[Mesa-dev] [PATCH] llvmpipe: Don't assume vector is 4 wide in lp_build_sin()/lp_build_cos()

Dave Airlie airlied at gmail.com
Fri Feb 17 10:28:34 PST 2012


On Fri, Feb 17, 2012 at 4:23 PM,  <jfonseca at vmware.com> wrote:
> From: José Fonseca <jfonseca at vmware.com>

Seems correct to me.

Reviewed-by: Dave Airlie <airlied at redhat.com>

>
> ---
>  src/gallium/auxiliary/gallivm/lp_bld_arit.c |  141 +++++++++++---------------
>  1 files changed, 60 insertions(+), 81 deletions(-)
>
> diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
> index a15dcb0..d379593 100644
> --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
> +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
> @@ -1662,23 +1662,6 @@ lp_build_rsqrt(struct lp_build_context *bld,
>  }
>
>
> -static inline LLVMValueRef
> -lp_build_const_v4si(struct gallivm_state *gallivm, unsigned long value)
> -{
> -   LLVMValueRef element = lp_build_const_int32(gallivm, value);
> -   LLVMValueRef elements[4] = { element, element, element, element };
> -   return LLVMConstVector(elements, 4);
> -}
> -
> -static inline LLVMValueRef
> -lp_build_const_v4sf(struct gallivm_state *gallivm, float value)
> -{
> -   LLVMValueRef element = lp_build_const_float(gallivm, value);
> -   LLVMValueRef elements[4] = { element, element, element, element };
> -   return LLVMConstVector(elements, 4);
> -}
> -
> -
>  /**
>  * Generate sin(a) using SSE2
>  */
> @@ -1686,29 +1669,27 @@ LLVMValueRef
>  lp_build_sin(struct lp_build_context *bld,
>              LLVMValueRef a)
>  {
> -   LLVMBuilderRef builder = bld->gallivm->builder;
>    struct gallivm_state *gallivm = bld->gallivm;
> +   LLVMBuilderRef builder = gallivm->builder;
>    struct lp_type int_type = lp_int_type(bld->type);
>    LLVMBuilderRef b = builder;
> -   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4);
> -   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4);
>
>    /*
>     *  take the absolute value,
>     *  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
>     */
>
> -   LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000);
> -   LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si");
> +   LLVMValueRef inv_sig_mask = lp_build_const_int_vec(gallivm, bld->type, ~0x80000000);
> +   LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, bld->int_vec_type, "a_v4si");
>
>    LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi");
> -   LLVMValueRef x_abs = LLVMBuildBitCast(b, absi, v4sf, "x_abs");
> +   LLVMValueRef x_abs = LLVMBuildBitCast(b, absi, bld->vec_type, "x_abs");
>
>    /*
>     * extract the sign bit (upper one)
>     * sign_bit = _mm_and_ps(sign_bit, *(v4sf*)_ps_sign_mask);
>     */
> -   LLVMValueRef sig_mask = lp_build_const_v4si(bld->gallivm, 0x80000000);
> +   LLVMValueRef sig_mask = lp_build_const_int_vec(gallivm, bld->type, 0x80000000);
>    LLVMValueRef sign_bit_i = LLVMBuildAnd(b, a_v4si, sig_mask, "sign_bit_i");
>
>    /*
> @@ -1716,7 +1697,7 @@ lp_build_sin(struct lp_build_context *bld,
>     * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
>     */
>
> -   LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516);
> +   LLVMValueRef FOPi = lp_build_const_vec(gallivm, bld->type, 1.27323954473516);
>    LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y");
>
>    /*
> @@ -1724,36 +1705,36 @@ lp_build_sin(struct lp_build_context *bld,
>     * emm2 = _mm_cvttps_epi32(y);
>     */
>
> -   LLVMValueRef emm2_i = LLVMBuildFPToSI(b, scale_y, v4si, "emm2_i");
> +   LLVMValueRef emm2_i = LLVMBuildFPToSI(b, scale_y, bld->int_vec_type, "emm2_i");
>
>    /*
>     * j=(j+1) & (~1) (see the cephes sources)
>     * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
