(int width) {
+ uint64_t dest_orig = 0x2726252423222120;
+ uint64_t src = 0x1716151413111110;
+ uint64_t dest, dest_copy;
+ void *dest_ptr = &dest;
+ void *dest_copy_ptr = &dest_copy;
+
+ /* sanity check against non-prefixed ops */
+ dest_copy = dest_orig;
+ switch (width) {
+ case 1:
+ asm(
+ "stb %1, 0(%0)"
+ : "+r" (dest_copy_ptr)
+ : "r" (src));
+ break;
+ case 2:
+ asm(
+ "sth %1, 0(%0)"
+ : "+r" (dest_copy_ptr)
+ : "r" (src));
+ break;
+ case 4:
+ asm(
+ "stw %1, 0(%0)"
+ : "+r" (dest_copy_ptr)
+ : "r" (src));
+ break;
+ case 8:
+ asm(
+ "std %1, 0(%0)"
+ : "+r" (dest_copy_ptr)
+ : "r" (src));
+ break;
+ default:
+ assert(false);
+ }
+
+ dest = dest_orig;
+ switch (width) {
+ case 1:
+ PST_CALL(PSTB, src, dest_ptr, 0, 0, 0);
+ break;
+ case 2:
+ PST_CALL(PSTH, src, dest_ptr, 0, 0, 0);
+ break;
+ case 4:
+ PST_CALL(PSTW, src, dest_ptr, 0, 0, 0);
+ break;
+ case 8:
+ PST_CALL(PSTD, src, dest_ptr, 0, 0, 0);
+ break;
+ default:
+ assert(false);
+ }
+
+ assert(dest == dest_copy);
+ check_pst(src, dest, dest_orig, width);
+}
+
+void test_pstb(void) {
+ test_pst(1);
+ test_pst_offset(1);
+}
+
+void test_psth(void) {
+ test_pst(2);
+ test_pst_offset(2);
+}
+
+void test_pstw(void) {
+ test_pst(4);
+ test_pst_offset(4);
+}
+
+void test_pstd(void) {
+ test_pst(8);
+ test_pst_offset(8);
+}
+
+#define PLBZ(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | 2<<24 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 34<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+#define PLHZ(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | 2<<24 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 40<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+#define PLHA(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | 2<<24 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 42<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+#define PLWZ(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | 2<<24 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 32<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+#define PLWA(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 41<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+#define PLD(_RT, _RA, _d0, _d1, _R) \
+ ".align 6;" \
+ ".long 1<<26 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 57<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+
+#define PL_CALL(op, src, src_ptr, dest, offset_upper18, offset_lower16, r) \
+ do { \
+ asm( \
+ op(%0, %2, offset_upper18, offset_lower16, r) \
+ : "+r" (dest) \
+ : "r" (src), "r" (src_ptr)); \
+ } while (0);
+
+void check_pl_z(uint64_t src, uint64_t dest, int width) {
+ uint64_t src_mask;
+
+ if (le) {
+ src_mask = (width == 8) ? -1UL : (1UL << (8*width)) - 1;
+ assert(dest == (src & src_mask));
+ } else {
+ src_mask = (width == 8) ? -1UL : -1UL << (8*(8-width));
+ assert(dest == (src & src_mask) >> (8*(8-width)));
+ }
+}
+
+void check_pl_a(uint64_t src, uint64_t dest, int width) {
+ uint64_t src_mask, sign_mask;
+
+ /* TODO: docs suggest testing high-order bit of src byte/halfword/etc, but
+ * QEMU seems to use high-order bit of src double in every case?
+ *
+ * but for le, it's based on the former? afa qemu goes???
