fix races in ztm (93521048) · Commits · Giacomo Mulas / NP_TMcode

src/libnptm/clu_subs.cpp

+19 −10

Original line number	Original line	Diff line number	Diff line
	@@ -2534,7 +2534,7 @@ void tqr(
	}		}

	void ztm(dcomplex *am, ParticleDescriptor c1) {		void ztm(dcomplex *am, ParticleDescriptor c1) {
	dcomplex gie, gle, a1, a2, a3, a4, sum1, sum2, sum3, sum4;		// dcomplex gie, gle, a1, a2, a3, a4, sum1, sum2, sum3, sum4;
	const dcomplex cc0 = 0.0 + 0.0 * I;		const dcomplex cc0 = 0.0 + 0.0 * I;
	// int i2 = 0; // old implementation		// int i2 = 0; // old implementation
	#ifdef USE_NVTX		#ifdef USE_NVTX
	@@ -2546,7 +2546,6 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	// C9 c9_para = new C9(c9);		// C9 c9_para = new C9(c9);
	dcomplex *gis_v = c1->gis[0];		dcomplex *gis_v = c1->gis[0];
	dcomplex *gls_v = c1->gls[0];		dcomplex *gls_v = c1->gls[0];
	double rac3j_local = (double ) malloc(c1->lmtpo*sizeof(double));
	int k2max = c1->li*(c1->li+2);		int k2max = c1->li*(c1->li+2);
	int k3max = c1->le*(c1->le+2);		int k3max = c1->le*(c1->le+2);
	// To parallelise, I run a linearised loop directly over k		// To parallelise, I run a linearised loop directly over k
	@@ -2559,6 +2558,8 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	// im = im -(2*l+1) and l = l+1 (there was a rounding error in a nearly exact root)		// im = im -(2*l+1) and l = l+1 (there was a rounding error in a nearly exact root)
	#ifdef USE_TARGET_OFFLOAD		#ifdef USE_TARGET_OFFLOAD
	#pragma omp target teams distribute parallel for simd collapse(3)		#pragma omp target teams distribute parallel for simd collapse(3)
			#else
			#pragma omp parallel for simd collapse(3)
	#endif		#endif
	for (int n2 = 1; n2 <= c1->nsph; n2++) { // GPU portable?		for (int n2 = 1; n2 <= c1->nsph; n2++) { // GPU portable?
	for (int k2 = 1; k2<=k2max; k2++) {		for (int k2 = 1; k2<=k2max; k2++) {
	@@ -2590,12 +2591,13 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	int i3 = l3 * l3 + im3 - 1;		int i3 = l3 * l3 + im3 - 1;
	int m3 = -l3 - 1 + im3;		int m3 = -l3 - 1 + im3;
	int vecindex = (i2 - 1) * c1->nlem + i3 - 1;		int vecindex = (i2 - 1) * c1->nlem + i3 - 1;
			double rac3j_local = (double ) malloc(c1->lmtpo*sizeof(double));
	gis_v[vecindex] = ghit_d(2, 0, n2, l2, m2, l3, m3, c1, rac3j_local);		gis_v[vecindex] = ghit_d(2, 0, n2, l2, m2, l3, m3, c1, rac3j_local);
	gls_v[vecindex] = ghit_d(2, 1, n2, l2, m2, l3, m3, c1, rac3j_local);		gls_v[vecindex] = ghit_d(2, 1, n2, l2, m2, l3, m3, c1, rac3j_local);
			free(rac3j_local);
	} // close k3 loop, former l3 + im3 loops		} // close k3 loop, former l3 + im3 loops
	} // close k2 loop, former l2 + im2 loops		} // close k2 loop, former l2 + im2 loops
	} // close n2 loop		} // close n2 loop
	free(rac3j_local);
	#ifdef USE_NVTX		#ifdef USE_NVTX
	nvtxRangePop();		nvtxRangePop();
	#endif		#endif
	@@ -2606,6 +2608,8 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	dcomplex *sam_v = c1->sam[0];		dcomplex *sam_v = c1->sam[0];