>     */
>
> -   LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1);
> +   LLVMValueRef all_one = lp_build_const_int_vec(gallivm, bld->type, 1);
>    LLVMValueRef emm2_add =  LLVMBuildAdd(b, emm2_i, all_one, "emm2_add");
>    /*
>     * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
>     */
> -   LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1);
> +   LLVMValueRef inv_one = lp_build_const_int_vec(gallivm, bld->type, ~1);
>    LLVMValueRef emm2_and =  LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and");
>
>    /*
>     * y = _mm_cvtepi32_ps(emm2);
>     */
> -   LLVMValueRef y_2 = LLVMBuildSIToFP(b, emm2_and, v4sf, "y_2");
> +   LLVMValueRef y_2 = LLVMBuildSIToFP(b, emm2_and, bld->vec_type, "y_2");
>
>    /* get the swap sign flag
>     * emm0 = _mm_and_si128(emm2, *(v4si*)_pi32_4);
>     */
> -   LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4);
> +   LLVMValueRef pi32_4 = lp_build_const_int_vec(gallivm, bld->type, 4);
>    LLVMValueRef emm0_and =  LLVMBuildAnd(b, emm2_add, pi32_4, "emm0_and");
>
>    /*
>     * emm2 = _mm_slli_epi32(emm0, 29);
>     */
> -   LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29);
> +   LLVMValueRef const_29 = lp_build_const_int_vec(gallivm, bld->type, 29);
>    LLVMValueRef swap_sign_bit = LLVMBuildShl(b, emm0_and, const_29, "swap_sign_bit");
>
>    /*
> @@ -1766,11 +1747,11 @@ lp_build_sin(struct lp_build_context *bld,
>     * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
>     */
>
> -   LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2);
> +   LLVMValueRef pi32_2 = lp_build_const_int_vec(gallivm, bld->type, 2);
>    LLVMValueRef emm2_3 =  LLVMBuildAnd(b, emm2_and, pi32_2, "emm2_3");
> -   LLVMValueRef poly_mask = lp_build_compare(bld->gallivm,
> +   LLVMValueRef poly_mask = lp_build_compare(gallivm,
>                                              int_type, PIPE_FUNC_EQUAL,
> -                                             emm2_3, lp_build_const_v4si(bld->gallivm, 0));
> +                                             emm2_3, lp_build_const_int_vec(gallivm, bld->type, 0));
>    /*
>     *   sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit);
>     */
> @@ -1781,9 +1762,9 @@ lp_build_sin(struct lp_build_context *bld,
>     * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
>     * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
>     */
> -   LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625);
> -   LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4);
> -   LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8);
> +   LLVMValueRef DP1 = lp_build_const_vec(gallivm, bld->type, -0.78515625);
> +   LLVMValueRef DP2 = lp_build_const_vec(gallivm, bld->type, -2.4187564849853515625e-4);
> +   LLVMValueRef DP3 = lp_build_const_vec(gallivm, bld->type, -3.77489497744594108e-8);
>
>    /*
>     * The magic pass: "Extended precision modular arithmetic"
> @@ -1818,9 +1799,9 @@ lp_build_sin(struct lp_build_context *bld,
>     * _PS_CONST(coscof_p1, -1.388731625493765E-003);
>     * _PS_CONST(coscof_p2,  4.166664568298827E-002);
>     */
> -   LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005);
> -   LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003);
> -   LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002);
> +   LLVMValueRef coscof_p0 = lp_build_const_vec(gallivm, bld->type, 2.443315711809948E-005);
> +   LLVMValueRef coscof_p1 = lp_build_const_vec(gallivm, bld->type, -1.388731625493765E-003);
> +   LLVMValueRef coscof_p2 = lp_build_const_vec(gallivm, bld->type, 4.166664568298827E-002);
>
>    /*
>     * y = *(v4sf*)_ps_coscof_p0;
> @@ -1839,10 +1820,10 @@ lp_build_sin(struct lp_build_context *bld,
