@@ -24,7 +24,7 @@ CFLAGS += -fno-unroll-loops
HEX_SRC=$(SRC_PATH)/tests/tcg/hexagon
VPATH += $(HEX_SRC)
-first: $(HEX_SRC)/first.S
+%: $(HEX_SRC)/%.S $(HEX_SRC)/crt.S
$(CC) -static -mv67 -nostdlib $^ -o $@
HEX_TESTS = first
@@ -42,4 +42,30 @@ HEX_TESTS += atomics
HEX_TESTS += fpstuff
HEX_TESTS += overflow
+HEX_TESTS += test_abs
+HEX_TESTS += test_bitcnt
+HEX_TESTS += test_bitsplit
+HEX_TESTS += test_call
+HEX_TESTS += test_clobber
+HEX_TESTS += test_cmp
+HEX_TESTS += test_dotnew
+HEX_TESTS += test_ext
+HEX_TESTS += test_fibonacci
+HEX_TESTS += test_hl
+HEX_TESTS += test_hwloops
+HEX_TESTS += test_jmp
+HEX_TESTS += test_lsr
+HEX_TESTS += test_mpyi
+HEX_TESTS += test_packet
+HEX_TESTS += test_reorder
+HEX_TESTS += test_round
+HEX_TESTS += test_vavgw
+HEX_TESTS += test_vcmpb
+HEX_TESTS += test_vcmpw
+HEX_TESTS += test_vlsrw
+HEX_TESTS += test_vmaxh
+HEX_TESTS += test_vminh
+HEX_TESTS += test_vpmpyh
+HEX_TESTS += test_vspliceb
+
TESTS += $(HEX_TESTS)
new file mode 100644
@@ -0,0 +1,14 @@
+#define SYS_exit_group 94
+
+ .text
+ .globl pass
+pass:
+ r0 = #0
+ r6 = #SYS_exit_group
+ trap0(#1)
+
+ .globl fail
+fail:
+ r0 = #1
+ r6 = #SYS_exit_group
+ trap0(#1)
new file mode 100644
@@ -0,0 +1,17 @@
+/* Purpose: test example, verify the soundness of the abs operation */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r1 = #-2
+ r2 = #2
+ }
+ {
+ r3 = abs(r1)
+ }
+ {
+ p0 = cmp.eq(r3, r2); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,40 @@
+/*
+ * Purpose: test example, verify the soundness of the cl[01] operations.
+ *
+ * The number 0x000001aa has 23 leading zeroes
+ * they become 55 when considered as 64 bit register
+ * and it has 1 trailing zero.
+ */
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #426
+ r1 = #0
+ }
+ {
+ r2 = cl0(r0)
+ }
+ {
+ p0 = cmp.eq(r2, #23); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ r2 = cl0(r1:0)
+ }
+ {
+ p0 = cmp.eq(r2, #55); if (p0.new) jump:t test3
+ jump fail
+ }
+
+test3:
+ {
+ r2 = ct0(r0)
+ }
+ {
+ p0 = cmp.eq(r2, #1); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,22 @@
+/* Purpose: test example, verify the soundness of the bitsplit operation */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r1 = #187
+ }
+ {
+ r3:2 = bitsplit(r1, #3)
+ }
+ {
+ p0 = cmp.eq(r2, #3); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r3, #23); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,64 @@
+/*
+ * Purpose: test function calls and duplex instructions.
+ * The string "Hello there, I'm a test string!" with the first letter replaced
+ * with a capital L should be printed out.