+ */
+ if (le) {
+ sign_mask = (src & (1UL << (width*8-1))) ? -1UL << (8*width) : 0;
+ } else {
+ sign_mask = (src & (1UL << 63)) ? -1UL << (8*width) : 0;
+ }
+
+ if (le) {
+ src_mask = (width == 8) ? -1UL : (1UL << (8*width)) - 1;
+ assert(dest == ((src & src_mask) | sign_mask));
+ } else {
+ src_mask = (width == 8) ? -1UL : -1UL << (8*(8-width));
+ assert(dest == (((src & src_mask) >> (8*(8-width))) | sign_mask));
+ }
+}
+
+void test_pl_a(int width, uint64_t src, uint64_t dest_orig) {
+ uint64_t dest = 0, dest_copy;
+ void *src_ptr = &src;
+ void *src_ptr_offset;
+
+ /* sanity check against non-prefixed ops */
+ dest_copy = dest_orig;
+
+ switch (width) {
+ case 2:
+ asm(
+ "lha %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ case 4:
+ asm(
+ "lwa %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ case 8:
+ asm(
+ "ld %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ default:
+ assert(false);
+ }
+
+ switch (width) {
+ case 2:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLHA, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLHA, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLHA, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLHA, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLHA, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ break;
+ case 4:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLWA, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLWA, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLWA, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLWA, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLWA, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ break;
+ case 8:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_a(src, dest, width);
+ break;
+ default:
+ assert(false);
+ }
+
+ assert(dest == dest_copy);
+}
+
+void test_pl_z(int width, uint64_t src, uint64_t dest_orig) {
+ uint64_t dest = 0, dest_copy;
+ void *src_ptr = &src;
+ void *src_ptr_offset;
+
+ /* sanity check against non-prefixed ops */
+ dest_copy = dest_orig;
+
+ switch (width) {
+ case 1:
+ asm(
+ "lbz %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ case 2:
+ asm(
+ "lhz %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ case 4:
+ asm(
+ "lwz %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ case 8:
+ asm(
+ "ld %0, 0(%2)"
+ : "+r" (dest_copy)
+ : "r" (src), "r" (src_ptr));
+ break;
+ default:
+ assert(false);
+ }
+
+ dest = dest_orig;
+ switch (width) {
+ case 1:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLBZ, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLBZ, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLBZ, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLBZ, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLBZ, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ break;
+ case 2:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLHZ, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLHZ, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLHZ, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLHZ, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLHZ, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ break;
+ case 4:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLWZ, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLWZ, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLWZ, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLWZ, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLWZ, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ break;
+ case 8:
+ dest = dest_orig;
+ src_ptr_offset = src_ptr;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 0, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 1;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 1, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0xFFFF;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr + 1;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0x3FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ dest = dest_orig;
+ src_ptr_offset = src_ptr - 0x1FFFFFFFF;
+ PL_CALL(PLD, src, src_ptr_offset, dest, 0x1FFFF, 0xFFFF, 0);
+ check_pl_z(src, dest, width);