>     * y = _mm_sub_ps(y, tmp);
>     * y = _mm_add_ps(y, *(v4sf*)_ps_1);
>     */
> -   LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5);
> +   LLVMValueRef half = lp_build_const_vec(gallivm, bld->type, 0.5);
>    LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp");
>    LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8");
> -   LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0);
> +   LLVMValueRef one = lp_build_const_vec(gallivm, bld->type, 1.0);
>    LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9");
>
>    /*
> @@ -1850,9 +1831,9 @@ lp_build_sin(struct lp_build_context *bld,
>     * _PS_CONST(sincof_p1,  8.3321608736E-3);
>     * _PS_CONST(sincof_p2, -1.6666654611E-1);
>     */
> -   LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4);
> -   LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3);
> -   LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1);
> +   LLVMValueRef sincof_p0 = lp_build_const_vec(gallivm, bld->type, -1.9515295891E-4);
> +   LLVMValueRef sincof_p1 = lp_build_const_vec(gallivm, bld->type, 8.3321608736E-3);
> +   LLVMValueRef sincof_p2 = lp_build_const_vec(gallivm, bld->type, -1.6666654611E-1);
>
>    /*
>     * Evaluate the second polynom  (Pi/4 <= x <= 0)
> @@ -1882,10 +1863,10 @@ lp_build_sin(struct lp_build_context *bld,
>     * y = _mm_andnot_ps(xmm3, y);
>     * y = _mm_add_ps(y,y2);
>     */
> -   LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, v4si, "y2_i");
> -   LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, v4si, "y_i");
> +   LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, bld->int_vec_type, "y2_i");
> +   LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, bld->int_vec_type, "y_i");
>    LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and");
> -   LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0);
> +   LLVMValueRef inv = lp_build_const_int_vec(gallivm, bld->type, ~0);
>    LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv");
>    LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and");
>    LLVMValueRef y_combine = LLVMBuildAdd(b, y_and, y2_and, "y_combine");
> @@ -1895,7 +1876,7 @@ lp_build_sin(struct lp_build_context *bld,
>     * y = _mm_xor_ps(y, sign_bit);
>     */
>    LLVMValueRef y_sign = LLVMBuildXor(b, y_combine, sign_bit_1, "y_sin");
> -   LLVMValueRef y_result = LLVMBuildBitCast(b, y_sign, v4sf, "y_result");
> +   LLVMValueRef y_result = LLVMBuildBitCast(b, y_sign, bld->vec_type, "y_result");
>    return y_result;
>  }
>
> @@ -1907,30 +1888,28 @@ LLVMValueRef
>  lp_build_cos(struct lp_build_context *bld,
>              LLVMValueRef a)
>  {
> -   LLVMBuilderRef builder = bld->gallivm->builder;
>    struct gallivm_state *gallivm = bld->gallivm;
> +   LLVMBuilderRef builder = gallivm->builder;
>    struct lp_type int_type = lp_int_type(bld->type);
>    LLVMBuilderRef b = builder;
> -   LLVMTypeRef v4sf = LLVMVectorType(LLVMFloatTypeInContext(bld->gallivm->context), 4);
> -   LLVMTypeRef v4si = LLVMVectorType(LLVMInt32TypeInContext(bld->gallivm->context), 4);
>
>    /*
>     *  take the absolute value,
>     *  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
>     */
>
> -   LLVMValueRef inv_sig_mask = lp_build_const_v4si(bld->gallivm, ~0x80000000);
> -   LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, v4si, "a_v4si");
> +   LLVMValueRef inv_sig_mask = lp_build_const_int_vec(gallivm, bld->type, ~0x80000000);
> +   LLVMValueRef a_v4si = LLVMBuildBitCast(b, a, bld->int_vec_type, "a_v4si");
>
>    LLVMValueRef absi = LLVMBuildAnd(b, a_v4si, inv_sig_mask, "absi");
> -   LLVMValueRef x_abs = LLVMBuildBitCast(b, absi, v4sf, "x_abs");