+ */
+
+#define SYS_write 64
+#define FD_STDOUT 1
+
+ .text
+ .globl test
+test:
+ {
+ jumpr r31
+ memb(r0+#0) = #76
+ }
+.Lfunc_end0:
+.Ltmp0:
+ .size test, .Ltmp0-test
+
+ .globl _start
+_start:
+ {
+ r0 = ##dummy_buffer
+ allocframe(#0)
+ call test
+ }
+ {
+ call write
+ }
+ {
+ deallocframe
+ jump pass
+ }
+.Lfunc_end1:
+.Ltmp1:
+ .size _start, .Ltmp1-_start
+
+write:
+ {
+ r6 = #SYS_write
+ r0 = #FD_STDOUT
+ r1 = ##dummy_buffer
+ r2 = #33
+ }
+ {
+ trap0(#1)
+ }
+ {
+ jumpr r31
+ }
+
+.Lfunc_end2:
+.Ltmp2:
+ .size write, .Ltmp2-write
+
+ .type dummy_buffer,@object
+ .data
+ .globl dummy_buffer
+ .p2align 3
+dummy_buffer:
+ .string "Hello there, I'm a test string!\n"
+ .space 223
+ .size dummy_buffer, 256
new file mode 100644
@@ -0,0 +1,29 @@
+/*
+ * Purpose: demonstrate the succesful operation of the register save mechanism,
+ * in which the caller saves the registers that will be clobbered, and restores
+ * them after the call.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ allocframe(#8)
+ {
+ r16 = #47
+ r17 = #155
+ }
+ memd(sp+#0) = r17:16
+ {
+ r16 = #255
+ r17 = #42
+ }
+ {
+ deallocframe
+ r17:16 = memd(sp+#0)
+ }
+ {
+ p0 = cmp.eq(r16, #47)
+ p0 = cmp.eq(r17, #155); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,31 @@
+/* Purpose: test a signed and unsigned comparison */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ jump signed
+ }
+
+ .globl signed
+signed:
+ {
+ r0 = #-2
+ r1 = #0
+ }
+ {
+ p0 = cmp.lt(r0, r1); if (p0.new) jump:t unsigned
+ jump fail
+ }
+
+ .globl unsigned
+unsigned:
+ {
+ r0 = #-2
+ r1 = #0
+ }
+ {
+ p0 = cmp.gtu(r0, r1); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,38 @@
+/* Purpose: test the .new operator while performing memory stores. */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ allocframe(#16)
+ }
+ {
+ r0 = #1
+ memw(sp+#0) = r0.new
+ }
+ {
+ r1 = #2
+ memw(sp+#4) = r1.new
+ }
+ {
+ r2 = #3
+ memw(sp+#8) = r2.new
+ }
+ {
+ r0 = memw(sp+#8)
+ }
+ {
+ r1 = memw(sp+#4)
+ }
+ {
+ r2 = memw(sp+#0)
+ }
+ {
+ r3 = mpyi(r1, r2)
+ }
+ {
+ deallocframe
+ p0 = cmp.eq(r3, #2); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,13 @@
+/* Purpose: test immediate extender instructions. */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r2 = ##-559038737
+ }
+ {
+ p0 = cmp.eq(r2, ##-559038737); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,30 @@
+/* Purpose: computes the Fibonacci series up to a constant number. */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r2 = #100
+ }
+ {
+ p0 = cmp.gt(r2, #0); if (!p0.new) jump:nt .LBB0_3
+ }
+ {
+ r3 = #0
+ r4 = #1
+ }
+.LBB0_2:
+ {
+ r5 = r4
+ }
+ {
+ p0 = cmp.gt(r2, r5); if (p0.new) jump:nt .LBB0_2
+ r4 = add(r3, r4)
+ r3 = r5
+ }
+.LBB0_3:
+ {
+ p0 = cmp.eq(r3, #144); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,16 @@
+/* Purpose: test example, verify the soundness of the high/low assignment */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0.H = #42
+ }
+ {
+ r0.L = #69
+ }
+ {
+ p0 = cmp.eq(r0, #2752581); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,19 @@
+/* Purpose: simple C Program to test hardware loops. */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ loop0(.LBB0_1, #10)
+ r2 = #0
+ }
+.LBB0_1:
+ {
+ r2 = add(r2, #1)
+ nop
+ }:endloop0
+ {
+ p0 = cmp.eq(r2, #10); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,22 @@
+/* Purpose: test example, verify the soundness of the jump operation */
+
+#define SYS_exit_group 94