+ break;
+ default:
+ assert(false);
+ }
+
+ assert(dest == dest_copy);
+}
+
+void test_plbz(void) {
+ test_pl_z(1, 0x8716151413111110, 0x0726252423222120);
+ test_pl_z(1, 0x1716151413111110, 0x0726252423222120);
+ test_pl_z(1, 0x1716151413111180, 0x0726252423222120);
+}
+
+void test_plhz(void) {
+ test_pl_z(2, 0x8716151483111110, 0x0726252423222120);
+ test_pl_z(1, 0x1716151413111110, 0x0726252423222120);
+ test_pl_z(1, 0x1716151413118110, 0x0726252423222120);
+}
+
+void test_plha(void) {
+ test_pl_a(2, 0x8716151483111110, 0x0726252423222120);
+ test_pl_a(2, 0x1716151413111110, 0x0726252423222120);
+ test_pl_a(2, 0x1716151413118110, 0x0726252423222120);
+}
+
+void test_plwz(void) {
+ test_pl_z(4, 0x8716151483111110, 0x0726252423222120);
+ test_pl_z(4, 0x1716151413111110, 0x0726252423222120);
+ test_pl_z(4, 0x1716151483111110, 0x0726252423222120);
+}
+
+void test_plwa(void) {
+ test_pl_a(4, 0x8716151483111110, 0x0726252423222120);
+ test_pl_a(4, 0x1716151413111110, 0x0726252423222120);
+ test_pl_a(4, 0x1716151483111110, 0x0726252423222120);
+}
+
+void test_pld(void) {
+ test_pl_a(8, 0x8716151483111110, 0x0726252423222120);
+ test_pl_a(8, 0x1716151413111110, 0x0726252423222120);
+}
+
+#define QUADWORD_HI 0x0f0e0d0c0b0a0908
+#define QUADWORD_LO 0x0706050403020100
+
+#define PSTQ(_RS, _RA, _d0, _d1, _R) \
+ ".long 1<<26 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 60<<26 | (" #_RS ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+
+void test_pstq(void) {
+ register uint64_t rs0 asm("r22");
+ register uint64_t rs1 asm("r23");
+ uint64_t storage[2] = { 0 };
+ void *src_ptr = storage;
+
+ if (le) {
+ /*
+ * MEM(EA, 16) <- RSp+1||RSp
+ * where RQ[15..0] = RSp+1||RSp = rs1[7..0] || rs0[7..0]
+ */
+ rs0 = QUADWORD_LO;
+ rs1 = QUADWORD_HI;
+ } else {
+ /*
+ * MEM(EA, 16) <- RSp||RSp+1
+ * where RQ[0..15] = RSp||RSp+1 = rs0[0..7] || rs1[0..7]
+ */
+ rs0 = QUADWORD_HI;
+ rs1 = QUADWORD_LO;
+ }
+
+ asm(
+ PSTQ(22, %0, 0, 0, 0)
+ : "+r" (src_ptr)
+ : "r" (rs0), "r" (rs1));
+
+ if (le) {
+ assert(storage[0] == QUADWORD_LO);
+ assert(storage[1] == QUADWORD_HI);
+ } else {
+ assert(storage[0] == QUADWORD_HI);
+ assert(storage[1] == QUADWORD_LO);
+ }
+
+ /* sanity check against stq */
+ asm(
+ "stq 22, 0(%0)"
+ : "+r" (src_ptr)
+ : "r" (rs0), "r" (rs1));
+
+ if (le) {
+ assert(storage[0] == QUADWORD_HI);
+ assert(storage[1] == QUADWORD_LO);
+ } else {
+ assert(storage[0] == QUADWORD_HI);
+ assert(storage[1] == QUADWORD_LO);
+ }
+}
+
+#define PLQ(_RT, _RA, _d0, _d1, _R) \
+ ".long 1<<26 | (" #_R ")<<20 | (" #_d0 ")\n" \
+ ".long 56<<26 | (" #_RT ")<<21 | (" #_RA ")<<16 | (" #_d1 ")\n"
+
+void test_plq(void) {
+ register uint64_t rdest0 asm("r20") = 7;
+ register uint64_t rdest1 asm("r21") = 8;
+ uint64_t dest0a = 7;
+ uint64_t dest0b = 7;
+ uint64_t dest1a = 7;
+ uint64_t dest1b = 7;
+ uint8_t src[16];
+ void *src_ptr = &src;
+ int i;
+
+ for (i = 0; i < 16; i++) {
+ src[i] = i;
+ }
+
+ /*
+ * PLQ:
+ *
+ * loads to RTp+1||RTp for little-endian
+ * RTp||RTp+1 for big-endian
+ *
+ * so we'd expect:
+ *
+ * value: 0x0f0e..08 || 0706..00
+ *
+ * little-endian:
+ *
+ * uint64_t storage[2] = { 0x0706050403020100,
+ * 0x0f0e0d0c0b0a0908 };
+ * plq 20,0(storage):
+ * r21[0..7] || r20[0..7]
+ * 0x0001020304050607 0x08090a0b0c0d0e0f
+ *
+ * big-endian:
+ *
+ * uint64_t storage[2] = { 0x0f0e0d0c0b0a0908,
+ * 0x0706050403020100 };
+ *
+ * plq 20,0(storage):
+ * r20[0..7] || r21[0..7]
+ * 0x0f0e0d0c0b0a0908 0x0706050403020100
+ *
+ * Note: According to spec, for GPRs at least, GPR byte ordering is always
+ * big-endian with regard to loads/stores. Hence the need to "reverse load"
+ * in the case of loading little-endian value into a register, as opposed
to
+ * simply assuming both the storage and the register would both use
+ * host-endian.
+ *
+ * But, this is just as far as the documentation goes, which is always
+ * left-to-right/big-endian byte ordering. The actual hardware register
+ * stores byte 0 in a little-endian to value to byte 0 in the register, so
+ * registers are loaded host-endian even though the documentation sort of
+ * suggests otherwise in some cases.