	#ifdef USE_TARGET_OFFLOAD		#ifdef USE_TARGET_OFFLOAD
	#pragma omp target teams distribute parallel for simd collapse(2)		#pragma omp target teams distribute parallel for simd collapse(2)
			#else
			#pragma omp parallel for simd collapse(2)
	#endif		#endif
	for (int i1 = 1; i1 <= c1->ndi; i1++) { // GPU portable?		for (int i1 = 1; i1 <= c1->ndi; i1++) { // GPU portable?
	for (int i3 = 1; i3 <= c1->nlem; i3++) {		for (int i3 = 1; i3 <= c1->nlem; i3++) {
	@@ -2615,7 +2619,6 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	dcomplex sum4 = cc0;		dcomplex sum4 = cc0;
	int i1e = i1 + c1->ndi;		int i1e = i1 + c1->ndi;
	int i3e = i3 + c1->nlem;		int i3e = i3 + c1->nlem;
	#pragma parallel for simd reduction(+:sum1,sum2,sum3,sum4)
	for (int i2 = 1; i2 <= c1->ndi; i2++) {		for (int i2 = 1; i2 <= c1->ndi; i2++) {
	int i2e = i2 + c1->ndi;		int i2e = i2 + c1->ndi;
	int vecindg_23 = (i2 - 1) * c1->nlem + i3 - 1;		int vecindg_23 = (i2 - 1) * c1->nlem + i3 - 1;
	@@ -2640,7 +2643,11 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	sam_v[vecind1e + i3e - 1] = sum4;		sam_v[vecind1e + i3e - 1] = sum4;
	} // i3 loop		} // i3 loop
	} // i1 loop		} // i1 loop
	#pragma omp parallel for collapse(2)		#ifdef USE_TARGET_OFFLOAD
			#pragma omp target teams distribute parallel for simd collapse(2)
			#else
			#pragma omp parallel for simd collapse(2)
			#endif
	for (int i1 = 1; i1 <= c1->ndi; i1++) {		for (int i1 = 1; i1 <= c1->ndi; i1++) {
	for (int i0 = 1; i0 <= c1->nlem; i0++) {		for (int i0 = 1; i0 <= c1->nlem; i0++) {
	int vecindex = (i1 - 1) * c1->nlem + i0 - 1;		int vecindex = (i1 - 1) * c1->nlem + i0 - 1;
	@@ -2650,7 +2657,9 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	} // i1 loop		} // i1 loop
	dcomplex *vec_am0m = c1->am0m[0];		dcomplex *vec_am0m = c1->am0m[0];
	#ifdef USE_TARGET_OFFLOAD		#ifdef USE_TARGET_OFFLOAD
	#pragma omp target parallel for collapse(2)		#pragma omp target teams distribute parallel for simd collapse(2)
			#else
			#pragma omp parallel for simd collapse(2)
	#endif		#endif
	for (int i0 = 1; i0 <= c1->nlem; i0++) {		for (int i0 = 1; i0 <= c1->nlem; i0++) {
	for (int i3 = 1; i3 <= c1->nlemt; i3++) {		for (int i3 = 1; i3 <= c1->nlemt; i3++) {
	@@ -2661,11 +2670,11 @@ void ztm(dcomplex *am, ParticleDescriptor c1) {
	int i1e = i1 + c1->ndi;		int i1e = i1 + c1->ndi;
	int vecind1 = (i1 - 1) * c1->nlemt;		int vecind1 = (i1 - 1) * c1->nlemt;
	int vecind1e = (i1e - 1) * c1->nlemt;		int vecind1e = (i1e - 1) * c1->nlemt;
	a1 = sam_v[vecind1 + i3 - 1];		dcomplex a1 = sam_v[vecind1 + i3 - 1];
	a2 = sam_v[vecind1e + i3 - 1];		dcomplex a2 = sam_v[vecind1e + i3 - 1];
	int vecindex = (i1 - 1) * c1->nlem + i0 - 1;		int vecindex = (i1 - 1) * c1->nlem + i0 - 1;
	gie = gis_v[vecindex];		dcomplex gie = gis_v[vecindex];
	gle = gls_v[vecindex];		dcomplex gle = gls_v[vecindex];
	sum1 += (a1 * gie + a2 * gle);		sum1 += (a1 * gie + a2 * gle);
	sum2 += (a1 * gle + a2 * gie);		sum2 += (a1 * gle + a2 * gie);
	} // i1 loop		} // i1 loop