+
+ .text
+ .globl _start
+
+_start:
+ {
+ jump pass
+ }
+ /*
+ * Inlined fail label in crt.S so we can fail without
+ * having a functioning jump
+ */
+ {
+ r0 = #1
+ r6 = #SYS_exit_group
+ }
+ {
+ trap0(#1)
+ }
new file mode 100644
@@ -0,0 +1,36 @@
+/* Purpose: test the soundness of the lsr operation */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #-56984
+ r1 = #2147483647
+ }
+ {
+ r2 = #0x19
+ }
+ {
+ r0 &= lsr(r1, r2)
+ }
+ {
+ p0 = cmp.eq(r0, #0x28); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ r0 = #0x0000000a
+ r1 = #0x00000000
+ }
+ {
+ r2 = #-1
+ }
+ {
+ r1:0 = lsl(r1:0, r2)
+ }
+ {
+ p0 = cmp.eq(r0, #0x5); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,17 @@
+/* Purpose: test a simple multiplication operation */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r1 = #4
+ r2 = #6
+ }
+ {
+ r3 = mpyi(r1, r2)
+ }
+ {
+ p0 = cmp.eq(r3, #24); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,29 @@
+/*
+ * Purpose: test that writes of a register in a packet are performed only after
+ * that packet has finished its execution.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ allocframe(#8)
+ }
+ {
+ r2 = #4
+ r3 = #6
+ }
+ {
+ memw(sp+#0) = r2
+ }
+ {
+ r3 = memw(sp+#0)
+ r0 = add(r2, r3)
+ }
+ {
+ deallocframe
+ p0 = cmp.eq(r3, #4)
+ p0 = cmp.eq(r0, #10); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,33 @@
+/*
+ * Purpose: demonstrate handling of .new uses appearing before the associated
+ * definition.
+ * Here we perform a jump that skips the code resetting R2 from 0xDEADBEEF to 0,
+ * only if P0.new is true, but P0 is assigned to 1 (R4) in the next instruction
+ * in the packet.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r2 = #-559038737
+ }
+ {
+ r4 = #1
+ }
+ {
+ if (p0.new) jump:nt skip
+ p0 = r4;
+ }
+
+fallthrough:
+ {
+ r2 = #0
+ }
+
+skip:
+ {
+ p0 = cmp.eq(r2, #-559038737); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,29 @@
+/*
+ * Purpose: test example, verify the soundness of the cround operation
+ * 106 = 0b1101010 with the comma at third digit is 12.5 which is crounded to 12
+ * but rounded to 13.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r1 = #200
+ }
+ {
+ r2 = round(r1, #4)
+ }
+ {
+ p0 = cmp.eq(r2, #13); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ r2 = cround(r1, #4)
+ }
+ {
+ p0 = cmp.eq(r2, #12); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,31 @@
+/*
+ * Purpose: test example, verify the soundness of the vavgw operation.
+ *
+ * 0x00030001 averaged with 0x00010003 results 0x00020002.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #3
+ r1 = #1
+ }
+ {
+ r2 = #1
+ r3 = #3
+ }
+ {
+ r1:0 = vavgw(r1:0, r3:2):crnd
+ }
+ {
+ p0 = cmp.eq(r0, #2); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r1, #2); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,30 @@
+/*
+ * Purpose: test example, verify the soundness of the vector compare bytes
+ * operation.
+ *
+ * Vector byte comparison between 0x1234567887654321 and 0x1234567800000000
+ * should result in 0b11110000 in binary, or 0xf0 in hex.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #0x87654321
+ r1 = #0x12345678
+ }
+ {
+ r2 = #0x00000000
+ r3 = #0x12345678
+ }
+ {
+ p2 = vcmpb.eq(r1:0, r3:2)
+ }
+ {
+ r4 = p2
+ }
+ {
+ p0 = cmp.eq(r4, #0xf0); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,30 @@
+/*
+ * Purpose: test example, verify the soundness of the vector compare words
+ * operation.
