aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorseyko <seyko2@gmail.com>2015-09-23 14:58:06 +0300
committerseyko <seyko2@gmail.com>2015-09-23 14:58:06 +0300
commiteb870b006c969d16002fc724d5bbfe33287801e5 (patch)
tree2f56b231d86158596287a70560ee5ee88ea229f7
parent8f620c8af8f664d8e17144b3f29f0c22d5ea190f (diff)
downloadtinycc-eb870b006c969d16002fc724d5bbfe33287801e5.tar.gz
tinycc-eb870b006c969d16002fc724d5bbfe33287801e5.tar.bz2
SSE opcodes to TCC assembler (i386, x86_64)
patch from Anaƫl Seghezzi a test program: ============================ #include <stdio.h> struct fl4{ float x, y, z, w; }; void asm_test(void) { struct fl4 v1, v2, v3; v1.x = 0.1; v1.y = 0.2; v1.z = 0.4; v1.w = 0.3; v2.x = 0.11; v2.y = 0.0; v2.z = 0.01; v2.w = 0.04; asm volatile ( "movups %0, %%xmm0;" "movups %1, %%xmm1;" "addps %%xmm1, %%xmm0;" "movups %%xmm0, %2" :: "g" (v1), "g" (v2), "g" (v3) : "memory"); printf("sse fl4 add : %f %f %f %f\n", v3.x, v3.y, v3.z, v3.w); printf("expected : %f %f %f %f\n", v1.x+v2.x, v1.y+v2.y, v1.z+v2.z, v1.w+v2.w); } int main() { asm_test(); } /* sse fl4 add : 0.210000 0.200000 0.410000 0.340000 expected : 0.210000 0.200000 0.410000 0.340000 */ ============================
-rw-r--r--i386-asm.h28
-rw-r--r--x86_64-asm.h28
2 files changed, 54 insertions, 2 deletions
diff --git a/i386-asm.h b/i386-asm.h
index 43c8867..486cffe 100644
--- a/i386-asm.h
+++ b/i386-asm.h
@@ -464,7 +464,33 @@ ALT(DEF_ASM_OP2(psrlq, 0x0f73, 2, OPC_MODRM, OPT_IM8, OPT_MMX ))
DEF_ASM_OP2(punpcklbw, 0x0f60, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
DEF_ASM_OP2(punpcklwd, 0x0f61, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
DEF_ASM_OP2(punpckldq, 0x0f62, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
- DEF_ASM_OP2(pxor, 0x0fef, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX ) /* must be last !OP0 */
+ DEF_ASM_OP2(pxor, 0x0fef, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+
+ /* sse */
+ DEF_ASM_OP2(movups, 0x0f10, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movups, 0x0f11, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(movaps, 0x0f28, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movaps, 0x0f29, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(movhps, 0x0f16, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movhps, 0x0f17, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(addps, 0x0f58, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(cvtpi2ps, 0x0f2a, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_SSE )
+ DEF_ASM_OP2(cvtps2pi, 0x0f2d, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_MMX )
+ DEF_ASM_OP2(cvttps2pi, 0x0f2c, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_MMX )
+ DEF_ASM_OP2(divps, 0x0f5e, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(maxps, 0x0f5f, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(minps, 0x0f5d, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(mulps, 0x0f59, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pavgb, 0x0fe0, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pavgw, 0x0fe3, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pmaxsw, 0x0fee, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pmaxub, 0x0fde, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pminsw, 0x0fea, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pminub, 0x0fda, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(rcpss, 0x0f53, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(rsqrtps, 0x0f52, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(sqrtps, 0x0f51, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(subps, 0x0f5c, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
#undef ALT
#undef DEF_ASM_OP0
diff --git a/x86_64-asm.h b/x86_64-asm.h
index 2f80e4b..df76bf0 100644
--- a/x86_64-asm.h
+++ b/x86_64-asm.h
@@ -439,7 +439,33 @@ ALT(DEF_ASM_OP2(psrlq, 0x0f73, 2, OPC_MODRM, OPT_IM8, OPT_MMX ))
DEF_ASM_OP2(punpcklbw, 0x0f60, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
DEF_ASM_OP2(punpcklwd, 0x0f61, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
DEF_ASM_OP2(punpckldq, 0x0f62, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
- DEF_ASM_OP2(pxor, 0x0fef, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX ) /* must be last !OP0 */
+ DEF_ASM_OP2(pxor, 0x0fef, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+
+ /* sse */
+ DEF_ASM_OP2(movups, 0x0f10, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movups, 0x0f11, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(movaps, 0x0f28, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movaps, 0x0f29, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(movhps, 0x0f16, 0, OPC_MODRM, OPT_EA | OPT_REG32, OPT_SSE )
+ALT(DEF_ASM_OP2(movhps, 0x0f17, 0, OPC_MODRM, OPT_SSE, OPT_EA | OPT_REG32 ))
+ DEF_ASM_OP2(addps, 0x0f58, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(cvtpi2ps, 0x0f2a, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_SSE )
+ DEF_ASM_OP2(cvtps2pi, 0x0f2d, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_MMX )
+ DEF_ASM_OP2(cvttps2pi, 0x0f2c, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_MMX )
+ DEF_ASM_OP2(divps, 0x0f5e, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(maxps, 0x0f5f, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(minps, 0x0f5d, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(mulps, 0x0f59, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pavgb, 0x0fe0, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pavgw, 0x0fe3, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(pmaxsw, 0x0fee, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pmaxub, 0x0fde, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pminsw, 0x0fea, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(pminub, 0x0fda, 0, OPC_MODRM, OPT_EA | OPT_MMX, OPT_MMX )
+ DEF_ASM_OP2(rcpss, 0x0f53, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(rsqrtps, 0x0f52, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(sqrtps, 0x0f51, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
+ DEF_ASM_OP2(subps, 0x0f5c, 0, OPC_MODRM, OPT_EA | OPT_SSE, OPT_SSE )
#undef ALT
#undef DEF_ASM_OP0