> +   LLVMValueRef x_abs = LLVMBuildBitCast(b, absi, bld->vec_type, "x_abs");
>
>    /*
>     * scale by 4/Pi
>     * y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
>     */
>
> -   LLVMValueRef FOPi = lp_build_const_v4sf(gallivm, 1.27323954473516);
> +   LLVMValueRef FOPi = lp_build_const_vec(gallivm, bld->type, 1.27323954473516);
>    LLVMValueRef scale_y = LLVMBuildFMul(b, x_abs, FOPi, "scale_y");
>
>    /*
> @@ -1938,46 +1917,46 @@ lp_build_cos(struct lp_build_context *bld,
>     * emm2 = _mm_cvttps_epi32(y);
>     */
>
> -   LLVMValueRef emm2_i = LLVMBuildFPToSI(b, scale_y, v4si, "emm2_i");
> +   LLVMValueRef emm2_i = LLVMBuildFPToSI(b, scale_y, bld->int_vec_type, "emm2_i");
>
>    /*
>     * j=(j+1) & (~1) (see the cephes sources)
>     * emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
>     */
>
> -   LLVMValueRef all_one = lp_build_const_v4si(bld->gallivm, 1);
> +   LLVMValueRef all_one = lp_build_const_int_vec(gallivm, bld->type, 1);
>    LLVMValueRef emm2_add =  LLVMBuildAdd(b, emm2_i, all_one, "emm2_add");
>    /*
>     * emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
>     */
> -   LLVMValueRef inv_one = lp_build_const_v4si(bld->gallivm, ~1);
> +   LLVMValueRef inv_one = lp_build_const_int_vec(gallivm, bld->type, ~1);
>    LLVMValueRef emm2_and =  LLVMBuildAnd(b, emm2_add, inv_one, "emm2_and");
>
>    /*
>     * y = _mm_cvtepi32_ps(emm2);
>     */
> -   LLVMValueRef y_2 = LLVMBuildSIToFP(b, emm2_and, v4sf, "y_2");
> +   LLVMValueRef y_2 = LLVMBuildSIToFP(b, emm2_and, bld->vec_type, "y_2");
>
>
>    /*
>     * emm2 = _mm_sub_epi32(emm2, *(v4si*)_pi32_2);
>     */
> -   LLVMValueRef const_2 = lp_build_const_v4si(bld->gallivm, 2);
> +   LLVMValueRef const_2 = lp_build_const_int_vec(gallivm, bld->type, 2);
>    LLVMValueRef emm2_2 = LLVMBuildSub(b, emm2_and, const_2, "emm2_2");
>
>
>    /* get the swap sign flag
>     * emm0 = _mm_andnot_si128(emm2, *(v4si*)_pi32_4);
>     */
> -   LLVMValueRef inv = lp_build_const_v4si(bld->gallivm, ~0);
> +   LLVMValueRef inv = lp_build_const_int_vec(gallivm, bld->type, ~0);
>    LLVMValueRef emm0_not = LLVMBuildXor(b, emm2_2, inv, "emm0_not");
> -   LLVMValueRef pi32_4 = lp_build_const_v4si(bld->gallivm, 4);
> +   LLVMValueRef pi32_4 = lp_build_const_int_vec(gallivm, bld->type, 4);
>    LLVMValueRef emm0_and =  LLVMBuildAnd(b, emm0_not, pi32_4, "emm0_and");
>
>    /*
>     * emm2 = _mm_slli_epi32(emm0, 29);
>     */
> -   LLVMValueRef const_29 = lp_build_const_v4si(bld->gallivm, 29);
> +   LLVMValueRef const_29 = lp_build_const_int_vec(gallivm, bld->type, 29);
>    LLVMValueRef sign_bit = LLVMBuildShl(b, emm0_and, const_29, "sign_bit");
>
>    /*
> @@ -1990,20 +1969,20 @@ lp_build_cos(struct lp_build_context *bld,
>     * emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
>     */
>
> -   LLVMValueRef pi32_2 = lp_build_const_v4si(bld->gallivm, 2);
> +   LLVMValueRef pi32_2 = lp_build_const_int_vec(gallivm, bld->type, 2);
>    LLVMValueRef emm2_3 =  LLVMBuildAnd(b, emm2_2, pi32_2, "emm2_3");
> -   LLVMValueRef poly_mask = lp_build_compare(bld->gallivm,
> +   LLVMValueRef poly_mask = lp_build_compare(gallivm,
>                                              int_type, PIPE_FUNC_EQUAL,
> -                                            emm2_3, lp_build_const_v4si(bld->gallivm, 0));
> +                                            emm2_3, lp_build_const_int_vec(gallivm, bld->type, 0));
>
>    /*
>     * _PS_CONST(minus_cephes_DP1, -0.78515625);
>     * _PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
>     * _PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
>     */
> -   LLVMValueRef DP1 = lp_build_const_v4sf(gallivm, -0.78515625);
> -   LLVMValueRef DP2 = lp_build_const_v4sf(gallivm, -2.4187564849853515625e-4);