+ *
+ * Vector word comparison between 0x1234567887654321 and 0x1234567800000000
+ * should result in 0b11110000 in binary, or 0xf0 in hex.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #0x87654321
+ r1 = #0x12345678
+ }
+ {
+ r2 = #0x00000000
+ r3 = #0x12345678
+ }
+ {
+ p2 = vcmpw.eq(r1:0, r3:2)
+ }
+ {
+ r4 = p2
+ }
+ {
+ p0 = cmp.eq(r4, #0xf0); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,20 @@
+/* Purpose: test the soundness of the vlsrw operation */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #0x00000001
+ r1 = #0x00000001
+ }
+ {
+ r1:0 = vlsrw(r1:0, #1)
+ }
+ {
+ r0 = add(r0, r1)
+ }
+ {
+ p0 = cmp.eq(r0, #0); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,35 @@
+/*
+ * Purpose: test example, verify the soundness of the vrmaxh operation.
+ *
+ * The maximum between 0x0002000300010005 and 0x0003000200020007 is
+ * 0x0003000300020007.
+ *
+ * input: r1 = 0x00010003 r0 = 0x00010005 r3 = 0x00030002 r2 = 0x00020007
+ * output: r1 = 0x00030003 r0 = 0x00020007
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #65541
+ r1 = #65539
+ }
+ {
+ r2 = #131079
+ r3 = #196610
+ }
+ {
+ r1:0 = vmaxh(r1:0, r3:2)
+ }
+ {
+ p0 = cmp.eq(r0, #131079); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r1, #196611); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,35 @@
+/*
+ * Purpose: test example, verify the soundness of the vrmaxh operation.
+ *
+ * The minimum between 0x0002000300010005 and 0x0003000200020007 is
+ * 0x0003000300020007.
+ *
+ * input: r1 = 0x00010003 r0 = 0x00010005 r3 = 0x00030002 r2 = 0x00020007
+ * output: r1 = 0x00010002 r0 = 0x00010005
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #65541
+ r1 = #65539
+ }
+ {
+ r2 = #131079
+ r3 = #196610
+ }
+ {
+ r1:0 = vminh(r1:0, r3:2)
+ }
+ {
+ p0 = cmp.eq(r0, #65541); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r1, #65538); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,28 @@
+/*
+ * Purpose: test example, verify the soundness of the vpmpyh operator.
+ *
+ * 0x01020304 vector polynomial multiplied with 0x04030201 results
+ * 0x000400060b060b04.
+ */
+
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #16909060
+ r1 = #67305985
+ }
+ {
+ r1:0 = vpmpyh(r0, r1)
+ }
+ {
+ p0 = cmp.eq(r0, #184945412); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r1, #262150); if (p0.new) jump:t pass
+ jump fail
+ }
new file mode 100644
@@ -0,0 +1,31 @@
+/*
+ * Purpose: test example, verify the soundness of the vspliceb operation
+ * the operation is a binary splice of two 64bit operators.
+ *
+ * vspliceb(0xffffffffffffffff,0x0000000000000000,5) = 0x000000ffffffffff.
+ */
+ .text
+ .globl _start
+
+_start:
+ {
+ r0 = #-1
+ r1 = #-1
+ }
+ {
+ r2 = #0
+ r3 = #0
+ }
+ {
+ r5:4 = vspliceb(r1:0, r3:2, #5)
+ }
+ {
+ p0 = cmp.eq(r4, #-1); if (p0.new) jump:t test2
+ jump fail
+ }
+
+test2:
+ {
+ p0 = cmp.eq(r5, #255); if (p0.new) jump:t pass
+ jump fail
+ }