+ */
+ asm(
+ PLQ(20, %2, 0, 0, 0)
+ : "=r" (rdest0), "=r" (rdest1)
+ : "r" (src_ptr));
+
+ dest0a = rdest0;
+ dest1a = rdest1;
+
+ /* loads to dest0||dest1 for both endians */
+ asm(
+ "lq 20, 0(%2)"
+ : "=r" (rdest0), "=r" (rdest1)
+ : "r" (src_ptr));
+
+ dest0b = rdest0;
+ dest1b = rdest1;
+
+ if (le) {
+ assert(dest0a == ((uint64_t*)src)[0]);
+ assert(dest1a == ((uint64_t*)src)[1]);
+ assert(dest0a == dest1b);
+ assert(dest1a == dest0b);
+ } else {
+ assert(dest0a == ((uint64_t*)src)[0]);
+ assert(dest1a == ((uint64_t*)src)[1]);
+ assert(dest0a == dest0b);
+ assert(dest1a == dest1b);
+ }
+
+ /* TODO: PC-relative and negative offsets just like all the others */
+}
+
+void test_plq2(void) {
+ register uint64_t rdest0 asm("r20") = 7;
+ register uint64_t rdest1 asm("r21") = 8;
+ register uint64_t rdest0b asm("r22") = 7;
+ register uint64_t rdest1b asm("r23") = 8;
+ uint64_t storage[2];
+ void *src_ptr = storage;
+
+ if (le) {
+ storage[0] = QUADWORD_LO;
+ storage[1] = QUADWORD_HI;
+ } else {
+ storage[0] = QUADWORD_HI;
+ storage[1] = QUADWORD_LO;
+ }
+
+ /*
+ * PLQ:
+ *
+ * loads to RTp+1||RTp for little-endian
+ * RTp||RTp+1 for big-endian
+ *
+ * loads into register using host-endian encoding
+ * calls it "reverse-order" for little-endian, but
+ * the byte-ordering is switched based on endianess
+ * so we still copy mem[0] to reg[0], etc., in all
+ * cases. i.e. storage endian encoding is maintained
+ * in the register encoding after load, even though
+ * documentation might still call it reverse and
+ * reference left-to-right byte ordering in some
+ * cases even for little-endian
+ *
+ * so we'd expect:
+ *
+ * value: 0x0f0e..08 || 0706..00
+ *
+ * little-endian:
+ *
+ * uint64_t storage[2] = { 0x0706050403020100,
+ * 0x0f0e0d0c0b0a0908 };
+ * plq 20,0(storage):
+ * RTquad[15..0] = r21[7..0] || r20[7..0]
+ * r21[7..0] || r20[7..0]
+ * 0x0f0e0d0c0b0a0908 0x0706050403020100
+ *
+ * big-endian:
+ *
+ * uint64_t storage[2] = { 0x0f0e0d0c0b0a0908,
+ * 0x0706050403020100 };
+ *
+ * plq 20,0(storage):
+ * RTquad[0..15] = r20[0..7] || r21[0..7]
+ * r20[0..7] || r21[0..7]
+ * 0x0f0e0d0c0b0a0908 0x0706050403020100
+ **/
+ asm(
+ PLQ(20, %2, 0, 0, 0)
+ : "=r" (rdest0), "=r" (rdest1)
+ : "r" (src_ptr));
+
+ if (le) {
+ assert(rdest0 == QUADWORD_LO);
+ assert(rdest1 == QUADWORD_HI);
+ } else {
+ assert(rdest0 == QUADWORD_HI);
+ assert(rdest1 == QUADWORD_LO);
+ }
+
+ /* sanity check against lq */
+ asm(
+ "lq 22, 0(%2)"
+ : "=r" (rdest0b), "=r" (rdest1b)
+ : "r" (src_ptr));
+
+ if (le) {
+ assert(rdest0 == rdest1b);
+ assert(rdest1 == rdest0b);
+ } else {
+ assert(rdest0 == rdest0b);
+ assert(rdest1 == rdest1b);
+ }
+}
+
+void test_plbz_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLBZ(%0, 0, 0, 8 /* skip plbz */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x1a\n"
+ ".byte 0x1b\n"
+ ".byte 0x1c\n"
+ ".byte 0x1d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ assert(dest == 0x1a);
+}
+
+void test_plhz_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLHZ(%0, 0, 0, 8 /* skip plhz */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x1a\n"
+ ".byte 0x1b\n"
+ ".byte 0x1c\n"