> -   LLVMValueRef DP3 = lp_build_const_v4sf(gallivm, -3.77489497744594108e-8);
> +   LLVMValueRef DP1 = lp_build_const_vec(gallivm, bld->type, -0.78515625);
> +   LLVMValueRef DP2 = lp_build_const_vec(gallivm, bld->type, -2.4187564849853515625e-4);
> +   LLVMValueRef DP3 = lp_build_const_vec(gallivm, bld->type, -3.77489497744594108e-8);
>
>    /*
>     * The magic pass: "Extended precision modular arithmetic"
> @@ -2038,9 +2017,9 @@ lp_build_cos(struct lp_build_context *bld,
>     * _PS_CONST(coscof_p1, -1.388731625493765E-003);
>     * _PS_CONST(coscof_p2,  4.166664568298827E-002);
>     */
> -   LLVMValueRef coscof_p0 = lp_build_const_v4sf(gallivm, 2.443315711809948E-005);
> -   LLVMValueRef coscof_p1 = lp_build_const_v4sf(gallivm, -1.388731625493765E-003);
> -   LLVMValueRef coscof_p2 = lp_build_const_v4sf(gallivm, 4.166664568298827E-002);
> +   LLVMValueRef coscof_p0 = lp_build_const_vec(gallivm, bld->type, 2.443315711809948E-005);
> +   LLVMValueRef coscof_p1 = lp_build_const_vec(gallivm, bld->type, -1.388731625493765E-003);
> +   LLVMValueRef coscof_p2 = lp_build_const_vec(gallivm, bld->type, 4.166664568298827E-002);
>
>    /*
>     * y = *(v4sf*)_ps_coscof_p0;
> @@ -2059,10 +2038,10 @@ lp_build_cos(struct lp_build_context *bld,
>     * y = _mm_sub_ps(y, tmp);
>     * y = _mm_add_ps(y, *(v4sf*)_ps_1);
>     */
> -   LLVMValueRef half = lp_build_const_v4sf(gallivm, 0.5);
> +   LLVMValueRef half = lp_build_const_vec(gallivm, bld->type, 0.5);
>    LLVMValueRef tmp = LLVMBuildFMul(b, z, half, "tmp");
>    LLVMValueRef y_9 = LLVMBuildFSub(b, y_8, tmp, "y_8");
> -   LLVMValueRef one = lp_build_const_v4sf(gallivm, 1.0);
> +   LLVMValueRef one = lp_build_const_vec(gallivm, bld->type, 1.0);
>    LLVMValueRef y_10 = LLVMBuildFAdd(b, y_9, one, "y_9");
>
>    /*
> @@ -2070,9 +2049,9 @@ lp_build_cos(struct lp_build_context *bld,
>     * _PS_CONST(sincof_p1,  8.3321608736E-3);
>     * _PS_CONST(sincof_p2, -1.6666654611E-1);
>     */
> -   LLVMValueRef sincof_p0 = lp_build_const_v4sf(gallivm, -1.9515295891E-4);
> -   LLVMValueRef sincof_p1 = lp_build_const_v4sf(gallivm, 8.3321608736E-3);
> -   LLVMValueRef sincof_p2 = lp_build_const_v4sf(gallivm, -1.6666654611E-1);
> +   LLVMValueRef sincof_p0 = lp_build_const_vec(gallivm, bld->type, -1.9515295891E-4);
> +   LLVMValueRef sincof_p1 = lp_build_const_vec(gallivm, bld->type, 8.3321608736E-3);
> +   LLVMValueRef sincof_p2 = lp_build_const_vec(gallivm, bld->type, -1.6666654611E-1);
>
>    /*
>     * Evaluate the second polynom  (Pi/4 <= x <= 0)
> @@ -2102,8 +2081,8 @@ lp_build_cos(struct lp_build_context *bld,
>     * y = _mm_andnot_ps(xmm3, y);
>     * y = _mm_add_ps(y,y2);
>     */
> -   LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, v4si, "y2_i");
> -   LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, v4si, "y_i");
> +   LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, bld->int_vec_type, "y2_i");
> +   LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, bld->int_vec_type, "y_i");
>    LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and");
>    LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv");
>    LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and");
> @@ -2114,7 +2093,7 @@ lp_build_cos(struct lp_build_context *bld,
>     * y = _mm_xor_ps(y, sign_bit);
>     */
>    LLVMValueRef y_sign = LLVMBuildXor(b, y_combine, sign_bit, "y_sin");
> -   LLVMValueRef y_result = LLVMBuildBitCast(b, y_sign, v4sf, "y_result");
> +   LLVMValueRef y_result = LLVMBuildBitCast(b, y_sign, bld->vec_type, "y_result");
>    return y_result;
>  }
>
> --
> 1.7.9
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list