+ ".byte 0x1d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ if (le) {
+ assert(dest == 0x1b1a);
+ } else {
+ assert(dest == 0x1a1b);
+ }
+}
+
+void test_plha_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLHA(%0, 0, 0, 8 /* skip plha */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x8a\n"
+ ".byte 0x8b\n"
+ ".byte 0x1c\n"
+ ".byte 0x1d\n"
+ ".byte 0x2a\n"
+ ".byte 0x2b\n"
+ ".byte 0x2c\n"
+ ".byte 0x2d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ if (le) {
+ assert(dest == 0xFFFFFFFFFFFF8b8a);
+ } else {
+ assert(dest == 0xFFFFFFFFFFFF8a8b);
+ }
+}
+
+void test_plwz_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLWZ(%0, 0, 0, 8 /* skip plwz */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x1a\n"
+ ".byte 0x1b\n"
+ ".byte 0x1c\n"
+ ".byte 0x1d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ if (le) {
+ assert(dest == 0x1d1c1b1a);
+ } else {
+ assert(dest == 0x1a1b1c1d);
+ }
+}
+
+void test_plwa_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLWA(%0, 0, 0, 8 /* skip plwa */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x8a\n"
+ ".byte 0x1b\n"
+ ".byte 0x1c\n"
+ ".byte 0x8d\n"
+ ".byte 0x2a\n"
+ ".byte 0x2b\n"
+ ".byte 0x2c\n"
+ ".byte 0x2d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ if (le) {
+ assert(dest == 0xFFFFFFFF8d1c1b8a);
+ } else {
+ assert(dest == 0xFFFFFFFF8a1b1c8d);
+ }
+}
+
+void test_pld_cia(void) {
+ uint64_t dest = 0;
+
+ asm(
+ PLD(%0, 0, 0, 8 /* skip pld */ + 4 /* skip b */, 1)
+ "b 1f\n"
+ ".byte 0x1a\n"
+ ".byte 0x1b\n"
+ ".byte 0x1c\n"
+ ".byte 0x1d\n"
+ ".byte 0x2a\n"
+ ".byte 0x2b\n"
+ ".byte 0x2c\n"
+ ".byte 0x2d\n"
+ "1: nop\n"
+ : "+r" (dest));
+
+ if (le) {
+ assert(dest == 0x2d2c2b2a1d1c1b1a);
+ } else {
+ assert(dest == 0x1a1b1c1d2a2b2c2d);
+ }
+}
+
+#define do_test(testname) \
+ if (debug) \
+ fprintf(stderr, "-> running test: " #testname "\n"); \
+ test_##testname(); \
+
+int main(int argc, char **argv)
+{
+ le = (htole16(1) == 1);
+
+ if (argc > 1 && !strcmp(argv[1], "-d")) {
+ debug = true;
+ }
+
+ do_test(pstb);
+ do_test(psth);
+ do_test(pstw);
+ do_test(pstd);
+ do_test(plbz);
+ do_test(plhz);
+ do_test(plha);
+ do_test(psth);
+ do_test(pld);
+
+ do_test(pstq);
+ do_test(plq);
+ do_test(plq2);
+
+ do_test(plbz_cia);
+ do_test(plhz_cia);
+ do_test(plha_cia);
+ do_test(plwz_cia);
+ do_test(plwa_cia);
+ do_test(pld_cia);
+
+ dprintf("All tests passed\n");
+ return 0;
+}
--
2.17.1
- [PATCH 1/7] target/ppc: Add infrastructure for prefixed instructions, Gustavo Romero, 2020/10/05
- [PATCH 3/7] tests/tcg: Add tests for prefixed load/store instructions,
Gustavo Romero <=
- [PATCH 4/7] target/ppc: Add support for paired vector load/store instructions, Gustavo Romero, 2020/10/05
- [PATCH 5/7] tests/tcg: Add tests for paired vector load/store instructions, Gustavo Romero, 2020/10/05
- [PATCH 6/7] target/ppc: Add support for prefixed load/store FP instructions, Gustavo Romero, 2020/10/05
- [PATCH 7/7] tests/tcg: Add tests for prefixed load/store FP instructions, Gustavo Romero, 2020/10/05
- Re: [PATCH 1/7] target/ppc: Add infrastructure for prefixed instructions, David Gibson, 2020/10/08